Statistics in Action

cover
http://acr.keymath.com/KeyPressPortalV3.0/ImportingCourses/SIA2/fr...
1 de 1
10-03-2009 13:32
Statistics in Action
1 de 1
http://acr.keypress.com/KeyPressPortalV3.0/ImportingCourses/SIA2/fr...
Statistics in Action
Understanding a World of Data
Second Edition
2008 Key Curriculum Press
23-03-2009 21:25
copyright
1 de 1
http://acr.keymath.com/KeyPressPortalV3.0/ImportingCourses/SIA2/fr...
Project Editor
Consulting Editor
Project Administrators
Editorial Assistants
Teacher Consultant and Writer
Mathematics Reviewer and
Accuracy Checker
AP Sample Test Contributor
AP Teacher Reviewers
Josephine Noah
Kendra Lockman
Elizabeth Ball, Aaron Madrigal
Aneesa Davenport, Nina Mamikunian
Corey Andreasen, North High School, Sheboygan, Wisconsin
Cindy Clements, Trinidad State Junior College, Trinidad, Colorado
Editorial Production Manager

Production Editor
Copyeditor
Production Supervisor
Production Coordinator
Text Designers
Compositor
Art Editor and Technical Artist
Cover Designers
Cover Photo Credit
Prepress and Printer
Joshua Zucker, Castilleja School, Palo Alto, California

Angelo DeMattia, Columbia High School, Maplewood, New Jersey
Beth Fox-McManus, formerly of Alan C. Pope High School, Marietta, Georgia
Dan Johnson, Silver Creek High School, San Jose, California
Gil Cuevas, University of Miami, Coral Gables, Florida
Genevieve Lau, Skyline College, San Bruno, California
Beatrice Lumpkin, Malcolm X College (retired), Chicago, Illinois
Christine Osborne
Kristin Ferraioli
Mary Roybal
Ann Rothenbuhler
Thomas Brierly
Graphic World, Thomas Brierly
Graphic World
LMP Media, Inc.
Jensen Barnes, Nidaul Uk
Getty Images/Alberto Incrocci
RR Donnelley
Textbook Product Manager

Executive Editor
Publisher
James Ryan
Casey FitzSimons
Steven Rasmussen
Multicultural Reviewers
2008 by Key Curriculum Press. All rights reserved.

No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any
form or by any means, electronic, photocopying, recording, or otherwise, without the prior
written permission of the publisher.
Key Curriculum Press is a registered trademark of Key Curriculum Press. Fathom Dynamic
Data is a trademark of KCP Technologies. All other registered trademarks and trademarks in this
book are the property of their respective holders.
Key Curriculum Press
1150 65th Street
Emeryville, CA 94608
editorial@keypress.com
www.keypress.com
Printed in the United States of America
10 9 8 7 6 5 4 3 2 1
11
ISBN: 978-1-55953-909-8
10
09
08
07
10-03-2009 16:53
About The Author
1 de 1
About the Authors

Ann E. Watkins is Professor of Mathematics at California State University,
Northridge (CSUN). She received her Ph.D. in education from the University of
California, Los Angeles. She is a former president of the Mathematical Association
of America (MAA) and a Fellow of the American Statistical Association (ASA).
Dr. Watkins has served as co-editor of College Mathematics Journal, as a member
of the Board of Editors of American Mathematical Monthly, and as Chair of the
Advanced Placement Statistics Development Committee. She was selected as the
19941995 CSUN Outstanding Professor and won the 1997 CSUN Award for the
Advancement of Teaching Effectiveness. Before moving to CSUN in 1990, she taught
for the Los Angeles Unified School District and at Pierce College in Los Angeles.
In addition to numerous journal articles, she is the co-author of books based on
work produced by the Activity-Based Statistics Project (co-authored with Richard
Scheaffer), the Quantitative Literacy Project, and the Core-Plus Mathematics Project.
Richard L. Scheaffer is Professor Emeritus of Statistics at the University of
Florida, where he served as chairman of the Department of Statistics for 12 years.

He received his Ph.D. in statistics from Florida State University. Dr. Scheaffers
research interests are in the areas of sampling and applied probability, especially
in their applications to industrial processes. He has published numerous papers
and is co-author of four college-level textbooks.
In recent years, much of his effort has been directed toward statistics education
at the elementary, secondary, and college levels. He was one of the developers of the
Quantitative Literacy Project where he helped form the basis of the data analysis
emphasis in mathematics curriculum standards recommended by the National
Council of Teachers of Mathematics. Dr. Scheaffer has also directed the task force
that developed the Advanced Placement Statistics Program and served as its first
Chief Faculty Consultant. Dr. Scheaffer is Fellow and past president of the American
Statistical Association, from which he received a Founders Award.
George W. Cobb is the Robert L. Rooke Professor of Statistics at Mt. Holyoke
College, where he served a three-year term as Dean of Studies. He received his Ph.D.
in statistics from Harvard University, and is an expert in statistics education with
a significant publication record. He chaired the joint committee on undergraduate
statistics of the Mathematical Association of America and the American Statistical
Association and is a Fellow of the American Statistical Association. He also led the
Statistical Thinking and Teaching Statistics (STATS) project of the Mathematical
Association of America, which helped professors of mathematics learn to teach
statistics. Over the past two decades, Dr. Cobb has frequently served as an expert
witness in lawsuits involving alleged employment discrimination.
iii
23-03-2009 21:20
About The Author
1 de 1
Acknowledgments
This book is a product of what we have learned from the statisticians and teachers
who have been actively involved in helping the introductory statistics course
evolve into one that emphasizes activity-based learning of statistical concepts
while reflecting modern statistical practice. This book is written in the spirit of the
recommendations from the MAAs STATS project, the ASAs Quantitative Literacy
and GAISE projects, and the College Boards AP Statistics course. We hope that
it adequately reflectsS the wisdom and experience of those with whom we have
worked and who have inspired and taught us.
We owe special thanks to Corey Andreasen, an outstanding high school
mathematics and AP Statistics teacher at North High School in Sheboygan,
Wisconsin, for his insight into what makes a topic teachable to high school
students. His careful review of the manuscript led to many clarifications of
wording and improvements in exercises, all of which will make it easier for you to
learn the material. Corey also has made substantial contributions to the solutions
and teachers notes, adding his unique perspective and sense of humor.
It has been an awesome experience to work with the Key Curriculum staff
and field-test teachers, who always put the interests of students and teachers first.
Their commitment to excellence has motivated us to do better than we ever could
have done on our own. Steve, Casey, Jim, Kristin B., Kristin F., and the rest of the
staff have been professional and astute throughout. Our deepest gratitude goes to
Cindy Clements and Josephine Noah, the editors of the first and second editions,
respectively, who have been a joy to work with. (Not all authors say thatand
mean itabout their editors.) Cindy and Josephine were outstanding high school
teachers before coming to Key. Their organizational skills, experience in the
classroom, and insight have improved every chapter of this text.
iv
23-03-2009 21:21
Contents v
1 de 1
< NOBR>
Contents
CHAPTER 1
CHAPTER
CHAPTER
A Note to Students from the Authors
ix
Statistical Reasoning: Investigating a

Claim of Discrimination
1.1 Discrimination in the Workplace: Data Exploration

1.2 Discrimination in the Workplace: Inference
Chapter Summary
4
11
20
Exploring Distributions
26
2.1 Visualizing Distributions: Shape, Center, and Spread

2.2 Graphical Displays of Distributions
2.3 Measures of Center and Spread
2.4 Working with Summary Statistics
2.5 The Normal Distribution
Chapter Summary
28
42
56
74
83
95
Relationships Between Two Quantitative Variables

104
3.1 Scatterplots
3.2 Getting a Line on the Pattern
3.3 Correlation: The Strength of a Linear Trend
3.4 Diagnostics: Looking for Features That the Summaries Miss
3.5 Shape-Changing Transformations
Chapter Summary
CHAPTER
Sample Surveys and Experiments

4.1 Why Take Samples, and How Not To
4.2 Random Sampling: Playing It Safe by Taking Chances
4.3 Experiments and Inference About Cause
4.4 Designing Experiments to Reduce Variability
Chapter Summary
106
116
140
162
179
203
216
219
231
243
262
278
v
23-03-2009 21:22
Contents vi
1 de 1
CHAPTER 5
Probability Models
286
5.1 Constructing Models of Random Behavior

5.2 Using Simulation to Estimate Probabilities
5.3 The Addition Rule and Disjoint Events
5.4 Conditional Probability
5.5 Independent Events
Chapter Summary
CHAPTER
288
301
315
325
339
349
Probability Distributions
356
6.1 Random Variables and Expected Value

6.2 The Binomial Distribution
6.3 The Geometric Distribution
Chapter Summary
CHAPTER
358
382
393
402
Sampling Distributions
408
7.1 Generating Sampling Distributions

7.2 Sampling Distribution of the Sample Mean
7.3 Sampling Distribution of the Sample Proportion
Chapter Summary
CHAPTER
410
426
446
458
Inference for Proportions
466
8.1 Estimating a Proportion with Confidence

8.2 Testing a Proportion
8.3 A Confidence Interval for the Difference of Two Proportions
8.4 A Significance Test for the Difference of Two Proportions
8.5 Inference for Experiments
Chapter Summary
CHAPTER
Inference for Means

9.1 A Confidence Interval for a Mean
9.2 A Significance Test for a Mean
9.3 When Things Arent Normal
9.4 Inference for the Difference Between Two Means
9.5 Paired Comparisons
Chapter Summary
vi
468
490
518
526
539
554
560
562
580
602
616
640
663
23-03-2009 21:22
Contents vii
1 de 1
CHAPTER 10
Chi-Square Tests
10.1 Testing a Probability Model: The Chi-Square Goodness-of-Fit Test
10.2 The Chi-Square Test of Homogeneity
10.3 The Chi-Square Test of Independence
Chapter Summary
CHAPTER
CHAPTER
11
12
Inference for Regression
736
739
755
771
790
Statistics in Action: Case Studies
798
Mums the Word!

Keeping Tabs on Americans
Baseball: Does Money Buy Success?
Martin v. Westvaco Revisited: Testing for Discrimination
Against Employees
Statistical Tables
Table A: Standard Normal Probabilities
Table B: t-Distribution Critical Values
Table C:
Critical Values
Table D: Random Digits
674
692
711
728
11.1 Variation in the Slope from Sample to Sample

11.2 Making Inferences About Slopes
11.3 Transforming for a Better Fit
Chapter Summary
12.1
12.2
12.3
12.4
APPENDIX
672
800
803
810
817
824
824
826
827
828
Glossary
829
Brief Answers to Selected Problems
839
Index
883
Photo Credits
894
vii
23-03-2009 21:22
Acknowledgments
1 de 1
A Note to Students
from the Authors
Data enter the conversation whether you talk about income, sports, health,
politics, the weather, or prices of goods and services. In fact, in this age of
information technology, data come at you at such a rapid rate that you can catch
only a glimpse of the masses of numbers. You cannot cope intelligently in this
quantitative world unless you have an understanding of the basic concepts of
statistics and have had practice making informed decisions using real data.
Statistics in Action is designed for students taking an introductory high
school statistics course and includes all of the topics in the Advanced Placement
(AP) Statistics syllabus. Beginning in Chapter 1 with a court case about age
discrimination, you will be immersed in real-world problems that can be solved
only with statistical methods. You will learn to explore, summarize, and display data;
design surveys and experiments; use probability to understand random behavior;
make inferences about populations by looking at samples from those populations;
and make inferences about the effect of treatments from designed experiments.
After completing your statistics course, you will be prepared to take the AP
Statistics Exam, to take a follow-up college-level course, and, above all, to make
informed decisions in this world of data.
You will be using this book in a first course in statistics, so, you arent required
to know anything yet about statistics. You may find that your success in statistics
results more from your perseverance in trying to understand what you read rather
than your skill with algebra. However, basic topics from algebra, such as slope,
linear equations, exponential equations, and the idea of a logarithm, will arise
throughout the book. Be prepared to review these as you go along.
Statistics from a Modern Perspective

Statistical work is more interactive than it was a generation ago. Computers and
graphing calculators have automated the graphical exploration of data and, in the
process, have made the practice of statistics a more visual enterprise. Statistical
techniques are also changing as simulations allow statisticians (and you) to shift
the emphasis from following recipes for calculations to paying more attention to
statistical concepts. Your instructor has selected this book so that you may learn
this modern, data-analytic approach to statistics and because he or she encourages
you to be an active participant in the classroom, wants you to see real data (if you
have only pretend data, you can only pretend to analyze it), believes that statistical
analyses must be tailored to the data, and uses graphing calculators or statistical
software for data analysis and for simulations.
ix
23-03-2009 21:24
What You Should Know About This Book
1 de 1
These features grow out of the vigorous changes that have been reshaping the
practice of statistics and the teaching of statistics over the last quarter century.
The most basic question to ask about any data set is, Where did the data come
from? Good data for statistical analysis must come from a good plan for data
collection. Thus, Statistics in Action treats the design and analysis of experiments
honestly and thoroughly and discusses how these methods of collecting data differ
from observational studies. It then follows through on this theme by relating the
statistical analysis to the manner in which the data were collected.
What You Should Know About This Book

Throughout Statistics in Action you will see many graphical displays, lots of real
data, activities that introduce each major topic, computer printouts, questions
for you to discuss with your class, and practice problems so that you can be sure
you understand the basics before you move on. The practice problems are found
at the end of every section, organized by topic. You should work every problem
for those topics that you wish to learn. The answers to all practice problems and
odd-numbered exercises are in the back of the book.
Also in the back of the book you will find a glossary of statistical notation, a
glossary of statistical terminology, statistical tables, and an index so that you can
locate topics quickly. Two tables are reprinted inside the back cover of the book.
You should keep in mind that the emphasis throughout is on the development
of statistical concepts. Even when looking for a numerical answer in a practice
problem or exercise, think about the underlying concept that is being illustrated.
Concepts are carefully developed in the written text, so read the material
thoughtfully and strive to fully understand what is being explained. Concepts
are also developed in the discussion questions. These are designed for in-class
discussions, but regard them as part of your reading assignment, even if class
time is too limited for full discussion. Through reading, discussing, and working
practice problems and exercises you will develop a profound understanding of
statistical thinking, and that will serve you well as a basis for a lifetime of coping
with a quantitative world.
Ann Watkins
Dick Scheaffer
George Cobb
23-03-2009 21:24
Lesson
2 de 3
http://acr.keypress.com/KeyPressPortalV3.0/Viewer/Lesson.htm
Statistical Reasoning:
Investigating a Claim
of Discrimination
Were older workers

discriminated against
during a companys
downsizing? When an
older worker felt he had
been unfairly laid off,
his lawyers called on a
statistician to help them
evaluate the claim.
23-03-2009 21:25
Lesson
3 de 3
In the year Robert Martin turned 54, the Westvaco Corporation, which makes
paper products, decided to downsize. They laid off several members of the
engineering department, including Robert Martin. Later that year, he sued
Westvaco, claiming he had been laid off because of his age. A major piece of
Martins case was based on a statistical analysis of the ages of the Westvaco
employees.
In the two sections of this chapter, you will get a chance to try your hand at
two very different kinds of statistical work, exploration and inference. Exploration
is an informal, open-ended examination of data. Your goal in the first section
will be to uncover and summarize patterns in data from Westvaco that bear on
the Martin case. You will try to formulate and answer basic questions such as
Were those who were laid off older on average than those who werent laid off?
You can use any toolsgraphs, averages, and so onthat you think might be
useful. Inference, which youll use in the second section, is quite different from
exploration in that it follows strict rules and focuses on judging whether the
patterns you found are the sort you would expect. Youll use inference to decide
whether the patterns you find in the Westvaco data are the sort you would expect
from a company that does not discriminate on the basis of age, or whether further
investigation into possible age discrimination is needed.
The purpose of this first chapter is to familiarize you with the ideas of
statistical thinking before you involve yourself with the details of statistical
methods. It is easy to get caught in the trap of doing rather than understanding, of
asking how rather than why. You cant do statistics unless you learn the methods,
but you must not get so caught up in the details of the methods that you lose
sight of what they mean. Doing and thinking, method and meaning, will compete
for your attention throughout this course.
In this chapter, you will learn the basic ideas of

exploring datauncovering and summarizing patterns
making inferences from datadeciding whether an observed feature of the
data could reasonably be attributed to chance alone
These ideas will remain key components of the statistical concepts youll develop
and study throughout this course.
23-03-2009 21:25
Lesson
1 de 7
1.1
Variables provide
information about cases.
Variability is what statistics

is all about.
A distribution is a record of
variability.
Discrimination in the Workplace: Data Exploration

Robert Martin was one of 50 people working in the engineering department of
Westvacos envelope division. One spring, Westvacos management went through
five rounds of planning for a reduction in their work force. In Round 1, they
eliminated 11 positions, and they eliminated 9 more in Round 2. By the time the
layoffs ended, after all five rounds, only 22 of the 50 workers had kept their jobs.
The average age in the department had fallen from 48 to 46.
After Martin, age 54, was laid off, he sued Westvaco for age discrimination.
Display 1.1 shows the data provided by Westvaco to Martins lawyers. The
statistical analysis in the lawsuit used all 50 employees in the engineering
department of the envelope division, with separate analyses for salaried and
hourly workers. Each row in Display 1.1 corresponds to one worker, and each
column corresponds to a characteristic of the worker: job title, whether hourly
or salaried, month and year of birth, month and year of hire, and age at birthday
in 1991. The next-to-last column (Round) tells how the worker fared in the
downsizing: 1 means chosen for layoff in Round 1 of planning for the reduction
in force, 2 means chosen in Round 2, and so on for Rounds 3, 4, and 5; 0 means
not chosen for layoff.
The subjects (or objects) of statistical examination often are called cases.
In the rows in Display 1.1, the cases are individual Westvaco employees. Their
characteristics, in the columns, are the variables. If you pick a row and read
across, you find information about a single case. (For example, Robert Martin, in
Row 44, was salaried, was born in September 1937, was hired in October 1967,
was chosen for layoff in Round 2, and turned 54 in 1991.) Although reading
across might seem the natural way to read the table, in statistics you will often
find it useful to pick a column and read down. This gives you information about
a single variable as you range through all the cases. For example, pick Age, read
down the column, and notice the variability in the ages. It is variability like
thisthe fact that individuals differthat can make it a challenge to see patterns
in data and figure out what they mean.
Imagine: If there had been no variabilityif all the workers had been of just
two ages, say, 30 and 50, and Westvaco had laid off all the 50-year-olds and kept
all the 30-year-oldsthe conclusion would be obvious and there would be no
need for statistics. But real life is more subtle than that. The ages of the laid-off
workers varied, as did the ages of the workers retained. Statistical methods were
designed to cope with such variability. In fact, you might define statistics as the
science of learning from data in the presence of variability.
Although the bare fact that the ages vary is easy to see in the data table, the
pattern of those ages is not so easy to see. This patternwhat the values are and
how often each occursis their distribution. In order to see that pattern, a graph
is better than a table. The dot plot in Display 1.2 shows the distribution of the
ages of the 36 salaried employees who worked in the engineering department just
before the layoffs began.
[To learn how to create a dot plot on your calculator, see Calculator Note 1A.]
Chapter 1 Statistical Reasoning: Investigating a Claim of Discrimination
23-03-2009 21:28
Lesson
2 de 7
Display 1.1 The data in Martin v. Westvaco. [ Source: Martin v. Envelope Division of Westvaco Corp.,
CA No. 92-03121-MAP, 850 Fed. Supp. 83 (1994).]
23-03-2009 21:28
Lesson
3 de 7
Display 1.2 Ages of the salaried workers. (Each dot represents a

worker; the age is shown by the position of the dot
along the scale below it.)
Display 1.2 provides some useful information about the variability in the ages,
but by itself doesnt tell anything about possible age discrimination in the layoffs.
For that, you need to distinguish between those salaried workers who lost their
jobs and those who didnt. The dot plot in Display 1.3, which shows those laid
off and those retained, provides weak evidence for Martins case. Those laid off
generally were older than those who kept their jobs, but the pattern isnt striking.
Display 1.3
Salaried workers: ages of those laid off and those

retained.
Display 1.3 shows that most salaried workers who were laid off were age 50
or older. However, this alone doesnt support Martins case because most of the
workers were age 50 or older to begin with.
One way to proceed is to make a summary table. The table shown here
classifies the salaried workers according to age and whether they were laid off
or retained. (Using 50 as the dividing age between younger and older is a
somewhat arbitrary, but reasonable, decision.)
To decide whether Martin has a case, compare the proportion of salaried

workers under 50 who were laid off (6 out of 16, or 0.375) with the proportion of
those 50 or older who were laid off (12 out of 20, or 0.60). These proportions are
quite different, an argument in favor of Martin.
Looking at the layoffs of the salaried workers round by round provides further
evidence in favor of Martin. The dot plots in Display 1.4 show the ages of the
salaried workers laid off and retained by round. These new dot plots use different
6
23-03-2009 21:28
Lesson
4 de 7
symbols for laid-off workers and for retained workers. For example, in the top dot
plot, the open circles represent the salaried workers whose jobs did not survive
Round 1. In this round, the four oldest workers were laid off, but only one worker
under age 50 was laid off. In the second round, one of the two oldest employees
was laid off. But then this pattern stopped. Again, the evidence favors Martin but
is far from conclusive.
Display 1.4
Salaried workers: ages of those laid off (open circles)

and those retained (solid dots) in each round.
You might feel as if the analysis so far ignores important facts, such as worker
qualifications. Thats true. However, the first step is to decide whether, based
on the data in Display 1.1, older workers were more likely to be laid off. If not,
Martins case fails. If so, it is then up to Westvaco to justify its actions.
Exploring the Martin v. Westvaco Data

D1. Suppose you were the judge in the Martin v. Westvaco case. How would you
use the information in Display 1.1 to decide whether Westvaco tended to lay
off older workers in disproportionate numbers (for whatever reason)?
D2. Display 1.5 (on the next page) is like Display 1.3 except that it gives data
for the hourly workers. Compare the plots for the hourly and salaried
workers. Which provides stronger evidence in support of Martins claim
of age discrimination?
1.1 Discrimination in the Workplace: Data Exploration 7
23-03-2009 21:28
Lesson
5 de 7
Display 1.5 Hourly workers: ages of those laid off and those
retained.
D3. Whenever you think you have a message from data, you should be careful
not to jump to conclusions. The patterns in the Westvaco data might be
realthey reflect age discrimination on the part of management. On
the other hand, the patterns might be the result of chancemanagement
wasnt discriminating on the basis of age but simply by chance happened
to lay off a larger percentage of older workers. Whats your opinion about
the Westvaco data: Do the patterns seem realtoo strong to be explained
by chance?
D4. The analysis up to this point ignores important facts such as worker
qualifications. Suppose Martin makes a convincing case that older workers
were more likely to be laid off. It is then up to Westvaco to justify its actions.
List several specific reasons Westvaco might give to justify laying off a
disproportionate number of older workers.
Summary 1.1: Data Exploration

Data exploration, or exploratory analysis, is a purposeful investigation to find
patterns in data, using tools such as tables and graphs to display those patterns
and statistical concepts such as di
23-03-2009 21:28
Lesson
6 de 7
Practice
Practice problems help you master basic concepts
group does the evidence more strongly
and computations. Throughout this textbook, you
favor Martins case?
should work all the practice problems for each
P3. Display 1.6 shows layoffs and retentions by
topic you want to learn. The answers to all practice
round for hourly workers. (There is no plot
problems are given in the back of the book.
for Round 5 because no hourly workers were
chosen for layoff in that round.) Compare
Exploring the Martin v. Westvaco Data
the pattern for the hourly workers with the
P1. Construct a dot plot similar to Display 1.3,
pattern for the salaried workers in Display
comparing the ages of hourly workers who
1.4 on page 7. For which group does the
lost their jobs during Rounds 13 to the ages
evidence more strongly favor Martins case?
of hourly workers who still had their jobs at
the end of Round 3. How do the ages differ?
P2. This summary table classifies the hourly
workers according to age and whether they
were laid off or retained.
a. What proportion of hourly workers

under age 50 were laid off? Were not
laid off?
b. What proportion of laid-off hourly
workers were under age 50? Were age 50
or older?
c. What two proportions should you
compute and compare in order to decide
whether older hourly workers were
disproportionately laid off? Make these
computations and give your conclusion.
d. Compare this table with the table for the
salaried workers on page 6. For which
Display 1.6 Hourly workers: ages of those laid off

(open circles) and those retained (solid
dots) in each round.
Exercises
E1. This summary table classifies salaried
workers as to whether they were laid off and
their age, this time using 40 as the cutoff
between younger and older workers.
a. What proportion of workers age 40 or

older were laid off? What proportion of
laid-off workers were age 40 or older?
b. What proportion of workers under age 40
were laid off? What proportion were not
laid off?
compute and compare in order to
1.1 Discrimination in the Workplace: Data Exploration 9
23-03-2009 21:28
Lesson
7 de 7
decide whether older workers were

E5. Refer to Display 1.1 on page 5.
disproportionately laid off? Make these
a. Create a summary table whose five cases
computations and give your conclusion.
are Round 1 through Round 5 and
d. Compare this table with the table for the
whose three variables are total number of
salaried workers on page 6, where 50 was
employees laid off in that round, number
the age cutoff. If you were Martins lawyer,
of employees laid off in that round who
would you present a table using 40 or 50
were 40 or older, and percentage laid off
in that round who were 40 or older.
as the cutoff?
E2. Explore whether hourly workers at Westvaco
b. Describe any patterns you find in the table
were more likely than salaried workers to
and what you think they might mean.
lose their jobs.
E6. Last hired, first fired is shorthand for
a. Start by constructing a summary table to
When you have to downsize, start by laying
display the relevant data.
off the newest person, then the person hired
next before that, and work back in reverse
b. Compute two proportions that will allow
order of seniority. (The person whos been
you to make this comparison.
working longest will be the last to be laid
c. What do you conclude from comparing
off.) Examine the Westvaco data.
the proportions?
a. How was seniority related to the
E3. Twenty-two workers kept their jobs. Explore
decisions about layoffs in the engineering
whether the age distributions are similar for
department at Westvaco?
the hourly and salaried workers who kept
b. What explanation(s) can you suggest for
their jobs.
any patterns you find?
a. Show the two age distributions on a pair
E7.
Many
tables in the media are arranged with
of dot plots that have the same scale. How
cases
as
rows and variables as columns. For
do these distributions differ?
Displays
1.7 and 1.8 in parts a and b, identify
b. Do your dot plots in part a support a
the cases and the variables. Then compute
claim that Westvaco was more inclined to
the values missing from each table.
keep older workers if they were salaried
rather than hourly?
E4. Consider these three facts from your work in
this section:
Salaried workers were more likely to keep
their jobs than were hourly workers.
Older workers were more likely to be laid
off than were younger workers.
Older workers were more likely to be
salaried than were younger workers.
Putting these three facts together, what can
you conclude? Is this evidence in favor of
Martins case, or does it help Westvaco?
10 Chapter 1 Statistical Reasoning: Investigating a Claim of Discrimination
23-03-2009 21:28
Lesson
1 de 9
a.
Display 1.7 Number of bases stolen in a single season by the top five Major
League baseball players. [Source: mlb.mlb.com.]
b.
Display 1.8 New York Stock Exchange Activity for October 29, 1929.
[Source: marketplace.publicradio.org.]
E8. Suppose you are studying the effects of

poverty and plan to construct a data set
whose cases are the villages in Bolivia. Name
some variables that you might study.
1.2
Discrimination in the Workplace: Inference

Overall, the exploratory work on the Westvaco data set in Section 1.1 shows that
older workers were more likely than younger workers to be laid of and were laid
off earlier. One of the main arguments in the court case, along the lines set out
in D3, was about what those patterns mean: Can you infer from the patterns that
Westvaco has some explaining to do, or are they the sort of patterns that tend to
happen even in the absence of discrimination?
A comprehensive analysis of Martin v. Westvaco will have to wait for its
reappearance among the case studies of Chapter 12, when youll be more familiar
with the concepts and tools of statistics. For now, you can get a pretty good idea
of how the analysis goes by working with a subset of the data.
1.2 Discrimination in the Workplace: Inference 11
23-03-2009 21:29
Lesson
2 de 9
The ages of the ten hourly workers involved in Round 2 of the layoffs,
arranged from youngest to oldest, were 25, 33, 35, 38, 48, 55, 55, 55, 56, and 64.
The three workers who were laid of were ages 55, 55, and 64. Display 1.9 shows
these data on a dot plot.
Display 1.9 Hourly workers: ages of those laid off and those
retained in Round 2.
Use a summary statistic

to condense the data.
To simplify the statistical analysis to come, it helps to condense the data into
a single number, called a summary statistic. One possible summary statistic is
the average, or mean, age of the three workers who lost their jobs:
Knowing what to make of the data requires balancing two points of view. On
one hand, the pattern in the data is pretty striking. Of the five workers under age
50, all kept their jobs. Of the five who were 55 or older, only two kept their jobs.
On the other hand, the number of workers involved is small: only three out of ten.
Should you take seriously a pattern involving so few cases? Imagine two people
taking sides in an argument that was at the center of the statistical part of the
Martin case.
Martin: Look at the pattern in the data. All three of the workers laid
off were much older than the average age of all workers. Thats
evidence of age discrimination.
Westvaco: Not so fast! Youre looking at only ten workers total, and only
three positions were eliminated. Just one small change and the
picture would be entirely di erent. For example, suppose it had
been the 25-year-old instead of the 64-year-old who was laid off.
Switch the 25 and the 64, and you get a totally different set of
averages. (Ages in red are those selected for layoff.)
Actual Data:
25
33
35
38
48
55
55
55
56
64
Altered Data:
25
33
35
38
48
55
55
55
56
64
23-03-2009 21:29
Lesson
3 de 9
See! Make just one small change, and the average age of the three
who were laid off is lower than the average age of the others.
Martin: Not so fast yourself! Of all the possible changes, you picked the
one most favorable to your side. If youd switched one of the
55-year-olds who got laid off with the 55-year-old who kept his
or her job, the averages wouldnt change at all. Why not compare
what actually happened with all the possibilities?
Westvaco: What do you mean?
Martin: Start with the ten workers, and pick three at random. Do this over
and over, to see what typically happens, and compare the actual
data with the results. then well find out how likely it is that the
average age of those laid off would be 58 or greater.
The dialogue between Martin and Westvaco describes one age-neutral
method for choosing which workers to lay off: Pick three workers completely at
random, with all sets of three having the same chance to be chosen.
Picking Workers at Random

D5. If you pick three of the ten ages at random, do you think you are likely to get
an average age of 58 or greater?
D6. If the probability of getting an average age of 58 or greater turns out to be
small, does this favor Martin or Westvaco?
Simulation requires a
chance model.
Activity 1.2a shows you how to estimate the probability of getting an average
age of 58 years or greater if you choose three workers at random. You will
use simulation, a procedure in which you set up a model of a chance process
(drawing three ages out of a box) that copies, or simulates, a real situation
(selecting three employees at random to lay off).
By Chance or by Design?
What youll need: paper or 3 5 cards, a box or other container
Lets test the process suggested by Martins advocate.
1. Create a model of a chance process. Write each of the ten ages on identical
pieces of paper or 3 5 cards, and put the ten cards in a box. Mix them
thoroughly, draw out three (the ones to be laid off), and record the ages.
(continued)
23-03-2009 21:29
Lesson
4 de 9
(continued)
2. Compute a summary statistic. Compute the average of the three numbers in
your sample to one decimal place.
3. Repeat the process. Repeat steps 1 and 2 nine more times.
4. Display the distribution. Pool your results with the rest of your class and
display the average ages on a dot plot.
5. Estimate the probability. Count the number of times your class computed
an average age of 58 years or greater. Estimate the probability that simply
by chance the average age of those chosen would be 58 years or greater.
6. Interpret your results. What do you conclude from your classs estimate in
step 5?
[See Calculator Note 1B to learn how to do this kind of simulation with your
calculator.]
The simulation tells what

kind of data to expect if
workers are selected at
random for layoff.
Your simulation was completely age-neutral. All sets of three workers

had exactly the same chance of being selected for layoff , regardless of age. The
simulation tells you what results are reasonable to expect from that sort of
age-blind process.
Shown here are the first 4 of 200 repetitions from such a simulation. (The
ages in red are those selected for layoff .) The average ages of the workers selected
for layoff 42.7, 48.0, 42.7, and 37.0are highlighted by the red dots in the
distribution of all 200 repetitions in Display 1.10.
Display 1.10 Results of 200 repetitions: the distribution of the

average age of the three workers chosen for layoff
by chance alone.
23-03-2009 21:29
Lesson
5 de 9
Here we take a closer

look at the logic.
Out of 200 repetitions, only 10, or 5%, gave an average age of 58 or greater. So
it is not at all likely that simply by chance youd pick workers as old as the three
Westvaco picked. Did the company discriminate? Theres no way to tell from the
numbers aloneWestvaco might have a good explanation. On the other hand,
if your simulation had told you that an average of 58 or greater is easy to get by
chance alone, then the data would provide no evidence of discrimination and
Westvaco wouldnt need to explain.
To better understand how this logic applies to Martin v. Westvaco, imagine a
realistic argument between the advocates for each side.
Martin: Look at the pattern in the data. All three of the workers laid off
were much older than average.
Westvaco: So what? You could get a result like that just by chance. If chance
alone can account for the pattern, theres no reason to ask us for
any other explanation.
Martin: Of course you could get this result by chance. The question is
whether its easy or hard to do so. If its easy to get an average
as large as 58 by drawing at random, Ill agree that we cant rule
out chance as one possible explanation. But if an average that
large is really hard to get from random draws, we agree that
its not reasonable to say that chance alone accounts for the
pattern. Right?
Westvaco: Right.
Martin: Here are the results of my simulation. If you look at the three
hourly workers laid o in Round 2, the probability of getting an
average age of 58 or greater by chance alone is only 5%. And if you
do the same computations for the entire engineering department,
the probability is a lot lower, about 1%. What do you say to that?
Westvaco: Well . . . Ill agree that its really hard to get an average age that
extreme simply by chance, but that by itself still doesnt prove
discrimination.
Martin: No, but I think it leaves you with some explaining to do!
In the actual case, Martin and Westvaco reached a settlement out of court
before the case went to trial.
The logic youve just seen is basic to all statistical inference, but its not easy
to understand. In fact, it took mathematicians centuries to come up with the
ideas. It wasnt until the 1920s that a brilliant British biological scientist and
mathematician, R. A. Fisher, realized that results of agricultural experiments
may be analyzed in a way similar to that in Activity 1.2a to see whether observed
differences should be attributed to chance alone or to treatment. Calculus, in
contrast, was first understood in 1665. Precisely because it is so important, the
logic of using randomization as a basis for statistical inference will be seen over
and over again throughout this book. Youll have lots of time to practice with it.
23-03-2009 21:29
Lesson
6 de 9
Key Steps in a Simulation

Here is a summary of the steps in a simulation:
1. Model. Set up a model in which chance is the only cause of being selected.
In Activity 1.2a the model for an age-neutral chance process was to draw
three numbers at random from the set of ten ages. You put the ages of the
workers who could be laid off in a box and selected three by random draw.
2. Repetition. Repeat the process.
In the activity, you repeated the process of drawing ages many times.
3. Distribution. Display the distribution of the summary statistics, and
determine how likely the actual result or one even more extreme would be.
In the activity, you used the average age to summarize the results, although
other summary statistics also could be used. For each repetition, you
computed the average age of those laid off and plotted that average on your
dot plot. The simulation showed that the chance of an average age of 58 or
greater was only about 0.05.
4. Conclusion. If the probability is small (and the definition of small will vary
depending on the situation), conclude that some explanation other than
just chance should be considered. If the probability isnt small, conclude
that you can reasonably attribute the result to chance alone.
In Round 2 of the Martin v. Westvaco case, the small probability of 0.05
(1 chance out of 20) meant that Westvaco had some explaining to do. However,
the Round 2 evidence alone wouldnt have been enough to serve as evidence of
discrimination in a court case (which requires a probability of 0.025 or less).
DISCUSSION
The Logic of Inference

D7. Why must you estimate the probability of getting an average age of 58
or greater rather than the probability of getting an average age of 58?
D8. How unlikely is too unlikely? The probability you estimated in Activity 1.2a
is in fact exactly equal to 0.05. In a typical court case, a probability of 0.025
or less is required to serve as evidence of discrimination.
a. Did the Round 2 layoffs of hourly workers in the Martin case meet the
court requirement?
b. If the probability in the Martin case had been 0.01 instead of 0.05, how
would that have changed your conclusions? 0.10 instead of 0.05?
D9. A friend wants to bet with you on the outcome of a coin flip. The coin looks
fair, but you decide to do a little checking. You flip the coin, and it lands
heads. You flip againalso heads. A third flipheads. Flipheads.
Flipheads. You continue and the coin lands heads 19 times in 20 flips.
a. Explain why the evidence19 heads in 20 flipsmakes it hard to believe
the coin is fair.
b. Design and carry out a simulation to estimate how unusual this result
would be if the coin were fair.
16
23-03-2009 21:29
Lesson
7 de 9
Summary 1.2: Inference

Inference is a statistical procedure that involves deciding whether an event can
reasonably be attributed to chance or whether you should look forand perhaps
investigatesome other explanation. In the Martin case, you used inference to
determine whether the relatively high average age of the laid-off hourly employees
in Round 2 could reasonably be due to chance.
Simulation is a useful device for inference.
First, you set up a model of a process in which chance is the only factor
influencing the outcomes.
The next stage is repetitionyou repeat the process.
Then you plot the distribution of the summary statistics in order to determine
how likely the actual result or one even more extreme would be.
Finally, you reachor infera conclusion. If the probability of getting a
summary statistic as extreme as that from your actual data is small, conclude
that chance isnt a reasonable explanation. If the probability isnt small,
conclude that you can reasonably attribute the result to chance alone.
In the Martin case, the probability was about 0.05, which was considered small
enough to warrant asking for an explanation from Westvaco but not small enough
to present in court as clear evidence of discrimination.
Practice
The Logic of Inference
P4. Suppose three workers were laid off from a
set of ten whose ages were the same as those
of the hourly workers in Round 2 in the
Martin case. This time, however, the ages of
those laid off were 48, 55, and 55.
25 33 35 38 48 55 55 55 56 64
a. Use the dot plot in Display 1.10 on page
14 to estimate the probability of getting
an average age as large as or larger than
that of those laid off in this situation.
b. What would your conclusion be if
Westvaco had laid off workers of these
three ages?
P5. At the beginning of Round 1, there were 14
hourly workers. Their ages were 22, 25, 33,
35, 38, 48, 53, 55, 55, 55, 55, 56, 59, and 64.
After the layoffs were complete, the ages
of those left were 25, 38, 48, and 56. Think
about how you would repeat Activity 1.2a
using these data.
a. What is the average age of the ten workers
laid off?
b. Describe a simulation for finding the

distribution of the average age of ten
workers laid off at random.
c. The results of 200 repetitions from a
simulation are shown in Display 1.11.
Suppose 10 workers are picked at random
for layoff from the 14 hourly workers.
Make a rough estimate of the probability
of getting, just by chance, the same or
larger average age as that of the workers
who actually were laid off (from part a).
d. Does this analysis provide evidence in
Martins favor?
Display 1.11 Results of 200 repetitions.
23-03-2009 21:29
Lesson
8 de 9
Exercises
E9. Revisit the idea of the simulation in Activity E10. The ages of the ten hourly workers left after
1.2a, this time for all 14 hourly workers and
Round 1 are given here. The ages of the four
using a different summary statistic. Use
workers laid off in Rounds 2 and 3 are shown
as your summary statistic the number of
in red. Their average age is 57.25.
hourly workers laid off who were 40 or older.
25 33 35 38 48 55 55 55 56 64
The ages listed here are those of the hourly
a. Describe how to simulate the chance of
workers, with the ages of those laid off in
getting an average age of 57.25 or more
red. Note that, of the ten hourly workers laid
using the methods of Activity 1.2a.
off by Westvaco, seven were age 40 or older.
b. Perform your simulation once and
22 25 33 35 38 48 53
compute the average age of the four
55 55 55 55 56 59 64
hourly workers laid off.
a. Write the 14 ages on 14 slips of paper
c. The dot plot in Display 1.13 shows
and draw 10 at random to be chosen for
the results of 200 repetitions of this
layoff. How many of the 10 are age 40 or
simulation. What is your estimate of the
older?
probability of getting an average age as
b. The dot plot in Display 1.12 shows the
great as or greater than Westvaco did if
results of 50 repetitions of this simulation.
four workers are picked at random for
Estimate the probability that, by chance,
layoff in Rounds 2 and 3 from the ten
seven or more of the ten hourly workers
hourly workers remaining after Round 1?
who were laid off would be age 40 or older.
Display 1.13 Results of 200 repetitions: mean ages

of three randomly selected workers.
Display 1.12 Results of 50 repetitions: the

distribution of the number of workers
age 40 or older from ten randomly
selected workers.
c. Do you conclude that the proportion of

laid-off workers age 40 or older could
reasonably be due to chance alone,
or should Westvaco be asked for an
explanation?
18
d. What is your conclusion?

E11. Mrs. Garcia was not happy when she found
that her baker had raised the price of a loaf
of breadand she let him know it. However,
she did buy her usual three loaves of bread.
They seemed a little light, so she asked that
they be weighed and that the other eight
loaves the baker could have given her also
be weighed. The other eight loaves weighed
14, 15, 15, 16, 16, 17, 17, and 18 ounces. The
three loaves Mrs. Garcia was given weighed
23-03-2009 21:29
Lesson
9 de 9
14, 15, and 16 ounces. The baker claimed

that he picked the three loaves at random.
In Display 1.14, each dot represents the
average weight of a random sample of three
loaves.
Display 1.14 Results of 200 repetitions: average

weight of three randomly selected
loaves.
Which of these conclusions should Mrs.

Garcia draw?
A. Because the probability of getting an
average weight as low as or lower than that
of Mrs. Garcias three loaves is small, Mrs.
Garcia should not be suspicious that the
baker deliberately gave her lighter loaves.
B. Because the probability of getting an
average weight as low as or lower than
that of Mrs. Garcias three loaves is small,
Mrs. Garcia should be suspicious that the
C. Because the probability of getting an
average weight as low as or lower than
that of Mrs. Garcias three loaves is
fairly large, Mrs. Garcia should not be
suspicious that the baker deliberately gave
her lighter loaves.
D. Because the probability of getting an
average weight as low as or lower than that
of Mrs. Garcias three loaves is fairly large,

Mrs. Garcia should be suspicious that the
E12. Snow in July? You have spent some time in
Oz. You think the date back in Kansas is
July 4, but you cant be sure because days
might not have the same length in Oz as
on Earth. A friendly tornado puts you and
your dog Toto down in Kansas. However,
you see snow falling (data). Which of these
inferences should you make?
A. If this is Kansas, it is very unlikely to
be snowing on July 4. Therefore, this
probably isnt Kansas.
B. If it is July 4, it is very unlikely to be
snowing in Kansas. Therefore, this
probably isnt July 4.
C. If it is snowing in Kansas on July 4, it is
time to go back to Oz.
D. If this is Kansas and it is July 4, it
probably isnt really snowing.
E13. For some situations, instead of using
simulation, it is possible to find exact
probabilities by counting equally likely
outcomes. Suppose only two out of the ten
hourly workers had been laid off in Round
2 and that those two workers were ages 55
and 64, with an average age of 59.5. It is
straightforward, though tedious, to list all
possible pairs of workers who might have
been chosen. Heres the beginning of a
systematic listing. The first nine outcomes
include the 25-year-old and one other. The
next eight outcomes include the 33-year-old
and one other, but not the 25-year-old
because the pair {25, 33} was already counted.
23-03-2009 21:29
Lesson
1 de 6
a. How many possible pairs are there?

(Dont list them all!)
b. How many pairs give an average age of
59.5 or greater? (Do list them.)
c. If the pair is chosen completely at
random, then all possibilities are equally
likely and the probability of getting an
average age of 59.5 or greater equals the
number of pairs with an average of 59.5
or more divided by the total number of
possible pairs. What is the probability?
d. Is the evidence of age discrimination
strong or weak?
E14. In this exercise, you will follow the same
steps as in E13 to find the probability of
getting an average age of 58 or greater when
drawing three hourly workers at random in
Round 2. The number of ways to pick three

different workers from ten to lay off is
[See Calculator Note 1C to learn how to

compute numbers of combinations.]
a. List the ways that give an average age of

58 or greater.
b. Compute the probability of getting an
average age of 58 or greater when three
workers are selected for layoff at random.
c. How does this probability compare to the
results of your class simulation in Activity
1.2a? Why do the two probabilities differ
(if they do)?
Chapter Summary
In this chapter, you explored the data from an actual case of alleged age
discrimination, looking for evidence you considered relevant. You then saw how
to use statistical reasoning to test the strength of the evidence: Are the patterns
in the data solid enough to support Martins claim of age discrimination, or are
they the sort that you would expect to occur even if there was no discrimination?
Along the way you made a substantial start at learning many of the most
important statistical terms and concepts: distribution, cases and variables,
summary statistic, simulation, and how to determine whether the result from
the real-life situation can reasonably be attributed to chance alone or whether an
explanation is called for.
You have practiced both thinking like a statistician and reporting your results
like a statistician. Throughout this textbook, you will be asked to justify your
answers in the real-world context. This includes stating assumptions, giving
appropriate plots and computations, and writing a conclusion in context.
The last chapter of this book includes a final look at the Martin case.
Review Exercises
E15. A teacher had two statistics classes, and
students could enroll in either the earlier
class or the later class. Final grades in the
courses are given here.
Earlier class:
99 95 69
91 79 67 64 54 68
47 53 86 100 95 45 41 59 66
Later class:
84 68 94 77 88 75 88 91 83
61 97 75 37 82 62 49 43 93
23-03-2009 21:30
Lesson
2 de 6
a. Display these data on dot plots so that

equal to the number of people age 50 or
you can compare the two classes.
older who were retained.
b. What conclusion can you draw from the
E. The table does not support a claim of
dot plots? Could the difference between
age discrimination because a larger
the two classes reasonably be attributed
percentage of people were 50 or older to
to chance, or do you think the teacher
begin with.
should look for an explanation?
E18. Display 1.15 contains information about
E16. Refer to the data in E15.
the planets in our solar system. The radius
of each planet is given in miles, and the
a. Make a table that divides the course
temperature is the average at the surface.
grades into fail (less than 60) and pass
What are the cases? What are the variables?
(60 or more) for the two classes.
b. What proportion of the students in the
earlier class passed? What proportion of
students who passed were in the earlier
class? What proportion of students
passed overall?
compute and compare in order to decide
whether a disproportionate number of
passing students enrolled in the earlier
class? Make these computations and give
your conclusion.
Display 1.15 Data about planets in our solar system.
E17. This table classifies the Westvaco workers by
[Source: solarsystem.nasa.gov.]
whether they were laid off and whether they
E19. Earlier you studied the summary table of
were under age 50 or were age 50 or older.
salaried workers classi ed according to age
and to whether they were laid off or retained,
using 50 as the dividing age. That table is
shown again here.
Choose the best conclusion to draw from
this table.
A. The table supports a claim of age
discrimination because most of the
people who were laid off were 50 or older.
B. The table supports a claim of age
discrimination because a larger
percentage of the people age 50 or older
were laid off than people under 50.
C. The table supports a claim of age
discrimination because a larger
percentage of the laid-off people were
age 50 or older than were under 50.
D. The table does not support a claim of age
discrimination because the number of
people under age 50 who were laid off is
The proportion of those under age 50 who

were laid off (6 out of 16, or 0.375) is smaller
than the proportion of those age 50 or older
who were laid off (12 out of 20, or 0.60).
The key question, however, is Is the actual
12 versus 6 split of those laid off consistent
with selecting workers at random for layoff ?
a. Which of these demonstrations would
help Martins case?
A. Showing that its not unusual to get
12 or more older workers if 18 workers
are selected at random for layoff
Chapter Summary
21
23-03-2009 21:30
Lesson
3 de 6
d. From this simulation, a total of 26 out

B. Showing that its pretty unusual to get
12 or more older workers if 18 workers
of 200 repetitions gave counts of 50
are selected at random for layoff
or older that were 12 or more. What
percentage of the repetitions is this? Is
b. To investigate this situation using
this percentage small enough to cast
simulation, follow these steps once and
serious doubt on a claim that those laid
record your results.
off were chosen by chance?
1. Make 36 identical white cards. Label
E20.
The
Eastbanko Company had fifteen workers
20 cards with an O for 50 or over
before laying off five of them. The ages of the
and 16 cards with a U for under 50.
fifteen workers were 22, 23, 25, 31, 34, 36,
2. Mix the cards in a bag and select the
37, 40, 41, 43, 44, 50, 55, 55, and 60, with the
18 to be laid off at random.
ages of the five laid-off workers in bold.
3. Count the number of Os among the
The dot plot in Display 1.17 gives the results
18 selected.
of a simulation with 600 repetitions for the
c. Display 1.16 shows the results of a
average age of five of these workers chosen
computer simulation of 200 repetitions
for layoff at random. Each dot represents
conducted according to the rules given
the average of five ages, rounded down to a
in part b. Where would your result from
whole number.
part b be placed on this dot plot?

distribution of the average age of
five randomly chosen workers.

distribution of the number of Os in
18 randomly selected cards.
23-03-2009 21:30
Lesson
4 de 6
a. Estimate the probability that if workers

E23. The Society for the Preservation of Wild
Gnus held a raffle last week and sold 50
were selected by chance alone for layoff ,
the average age of those laid off would be
tickets. The two lucky participants whose
as large as or larger than the average age
tickets were drawn received all-day passes
of those in the actual layoffs.
to the Wild Gnu Park in Florida. But there
was a near riot when the winners were
b. If the 60-year-old sues for age
announcedboth winning tickets belonged
discrimination, would Eastbanko have
to society president Filbert Newmans
some explaining to do?
cousins. After some intense questioning
E21. In E9, you conducted a simulation to
by angry ticket holders, it was determined
estimate the probability that, just by chance,
that only 4 of the 50 tickets belonged to
seven or more of the ten hourly workers
Newmans cousins and the other 46 tickets
who were laid off would be age 40 or older.
belonged to people who were not part of his
The ages of the fourteen hourly workers were
family. Newmans final comment to the press
22, 25, 33, 35, 38, 48, 53, 55, 55, 55, 55, 56,
was Hey kids, I guess we were just lucky.
59, and 64.
Deal with it.
a. How many ways can 10 workers be
One member of the Gnu Society was taking
selected from 14 workers for layoff ?
a statistics class and decided to deal with it
b. If there are a total of 10 layoff s, what
by simulating the drawing. He put 50 tickets
numbers of older workers would it have
in a bowl; 4 of the tickets were marked C
been possible to lay off ?
for cousin and 46 were marked N for
c. Using your calculator, find the number of
not a cousin. The statistics student drew
ways that you can lay off
two tickets at random and kept track of the
number of cousins picked. A er doing this
i. seven older workers and three
1000 times, the student found that 844 draws
younger workers
resulted in two Ns, 149 in one N and one C,
ii. eight older workers and two younger
and only 7 in two Cs.
workers
a. Use the results of the simulation to
iii. nine older workers and one younger
estimate the probability that, in a fair
worker
drawing, both winning tickets would be
d. What is the probability that you will get
held by Newmans cousins.
7 or more workers age 40 or older if you
b.
Using the probability you estimated
select 10 of the 14 workers completely at
in
part a, write a short paragraph that
random for layoff ?
the statistics student can send to other
E22. Refer to your reasoning in E14, where you
members of the Gnu Society.
computed the probability that the three
c.
Is
it possible that Newmans cousins won
workers laid off in Round 2 would have an
the
prizes by chance alone? Explain.
average age of 58 or greater. Describe how
d. Using reasoning like that in E13 and E14,
your reasoning and conclusions would be
compute the exact probability that, in a
different if the workers ages were 25, 33, 35,
fair drawing, both winning tickets would
38, 48, 55, 55, 55, 55, and 55, and the three
be held by Newmans cousins.
workers chosen for layoff were all age 55. Is
the evidence stronger or weaker for Martin
in this situation than in E14?
Chapter Summary
23
23-03-2009 21:30
Lesson
5 de 6
No, because half of hourly workers were

AP1. This plot shows the ages of the part-time
and full-time students who receive financial
laid off, but more than half of salaried
aid at a small college. Which of the following
workers were laid off.
is a conclusion about students at this college AP3. This table shows the number of male
that cannot be drawn from the plot alone?
and female applicants who applied and
were either admitted to or rejected from
a graduate program. What proportion of
admitted applicants were female?
Part-time students who receive financial

aid tend to be older than full-time
students who receive financial aid.
A larger proportion of part-time
students than full-time students receive
financial aid.
The oldest student receiving financial aid
is a full-time student.
No student under age 18 receives
financial aid.
More part-time students than full-time
students receive financial aid.
AP2. This table classifies hourly and salaried
workers as to whether they were laid off .
Do the data support a claim that hourly
workers are being treated unfairly?
Yes, because most people laid off were

hourly workers.
Yes, because a bigger proportion of
hourly workers were laid off than
salaried workers.
No, because half of hourly workers were
laid off and half were not.
No, because more than half of workers
were hourly and less than half salaried.
AP4. For the data in AP3, in order to determine

if there is evidence to continue investigating
whether the graduate admissions process
discriminates against females, a study
takes a random sample of 25 out of the
70 applicants to be the admitted group.
The proportion of females in the sample
was computed. This process was repeated
for a total of 50 random samples and the
results are graphed below. What is the best
conclusion to draw from this simulation?
The actual proportion of females among

those admitted is very near the center of
this distribution, so there is no evidence
of discrimination.
those admitted is very near the center
of this distribution, so there is strong
evidence of discrimination in favor of
female applicants.
23-03-2009 21:30
Lesson
6 de 6
b. Construct dot plots for compounds

A and B. Does it now appear that
compound B tends to give larger
measurements than compound A?
c. Find the average increase in lung
capacity for compound A and for
compound B. When you compare
these means, does it look to you as if
compound B is better than compound A
at opening up the lungs?
AP6. Refer to AP5. Your task now is to see
whether the observed difference in the
means of each treatment group reasonably
could be attributed to chance alone.
a. Place the ten measurements on separate
slips of paper and mix them in a bag.
Investigative Tasks
Select five at random to play the role of
the A treatment group; the other five
AP5. People with asthma often use an inhaler
play the role of the B treatment group.
to help open up their lungs and breathing
This time you will use as your summary
passages. A pharmaceutical company has
statistic the difference between the
come up with a new compound to put in
means of each treatment group. Calculate
the inhaler that, they believe, will open
this difference, mean (compound B)
up the lungs of the user even more than
mean
(compound A), for your sample.
the standard compound tends to do. Ten
b.
The
dot
plot in Display 1.19 shows
volunteers with asthma are randomly split
the results of 50 repetitions of this
into two groups: one group uses the new
simulation. Compute the difference
compound B and the other uses the standard
between the means for the actual data.
compound A. The measurements listed in
Locate this difference on the dot plot.
Display 1.18 are the increase in lung capacity
How many simulated differences exceed
(in liters) 1 hour after the use of the inhaler.
the actual difference? What proportion?
those admitted is very near the center
of this distribution, so there is strong
evidence of discrimination against
female applicants.
those admitted is quite a bit above the
center of this distribution, so there is
strong evidence of discrimination in
favor of female applicants.
those admitted is quite a bit below the
center of this distribution, so there
is strong evidence of discrimination
against female applicants.
Display 1.18 Increase in lung capacity, in liters,

1 hour after use of an inhaler
containing
a compound.
a. From simply studying the data in the

table, do you think compound B does
better than compound A in increasing
lung capacity?

distribution of the difference between
the means of two randomly selected
groups of five values.
c. In light of this simulation, do you think

it is reasonable to attribute the actual
difference to chance alone? Explain.
AP Sample Test 25
23-03-2009 21:30
Lesson
1 de 2
Exploring Distributions
What does the

distribution of female
heights look like?
Statistics gives you
the tools to visualize
and describe large sets
of data.
23-03-2009 21:31
Lesson
2 de 2
Raw dataa long list of valuesare hard to make sense of. Suppose, for
example, that you are applying to the University of Michigan at Ann Arbor and
wonder how your SAT I math score of 650 compares with those of the students
who attend that university. If all you have are raw dataa list of the SAT I math
scores of the 25,000 students at the University of Michiganit would take a lot
of time and effort to make sense of the numbers.
Suppose instead that you read the summary in the universitys guide, which
says that the middle 50% of the SAT math scores were between 630 and 720, with
half above 680 and half below. Now you know that your score of 650, though in
the bottom half of the scores, is not far from the center value of 680 and is above
the bottom quarter.
Notice that the summary of the scores gives you two different kinds of
information for the middle 50%: the center, 680, and the spread, from 630 to 720.
Often thats all you need, especially if the shape of the distribution is one of a few
standard shapes youll learn about in this chapter.
These three featuresshape, center, and spreadcan sometimes take
you a surprisingly long way in data analysis. For example, in Chapter 1 you did
a simulation to answer this question: If you choose three people at random
from a set of ten people and compute the average age of the ones you choose,
how likely is it that you get an average of 58 years or more? Generally, you
dont need to do all this work! Using shape, center, and spread, you can get an
answer without doing a simulation. This remarkable fact, which first began to
come to light in the late 1600s, helped make statistical inference possible in the
20th century before the age of computers. In the next several chapters, youll learn
how to make good use of this fact.
In this chapter, you begin your systematic study

of distributions by learning how to
make and interpret different kinds of plots

describe the shapes of distributions
choose and compute a measure of center
choose and compute a useful measure of spread (variability)
work with the normal distribution
23-03-2009 21:31
2.1
Visualizing Distributions: Shape, Center, and Spread

Summaries simplify. In fact, summaries sometimes can oversimplify, which means
it is important to know when to use summaries and which summaries to use.
Often the right choice depends on the shape of your distribution. To help you
build your visual intuition about how shape and summaries are related, this first
section of the chapter introduces various shapes and asks you to estimate some
summary values visually. Later sections will tell you how to compute summary
values numerically.
Distributions come in a variety of shapes. Four of the most common shapes
are illustrated in the rest of this section.
Uniform (Rectangular) Distributions

The uniform distribution
is rectangular.
The plot to the left shows the shape of a uniform or rectangular distribution, in
which all values occur equally often. How uniform is a sample of values taken
from a uniform distribution? In the next activity, you will find out.
Distributing Digits
What youll need: one page from a phone book for each member of the class,
and a box of slips of paper, with one slip for each member of your class, half
labeled phone book and half fake it
1. Suppose your class made a dot plot of the last digits of every phone
number in the phone book. (This would take a very long time!) Sketch what
you think this plot would look like.
2. Draw a slip of paper from the box.
If the slip you drew says phone book, use the page from the phone
book, start at a random spot, and write down the last digit of each of the
next 30 phone numbers. Using a full sheet of paper, plot your 30 digits
on a dot plot, using a scale like the one here. Use big dots so that they
can be seen from across the room.
If the slip you drew says fake it, dont use the page from the phone
book but instead make up and plot 30 digits on a dot plot using a scale
(Continued)
28 Chapter 2 Exploring Distributions
like the one on the previous page. Try to make the distribution look like
the digits might have come from the phone book.
Write your name on the front of the plot, but not which method you
used. Dont consult with other students while you are doing this step.
3.Post the dot plots around the room and compare. Which plots are you
confident came from the phone book? Which are you confident came from
made-up digits? (Dont say anything about your own plot.)
4.Find your plot and write a large P (for phone book) or F (for faked it)
on the front. Check your predictions from step 3. What differences do you
see in the two groups of plots?
5.In this activity, it was important that you sampled from the phone book
in such a way that all digits were equally likely to occur. Why did step 1
specify that you use the last digit of the phone number and not, say,
the first?
The number of births per month in a year is another set of data you might expect
to be fairly uniform. Or, is there a reason to believe that more babies are born in
one month than in another? Display 2.1 shows a table and plot of U.S. births (in
thousands) for 2003.
Display 2.1
An example of a (roughly) uniform distribution:

births per month in the United States, 2003.
[Source: Centers for Disease Control and Prevention.]
The plot shows that there is actually little change from month to month;
that is, we see a roughly uniform distribution of births across the months. To
summarize this distribution, you might write The distribution of births is
roughly uniform over the months January through December, with about
340,000 births per month.
29
Computers and many calculators generate random numbers between 0 and

1 with a uniform distribution. Display 2.2 shows a dot plot of 1000 random
numbers generated by statistical software. There is some variability in the
frequencies, but, as expected, about 20% of the random numbers fall between, for
example, 0.2 and 0.4. [See Calculator Note 2A to learn how to create a distribution of
random numbers using your calculator.]
Display 2.2
Dot plot of 1000 random numbers from a uniform

distribution. Each dot represents two points.
Uniform Distributions
D1. Think of other situations that you would expect to be uniform distributions
a. over the days of the week
b. over the digits 0, 1, 2, . . . , 9
D2. Think of situations that you would expect to be very nonuniform
distributions
a. over the months of the year
b. over the days of the month
c. over the digits 0, 1, 2, . . . , 9
d. over the days of the week
Normal Distributions
Activity 2.1b introduces one of the most important common shapes of
distributions and one of the common ways this shape is produced. What happens
when different people measure the same distance or the same feature of very
similar objects? In the activity, youll measure a tennis ball with a ruler, but the
results you get will reflect what happens even if you use very precise instruments
under carefully controlled conditions. For example, a 10-gram platinum weight
is used for calibration of scales all across the United States. When scientists at the
National Institute of Standards and Technology use an analytical balance for the
weights weekly weighing, they face a similar challenge due to variability.
Measuring Diameters
What youll need: a tennis ball, a ruler with a centimeter scale
1. With your partner, plan a method for measuring the diameter of the tennis
ball with the centimeter ruler.
(continued)
30
Chapter 2 Exploring Distributions
2. Using your method, make two measurements of

the diameter of your tennis ball to the nearest
millimeter.
3. Combine your data with those of the rest of the
class and make a dot plot. Speculate first, about
the shape you expect for the distribution.
4. Shape. What is the approximate shape of the
plot? Are there clusters and gaps or unusual
values (outliers) in the data?
5. Center and spread. Choose two numbers that
seem reasonable for completing this sentence:
Our typical diameter measurement is about ?, give or take about
?. (More than one reasonable set of choices is possible.)
6. Discuss some possible reasons for the variability in the measurements.
How could the variability be reduced? Can the variability be eliminated
entirely?
The normal distribution

is bell-shaped..
The measurements of the diameter of a tennis ball taken by your class in

Activity 2.1b probably were not uniform. More likely, they piled up around
some central value, with a few measurements far away on the low side and a
few far away on the high side. This common bell shape has an idealized version
the normal distributionthat is especially important in statistics.
Pennies minted in the United States are supposed to weigh 3.110 g, but a
tolerance of 0.130 g is allowed in either direction. Display 2.3 shows a plot of the
weights of 100 pennies.
Display 2.3
Weights of pennies. [Source: W. J. Youden, Experimentation and

Measurement (National Science Teachers Association, 1985), p. 108.]
The smooth curve superimposed on the graph of the pennies is an example

of a normal curve. No real-world example matches the curve perfectly, but many
plots of data are approximately normal. The idealized normal shape is perfectly
symmetricthe right side is a mirror image of the left side, as in Display 2.4.
There is a single peak, or mode, at the line of symmetry, and the curve drops off
smoothly on both sides, flattening toward the x-axis but never quite reaching it
and stretching infinitely far in both directions. On either side of the mode are
inflection points, where the curve changes from concave down to concave up.
2.1
31
Display 2.4 A normal curve, showing the line of symmetry, mode,

mean, inflection points, and standard deviation (SD).
Use the mean and
standard deviation to
describe the center
and spread of a normal
distribution.
You should use the mean (or average) to describe the center of a normal
distribution. The mean is the value at the point where the line of symmetry
intersects the x-axis. You should use the standard deviation, SD for short, to
describe the spread. The SD is the horizontal distance from the mean to an
inflection point.
It is difficult to locate inflection points, especially when curves are drawn
by hand. A more reliable way to estimate the standard deviation is to use areas.
For a normal curve, 68% (roughly) of the total area under the curve is between
the vertical lines through the two inflection points. In other words, the interval
between one standard deviation on either side of the mean accounts for roughly
68% of the area under the normal curve.
Example: Averages of Random Samples

Display 2.5 shows the distribution of average ages computed from 200 sets of
five workers chosen at random from the ten hourly workers in Round 2 of the
Westvaco case discussed in Chapter 1. Notice that, apart from the bumpiness, the
shape is roughly normal. Estimate the mean and standard deviation.
Display 2.5 Distribution of average age for groups of five workers

drawn at random.
32
Solution
The curve in the display has center at 47, and the middle 68% of dots fall roughly
between 43 and 51. Thus, the estimated mean is 47, and the estimated standard
deviation is 4. A typical random sample of five workers has average age 47 years,
give or take about 4 years.
[You can graph a normal curve on your calculator by specifying the mean and standard
deviation. See Calculator Note 2B.]
In this section, youve seen the three most common ways normal distributions
arise in practice:
through variation in measurements (diameters of tennis balls)
through natural variation in populations (weights of pennies)
through variation in averages computed from random samples (average ages)
All three scenarios are common, which makes the normal distribution especially
important in statistics.
D3. Determine these summary statistics visually.
a. Estimate the mean and standard deviation of the penny weight data in
Display 2.3, and use your estimates to write a summary sentence.
b. Estimate the mean and standard deviation of your class data from
Activity 2.1b.
Skewed Distributions
Skewed left
Both the uniform (rectangular) and normal distributions are symmetric. That
is, if you smooth out minor bumps, the right side of the plot is a mirror image
of the left side. Not all distributions are symmetric, however. Many common
distributions show bunching at one end and a long tail stretching out in the other
direction. These distributions are called skewed. The direction of the tail tells
whether the distribution is skewed right (tail stretches right, toward the high
values) or skewed left (tail stretches left , toward the low values).
Skewed right
Display 2.6 Weights of bears in pounds. [Source: MINITAB data set from
MINITAB Handbook, 3rd ed.]
The dot plot in Display 2.6 shows the weights, in pounds, of 143 wild bears. It is
skewed right (toward the higher values) because the tail of the distribution stretches
out in that direction. In everyday conversation, you might describe the two parts of
2.1
33
the distribution as normal and abnormal. Usually, bears weigh between about
50 and 250 lb (this part of the distribution even looks approximately normal), but
if someone shouts Abnormal bear loose! you should run for coverthat unusual
bear is likely to be big! The unusualness of the distribution is all in one direction.
Often the bunching in a skewed distribution happens because values bump
up against a walleither a minimum that values cant go below, such as 0 for
measurements and counts, or a maximum that values cant go above, such as 100
for percentages. For example, the distribution in Display 2.7 shows the grade-point
averages of college students (mostly first-year students and sophomores) taking an
introductory statistics course at the University of Florida. It is skewed left (toward
the smaller values). The maximum grade-point average is 4.0, for all As, so the
distribution is bunched at the high end because of this wall. A GPA of 0.0 wouldnt
be called a wall, even though GPAs cant go below 0.0, because the values arent
bunched up against it. The skew is to the left : An unusual GPA would be one that
is low compared to most GPAs of students in the class.
Display 2.7 Grade-point averages of 62 statistics students. Each

dot represents two points.
Use the median along
with the lower and
upper quartiles to
describe the center
and spread of a skewed
distribution.
Because there is no line of symmetry in a skewed distribution, the ideas of

center and spread are not as clear-cut as they are for a normal distribution. To get
around this problem, typically you should use the median to describe the center
of a skewed distribution. To estimate the median from a dot plot, locate the value
that divides the dots into two halves, with equal numbers of dots on either side.
You should use the lower and upper quartiles to indicate spread. The lower
quartile is the value that divides the lower half of the distribution into two halves,
with equal numbers of dots on either side. The upper quartile is the value that
divides the upper half of the distribution into two halves, with equal numbers
of dots on either side. The three valueslower quartile, median, and upper
quartiledivide the distribution into quarters. This allows you to describe a
distribution as in the introduction to this chapter: The middle 50% of the SAT
math scores were between 630 and 720, with half above 680 and half below.
Example: Median and Quartiles for Bear Weights

Divide the bears weights in Display 2.8 into four groups of equal size, and
estimate the median and quartiles. Write a short summary of this distribution.
Solution
There are 143 dots in Display 2.8, so there are about 71 or 72 dots in each half and
35 or 36 in each quarter. The value that divides the dots in half is about 155 lb.
The values that divide the two halves in half are roughly 115 and 250. Thus, the
middle 50% of the bear weights are between about 115 and 250, with half above
about 155 and half below.
34
Display 2.8 Estimating center and spread for the weights

of bears.
D4. Decide whether each distribution described will be skewed. Is there a wall
that leads to bunching near it and a long tail stretching out away from it? If
so, describe the wall.
a. the sizes of islands in the Caribbean
b. the average per capita incomes for the nations of the United Nations
c. the lengths of pant legs cut and sewn to be 32 in. long
d. the times for 300 university students of introductory psychology to
complete a 1-hour timed exam
e. the lengths of reigns of Japanese emperors
D5. Which would you expect to be the more common direction of skew, right or
left ? Why?
Bimodal Distributions
A bimodal distribution
has two peaks.
Many distributions, including the normal distribution and many skewed

distributions, have only one peak (unimodal), but some have two peaks
(bimodal) or even more. When your distribution has two or more obvious peaks,
or modes, it is worth asking whether your cases represent two or more groups.
For example, Display 2.9 shows the life expectancies of females from countries
on two continents, Europe and Africa.
Display 2.9 Life expectancy of females by country on two

continents. [Source: Population Reference Bureau, World
Population Data Sheet, 2005.]
2.1
35
Europe and Africa differ greatly in their socioeconomic conditions, and the
life expectancies reflect those conditions. If you make a separate plot for each of
the two continents, the two peaks become essentially one peak in each plot, as in
Display 2.10.
Display 2.10 Life expectancy of females in Africa and Europe.
Although it makes sense to talk about the center of the distribution of life
expectancies for Europe or for Africa, notice that it doesnt really make sense to
talk about the center of the distribution for both continents together. You could
possibly tell the locations of the two peaks, but finding the reason for the two
modes and separating the cases into two distributions communicates even more.
Other Features: Outliers, Gaps, and Clusters

An unusual value, or outlier, is a value that stands apart from the bulk of the data.
Outliers always deserve special attention. Sometimes they are mistakes (a typing
mistake, a measuring mistake), sometimes they are atypical for other reasons
(a really big bear, a faulty lab procedure), and sometimes unusual features of the
distribution are the key to an important discovery.
In the late 1800s, John William Strutt, third Baron Rayleigh (English,
18421919), was studying the density of nitrogen using samples from the air
outside his laboratory (from which known impurities were removed) and samples
produced by a chemical procedure in his lab. He saw a pattern in the results that
you can observe in the plot of his data in Display 2.11.
Lord Rayleigh
Display 2.11 Lord Rayleighs densities of nitrogen. [Source:

Proceedings of the Royal Society 55 (1894).]
Lord Rayleigh saw two clusters separated by a gap. (There is no formal

definition of a gap or a cluster; you have to use your best judgment about them.
For example, some people call a single outlier a cluster of one; others dont. You
also could argue that the value at the extreme right is an outlier, perhaps because
of a faulty measurement.)
When Rayleigh checked the clusters, it turned out that the ten values to
the left had all come from the chemically produced samples and the nine to the
right had all come from the atmospheric samples. What did this great scientist
conclude? The air samples on the right might be denser because of something
in them besides nitrogen. This hypothesis led him to discover inert gases in the
atmosphere.
Summary 2.1: Visualizing Distributions

Distributions have different shapes, and different shapes call for different
summaries.
If your distribution is uniform (rectangular), its often enough simply to tell
the range of the set of values and the approximate frequency with which each
value occurs.
If your distribution is normal (bell-shaped), you can give a good summary
with the mean and the standard deviation. The mean lies at the center of
the distribution, and the standard deviation (SD) is the horizontal distance
from the center to the points of inflection, where the curvature changes. To
estimate the SD, find the distance on either side of the mean that defines the
interval enclosing about 68% of the cases.
If your distribution is skewed, you can give the values (median and quartiles)
that divide the distribution into fourths.
If your distribution is bimodal, reporting a single center isnt useful. One
reasonable summary is to locate the two peaks. However, it is even more
useful if you can find another variable that divides your set of cases into two
groups centered at the two peaks.
Later in this chapter, you will study the various measures of center and spread
in more detail and learn how to compute them.
Practice
Practice problems help you master basic concepts
and computations. Throughout this textbook, you
should work all the practice problems for each
topic you want to learn. The answers to all practice
problems are given in the back of the book.
Uniform (Rectangular) Distributions
P1. This diagram shows a uniform distribution
on [0, 2], the interval from 0 through 2.
a. What value divides the distribution in

half, with half the numbers below that
value and half above?
b. What values divide the distribution into
quarters?
c. What values enclose the middle 50% of
the distribution?
d. What percentage of the values lie between
0.4 and 0.7?
e. What values enclose the middle 95% of
the distribution?
37
P2. The plot in Display 2.12 gives the number

of deaths in the United States per month in
2003. Does the number of deaths appear to
be uniformly distributed over the months?
Give a verbal summary of the way deaths are
distributed over the months of the year.
a. SAT math scores

b. ACT scores
c. heights of women attending college
d. single-season batting averages for
professional baseball players in the 1910s
P4. Estimate the median and quartiles for the
distribution of GPAs in Display 2.7 on
page 34. Then write a verbal summary of
the same form as in the example.
P5. Match each plot in Display 2.14 with its
median and quartiles (the set of values that
divide the area under the curve into fourths).
a. 15, 50, 85
b. 50, 71, 87
c. 63, 79, 91
d. 35, 50, 65
e. 25, 50, 75
Display 2.12 Deaths per month, 2003. [Source: Centers

for Disease Control and Prevention, National
Vital Statistics Report, 2004.]
P3. For each of the normal distributions
in Display 2.13, estimate the mean and
standard deviation visually, and use your
estimates to write a verbal summary of the
form A typical SAT score is roughly (mean),
give or take (SD) or so.
Display 2.14 Five distributions with different

shapes.
Display 2.13 Four distributions that are
approximately normal.
Exercises
E1. Describe each distribution as bimodal,
skewed right, skewed left , approximately
normal, or roughly uniform.
a. ages of all people who died last year in the
United States
b. ages of all people who got their first
drivers license in your state last year
c. SAT scores for all students in your state
taking the test this year
d. selling prices of all cars sold by General
Motors this year
E2. Describe each distribution as bimodal,
skewed right, skewed left , approximately
normal, or roughly uniform.
a. the incomes of the worlds 100 richest
people
b. the birthrates of Africa and Europe
c. the heights of soccer players on the last
Womens World Cup championship team
d. the last two digits of telephone numbers
in the town where you live
e. the length of time students used to
complete a chapter test, out of a
50-minute class period
b. a roughly normal distribution with mean

15 and standard deviation 5
c. a distribution that is skewed left , with half
its values above 20 and half below and
with the middle 50% of its values between
10 and 25
d. a distribution that is skewed right, with
the middle 50% of its values between 100
and 1000 and with half the values above
200 and half below
E4. The U.S. Environmental Protection Agencys
National Priorities List Fact Book tells
the number of hazardous waste sites for
each U.S. state and territory. For 2006, the
numbers ranged from 1 to 138, the middle
50% of the values were between 11 and 32,
half the values were above 18, and half were
below 18. Sketch what the distribution might
look like. [Source: U.S. Environmental Protection
Agency, www.epa.gov, 2006.]
E5. The dot plot in Display 2.15 gives the ages of

the officers who attained the rank of colonel
in the Royal Netherlands Air Force.
a. What are the cases? Describe the
variables.
b. Describe this distribution in terms of
shape, center, and spread.
c. What kind of wall might there be that
causes the shape of the distribution?
Generate as many possibilities as you can.
The 2003 Womens World Cup Championship

team, from Germany
E3. Sketch these distributions.

a. a uniform distribution that shows the sort
of data you would get from rolling a fair
die 6000 times
Display 2.15 Ages of colonels. Each dot represents

two points. [Source: Data and Story Library at
Carnegie-Mellon University, lib.stat.cmu.edu.]
39
E6. The dot plot in Display 2.16 shows the

distribution of the number of inches of
rainfall in Los Angeles for the seasons
18991900 through 19992000.
E9. Make up a scenario (name the cases and

variables) whose distribution you would
expect to be
a. skewed right because of a wall. What is
responsible for the wall?
b. skewed left because of a wall. What is
responsible for the wall?
E10. The plot in Display 2.18 shows the last
digit of the Social Security numbers of the
students in a statistics class. Describe this
distribution.
Display 2.16 Los Angeles rainfall.[Source: National

Weather Service.]
a. What are the cases? Describe the

variables.
b. Describe this distribution in terms of
c. What kind of wall might there be that
causes the shape of the distribution?
Generate as many possibilities as you can.
E7. The distribution in Display 2.17 shows
measurements of the strength in pounds of
22s yarn (22s refers to a standard unit for
measuring yarn strength). What is the basic
shape of this distribution? What feature
makes it uncharacteristic of distributions
with that shape?
Display 2.18 Last digit of a sample of Social Security

numbers.
E11. Although a uniform distribution gives a

reasonable approximation of the actual
distribution of births over months (Display
2.1 on page 29), you can blow up the graph
to see departures from the uniform pattern,
as in Display 2.19. Do these deviations from
the uniform shape form their own pattern,
or do they appear haphazard? If you think
theres a pattern, describe it.
Display 2.17 Strength of yarn. [Source: Data and

Story Library at Carnegie-Mellon University,
lib.stat.cmu.edu.]
E8. Sketch a normal distribution with mean 0

and standard deviation 1. You will study this
standard normal distribution in Section 2.5.
Display 2.19 A blow up of the distribution of births

over months, showing departures
from the uniform pattern.
E12.Draw a graph similar to that in Display 2.19

for the data on deaths in the United States
listed in Display 2.20, and summarize what
you find.
Display 2.20 Deaths in the United States, 2003.

[Source: Centers for Disease Control and
Prevention.]
E13.How do countries compare with respect to

the value of the goods they produce? Display
2.21 shows gross domestic product (GDP)
per capita, a measure of the total value of
all goods and services produced divided
by the number of people in a country, and
the average number of people per room in
housing units, a measure of crowdedness, for
a selection of countries in Asia, Europe, and
North America. Youll analyze these data in
parts ad.
Display 2.21 Per capita GDP and crowdedness for

a selection of countries. [Source: United
Nations, unstats.un.org.]
A family in Albania
41
A dot plot of the per capita GDP data is

shown in Display 2.22.
d. Is it surprising to find clusters and gaps

in data that measure an aspect of the
economies of the countries?
E14.The dot plot in Display 2.23 gives a look
at how the countries listed in Display 2.21
compare in terms of the crowdedness of
their residents.
Display 2.22 Dot plot of per capita GDP.
a. Describe this distribution in terms of

b. Which two countries have the highest
per capita GDP? Do they appear to be
outliers?
c. A rather large gap appears near the
middle of the distribution. Which of the
two clusters formed by this gap contains
mostly Western European and North
American countries? In what part of the
world are most of the countries in the
other cluster?
2.2
Plots should present the
essentials quickly and
clearly.
Display 2.23 Dot plot of crowdedness.
a. Describe this distribution in terms of

b. Which countries appear to be outliers?
Are they the same as the countries that
appeared to be outliers for the per capita
GDP data?
c. Where on the dot plot is the cluster that
contains mostly Western European and
North American countries?
Graphical Displays of Distributions

As you saw in the previous section, the best way to summarize a distribution often
depends on its shape. To see the shape, you need a suitable graph. In this section,
youll learn how to make and interpret three kinds of plots for quantitative variables
(dot plot, histogram, and stemplot) and one plot for categorical variables (bar chart).
Cases and Variables, Quantitative and Categorical

Pet cats typically live about 12 years, but some have been known to live 28 years.
Is that typical of domesticated predators? What about domesticated nonpredators,
such as cows and guinea pigs? What about wild mammals? The rhinoceros, a
nonpredator, lives an average of 15 years, with a maximum of about 45 years.
The grizzly bear, a wild predator, lives an average of 25 years, with a maximum of
about 50 years. Do meat-eaters typically outlive vegetarians in the wild? Often you
can find answers to questions like these in a plot of the data.
Many of the examples in this section are based on the data about mammals in
Display 2.24. Each row (type of mammal) is a case. As you learned in Chapter 1,
the cases in a data set are the people, cities, mammals, or other items being studied.
42
Measurements and other properties of the cases are organized into columns, with one
column for each variable. Thus, average longevity and speed are variables, and, for
example, 30 mi/h is the value of the variable speed for the case grizzly bear.
Display 2.24 Facts on mammals. (Asterisks (*) mark missing

values.) [Source: World Almanac and Book of Facts, 2001, p. 237.]
43
Counts of how many and measurements of how much are called quantitative
variables. Speed is a quantitative variable because speed is measured on a numerical
scale. A variable that groups cases into categories is called a categorical variable.
Predator is a categorical variable because it groups the mammals into those who
eat other animals and those who dont. Although the categories are coded 1 if a
mammal preys on other animals and 0 if it does not, these numbers just indicate
the appropriate category and are not meant to be quantitative. In Display 2.24, the
asterisks (*) mark missing values.
Cases and Variables, Quantitative and Categorical

D6. Classify each variable in Display 2.24 as quantitative or categorical.
More About Dot Plots

Dot plots show
individual cases as dots.
Youve already seen dot plots, beginning in Chapter 1. As the name suggests, dot
plots show individual cases as dots (or other plotting symbols, such as x). When
you read a dot plot, keep in mind that different statistical software packages make
dot plots in different ways. Sometimes one dot represents two or more cases, and
sometimes values have been rounded. With a small data set, different rounding rules
can give different shapes.
Display 2.25 shows a dot plot of the speeds of the mammals from Display
2.24. The gap between the cheetahs speed and that of the other mammals shows up
clearly in the dot plot but not in the list of speeds in Display 2.24. Discoveries like
this demonstrate why you should always plot your data.
Display 2.25 Dot plot of the speeds of mammals.

When are dot plots most
useful?
As you saw in Section 2.1, a dot plot shows shape, center, and spread. Dot
plots tend to work best when
you have a relatively small number of values to plot
you want to see individual values, at least approximately
you want to see the shape of the distribution
you have one group or a small number of groups you want to compare
Histograms
Histograms show groups
of cases as bars.
A dot plot shows individual cases as dots above a number line. To make a
histogram, you divide the number line into intervals, called bins, and over
each bin construct a bar that has a height equal to the number of cases in that
bin. In fact, you can think of a histogram as a dot plot with bars drawn around
the dots and the dots erased. The height of the bar becomes a visual substitute
for the number of dots. The plot in Display 2.26 is a histogram of the mammal
speeds. Like the dot plot of a distribution, a histogram shows shape, center, and
spread. The vertical axis gives the number of cases (the frequency or count)
represented by each bar. For example, four mammals have speeds of from 30 mi/h
up to 35 mi/h.
Display 2.26 Histogram of mammal speeds.

Borderline values go in
the bar to the right.
Most calculators and statistical soft ware packages place a value that falls at the
dividing line between two bars into the bar to the right. For example, in Display
2.26, the bar going from 30 to 35 contains cases for which 30 speed < 35.
Changing the width of the bars in your histogram can sometimes change your
impression of the shape of the distribution. For example, the histogram of the
speeds of mammals in Display 2.27 has fewer and wider bars than the histogram
in Display 2.26. It shows a more symmetric, bell-shaped distribution, and there
appears to be one peak rather than two. There is no right answer to the question
of which bar width is best, just as there is no rule that tells a photographer
when to use a zoom lens for a close-up. Different versions of a picture bring out
different features. The job of a data analyst is to find a plot that shows important
features of the distribution.
Display 2.27 Speeds of mammals using a histogram with

wider bars.
45
You can use your calculator to quickly display histograms with different bar
widths. [See Calculator Note 2C.] Shown here are the mammal speed data. The
numbers below the calculator screens indicate the window settings (minimum x,
maximum x, x-scale, minimum y, maximum y, y-scale).
When are histograms

most useful?
Relative frequency
histograms show
proportions instead
of counts.
Histograms work best when

you have a large number of values to plot
you dont need to see individual values exactly
you want to see the general shape of the distribution
you have only one distribution or a small number of distributions you want to
compare
you can use a calculator or computer to make the plots for you
A histogram shows frequencies on the vertical axis. To change a histogram
into a relative frequency histogram, divide the frequency for each bar by the
total number of values in the data set and show these relative frequencies on the
vertical axis.
Example: Converting Frequencies to Relative Frequencies

Four of the 18 mammals have speeds from 30 mi/h up to 35 mi/h. Convert the
frequency 4 to a relative frequency.
Solution
Four out of 18 is
or approximately 0.22. So about 0.22 of the mammals have
speeds in this range.
Example: Relative Frequency of Life Expectancies

Display 2.28 shows the relative frequency
distribution of life expectancies for
203 countries around the world. How
many countries have a life expectancy
of at least 70 but less than 75 years? What
proportion of the countries have a life
expectancy of 70 years or more?
Display 2.28
Life expectancies of people in countries around the

world. [Source: Population Reference Bureau, World Population
Data Sheet, 2005.]
Solution
The bar including 70 years and up to 75 years has a relative frequency of about
0.30, so the number of countries with a life expectancy of at least 70 years but less
than 75 years is about 0.30 203, or about 61.
The proportion of countries with a life expectancy of 70 years or greater is the
sum of the heights of the three bars to the right of 70about 0.30 + 0.19 + 0.07,
or 0.56.
Histograms
D7. In what sense does a histogram with narrow bars, as in Display 2.26, give
you more information than a histogram with wider bars, as in Display 2.27?
In light of your answer, why dont we always make histograms with very
narrow bars?
D8. Does using relative frequencies change the shape of a histogram? What
information is lost and gained by using a relative frequency histogram rather
than a frequency histogram?
Stemplots
The plot in Display 2.29 is a stem-and-leaf plot, or stemplot, of the mammal
speeds. It shows the key features of the distribution and preserves all theoriginal
numbers.
Display 2.29
Stemplot of mammal speeds.
2.2
47
A stemplot shows cases

as digits.
In Display 2.29, the numbers on the left , called the stems, are the tens digits
of the speeds. The numbers on the right, called the leaves, are the ones digits of
the speeds. The leaf for the speed 39 mi/h is printed in bold. If you turn your book
90 counterclockwise, you will see that a stemplot looks something like a dot plot
or histogram; you can see the shape, center, and spread of the distribution.
The stemplot in Display 2.30 displays the same information, but with
split stems: Each stem from the original plot has become two stems. If the ones
digit is 0, 1, 2, 3, or 4, it is placed on the first line for that stem. If the ones digit is
5, 6, 7, 8, or 9, it is placed on the second line for that stem.
Display 2.30 Stemplot of mammal speeds, using split stems.
Spreading out the stems in this way is similar to changing the width of the
bars in a histogram. The goal here, as always, is to find a plot that conveys the
essential pattern of the distribution as clearly as possible.
You have compared two data distributions by constructing dot plots on
the same scale (see, for example, Display 2.10). Another way to compare two
distributions is to construct a back-to-back stemplot. Such a plot for the speeds of
predators and nonpredators is shown in Display 2.31. The predators tend to have
the faster speedsor, at least, there are no slow predators!
Display 2.31 Back-to-back stemplot of mammal speeds for

predators and nonpredators.
48
Usually, only two digits are plotted on a stemplot, one digit for the stem and
one digit for the leaf. If the values contain more than two digits, the values may
be either truncated (the extra digits simply cut off ) or rounded. For example, if
the speeds had been given to the nearest tenth of a mile, 32.6 mi/h could be either
truncated to 32 mi/h or rounded to 33 mi/h.
As with the other types of plots, the rules for making stemplots are flexible.
Do what seems to work best to reveal the important features of the data.
The stemplot of mammal speeds in Display 2.32 was made by statistical
software. Although it looks a bit different from the handmade plot in Display 2.30,
it is essentially the same. In the first two lines, N = 18 means that 18 cases were
plotted; N* = 21 means that there were 21 cases in the original data set for which
speeds were missing; and Leaf Unit = 1.0 means that the ones digits were graphed
as the leaves. The numbers in the left column keep track of the number of cases,
counting in from the extremes. The 2 on the left in the first line means that there
are two cases on that stem. If you skip down three lines, the 4 on the left means
that there are a total of four cases on the first four stems (11, 12, 20, and 25).
Display 2.32 Stem-and-leaf plot of mammal speeds made by

statistical software.
When are stem-and-leaf

plots most useful?
Stemplots are useful when

you are plotting a single quantitative variable
you have a relatively small number of values to plot
you would like to see individual values exactly, or, when the values contain
more than two digits, you would like to see approximate individual values
you want to see the shape of the distribution clearly
you have two (or sometimes more) groups you want to compare
Stemplots
D9. What information is given by the numbers in the bottom half of the far
left column of the plot in Display 2.32? What does the 2 in parentheses
indicate?
49
D10.How might you construct a stemplot of the data on gestation periods for the
mammals listed in Display 2.24? Construct the stemplot and describe the
shape of the distribution.
Do Units of Measurement Affect Your Estimates?
In this activity, you will see whether you and your class estimate lengths better
in feet or in meters.
1. Your instructor will split the class randomly into two groups.
2. If you are in the fist group, you will estimate the length of your classroom
in feet. If you are in the second group, you will estimate the length of the
classroom in meters (without estimating first in feet and then converting to
meters). Do this by looking at the length of the room; no pacing the length
of the room is allowed.
3. Find an appropriate way to plot the two data sets so that you can compare
their shapes, centers, and spreads.
4. Do the students in your class tend to estimate more accurately in feet or in
meters? What is the basis for your decision?
5. Why split the class randomly into two groups instead of simply letting the
left half of the room estimate in feet and the right half estimate in meters?
Bar Charts for Categorical Data
Bar charts show

the frequencies of
categorical data as
heights of bars.
You now have three different types of plots to use with quantitative variables.
What about categorical variables? You could make a dot plot, or you could make
what looks like a histogram but is called a bar chart or bar graph. There is one
bar for each category, and the height of the bar tells the frequency. A bar chart has
categories on the horizontal axis, whereas a histogram has measurementsvalues
from a quantitative variable.
The bar chart in Display 2.33 shows the frequency of mammals that fall into
the categories wild and domesticated, coded 1 and 0, respectively. (Note that
the bars are separated so that there is no suggestion that the variable can take on a
value of, say, 0.5.)
Display 2.33 Bar chart showing frequency of domesticated (0)

and wild (1) mammals.
50
The relative frequency bar chart in Display 2.34 shows the proportion of the
female labor force age 25 and older in the United States who fall into various
educational categories. The coding used in the display is as follows:
1: none8th grade
2: 9th grade11th grade
3: high school graduate
4: some college, no degree
5: associate degree
6: bachelors degree
7: masters degree
8: professional degree
9: doctorate degree
Display 2.34 The female labor force age 25 years and older by
educational attainment [Source: U.S. Census Bureau, March
2005 Current Population Survey, www.census.gov.]
The educational categories in Display 2.34 have a natural order from least
education to most education and are coded with the numbers 1 through 9. Note
that if you compute the mean of this distribution, there is no reasonable way to
interpret it. However, it does make sense to summarize this distribution using the
mode: More women fall into the category high school graduate than into any
other category. Thus, the numbers 1 through 9 are best thought of as representing
an ordered categorical variable, not a quantitative variable.
You will learn more about the analysis of categorical data in Chapter 10.
Bar Charts
D11.In the bar chart in Display 2.33, would it matter if the order of the bars were
reversed? In the bar chart in Display 2.34, would it matter if the order of
the first two bars were reversed? Comment on how we might define two
different types of categorical variables.
Summary 2.2: Graphical Displays of Distributions

When a variable is quantitative, you can use dot plots, stemplots (stem-and-leaf
plots), and histograms to display the distribution of values. From each plot, you
can see the shape, center, and spread of the distribution. However, the amount of
2.2
51
detail varies, and you should choose a plot that fits both your data set and your
reason for analyzing it.
Stemplots can retain the actual data values.
Dot plots are best used with a small number of values and show roughly
where the values lie on a number line.
Histograms show only intervals of values, losing the actual data values, and
are most appropriate for large data sets.
A bar chart shows the distribution of a categorical variable.
When you look at a plot, you should attempt to answer these questions:
Where did this data set come from?
What are the cases and the variables?
What are the shape, center, and spread of this distribution? Does the
distribution have any unusual characteristics, such as clusters, gaps, or
outliers?
What are possible interpretations or explanations of the patterns you see in
the distribution?
Practice
More About Dot Plots
P6. In the listing of the Westvaco data in
Display 1.1 on page 5, which variables are
quantitative? Which are categorical?
P7. Select a reasonable scale, and make a dot
plot of the gestation periods of the mammals
listed in Display 2.24 on page 43. Write a
sentence using shape, center, and spread
to summarize the distribution of gestation
periods for the mammals. What kinds of
mammals have longer gestation periods?
Histograms
P8. Make histograms of the average longevities
and the maximum longevities from Display
2.24. Describe how the distributions differ in
terms of shape, center, and spread. Why do
these differences occur?
P9. Convert your histograms from P8 of
the average longevities and maximum
longevities of the mammals to relative
frequency histograms. Do the shapes of the
histograms change?
P10. Using the relative frequency histogram of
life expectancy in countries around the
world (Display 2.28 on page 47), estimate
the proportion of countries with a life
expectancy of less than 50 years. Then
estimate the number of countries with a life
expectancy of less than 50 years. Describe
the shape, center, and spread of this
distribution.
Stemplots
P11. Make a back-to-back stemplot of the average
longevities and maximum longevities from
Display 2.24 on page 43. Compare the two
distributions.
52
Bar Charts for Categorical Data

P12. Display 2.35, educational attainment of
the male labor force, is the counterpart of
Display 2.34. What are the cases, and what
is the variable? Describe the distribution
you see here. How does the distribution of
female education compare to the distribution
of male education? Why is it better to look
at relative frequency bar charts rather
than frequency bar charts to make this
comparison?
Display 2.35 The male labor force age 25 years
and older by educational attainment.
[Source: U.S. Census Bureau, March 2005 Current
Population Survey, www.census.gov.]
P13. Using the Westvaco data in Display 1.1

on page 5, make a bar chart showing the
number of workers laid off in each round. In
addition to a bar showing layoffs, for each of
the five rounds, include a bar showing the
number of workers not laid off . Then make
a relative frequency bar chart. Describe any
patterns you see.
Exercises
E15. The dot plot in Display 2.36 shows the
distribution of the ages of pennies
in a sample collected by a statistics
class.
a. Where did this data set come from? What

are the cases and the variables?
b. What are the shape, center, and spread of
this distribution?
c. Does the distribution have any unusual
characteristics? What are possible
interpretations or explanations of the
patterns you see in the distribution? That
is, why does the distribution have the
shape it does?
E16. Suppose you collect this information for
each student in your class: age, hair color,
number of siblings, gender, and miles he
or she lives from school. What are the
cases? What are the variables? Classify
each variable as quantitative or
categorical.
Display 2.36 Age of pennies. Each dot represents

four points.
2.2
53
E17. Using your knowledge of the variables and

what you think the shape of the distribution
might be, match each variable in this list
with the appropriate histogram in Display
2.37.
E20. Convert the histogram in Display 2.38 into a

relative frequency histogram.
I. scores on a fairly easy examination in

statistics
II. heights of a group of mothers and their
12-year-old daughters
III. numbers of medals won by medal
winning countries in the 2004 Summer
Olympics
IV. weights of grown hens in a barnyard
.
Display 2.38 Ages of 1000 people.
E21. Display 2.39 shows the distribution of the

heights of U.S. males between the ages of
18 and 24. The heights are rounded to the
nearest inch.
Display 2.37 Four histograms with different shapes.
E18. Using the technology available to you, make

histograms of the average longevity and
maximum longevity data in Display 2.24
on page 43, using bar widths of 4, 8, and
16 years. Comment on the main features
of the shapes of these distributions and
determine which bar width appears to
display these features best.
E19. Rewrite each sentence so that it states a
relative frequency rather than a count.
a. Six students in a class of 30 got an A.
b. Out of the 50,732 people at a concert,
24,021 bought a T-shirt.
54
Display 2.39 Heights of males, age 18 to 24. [ Source:

U.S. Census Bureau, Statistical Abstract of the
United States, 1991.]
a. Draw a smooth curve to approximate the

histogram.
b. Without doing any computing, estimate
the mean and standard deviation.
c. Estimate the proportion of men age 18 to
24 who are 74 in. tall or less.
d. Estimate the proportion of heights that
fall below 68 in.
e. Why should you say that the distribution
of heights is approximately normal
rather than simply saying that it is
normally distributed?
E22. The histogram in Display 2.40 shows the

distribution of SAT I math scores for
20042005.
a. Without doing any computing, estimate
the mean and standard deviation.
b. Roughly what percentage of the SAT I
math scores would you estimate are
within one standard deviation of the
mean?
c. For SAT I critical reading scores, the
shape was similar, but the mean was 10
points lower and the standard deviation
was 2 points smaller. Draw a smooth
curve to show the distribution of SAT I
critical reading scores.
E24. The plots in Display 2.41 show a form of

back-to-back histogram called a population
pyramid. Describe how the population
distribution of the United States differs from
the population distribution of Mexico.
Display 2.40 Relative frequency histogram of SAT I

math scores, 20042005. [Source: College
Board Online, www.collegeboard.org.]
E23. In this section, you looked at various

characteristics of mammals.
a. Would you predict that wild mammals or
domesticated mammals generally have
greater longevity?
b. Using the data in Display 2.24 on page 43,
make a back-to-back stemplot to compare
the average longevities.
c. Write a short summary comparing the
two distributions.
Display 2.41 Population pyramids for the United

States and Mexico, 2005. [Source: U.S.
Census Bureau, International Data Base,
www.census.gov.]
55
Lesson
1 de 18
http://acr.keymath.com/KeyPressPortalV3.0/Viewer/Lesson.htm
E25. Examine the grouped bar chart in Display

2.42, which summarizes some of the
information from Display 2.24 on page 43.
Display 2.42 Bar chart for nonpredators and

predators, showing frequency of wild
and domesticated mammals.
a. Describe what the height of the first three

bars represents.
b. How can you tell from this bar chart
whether a predator from the list in
Display 2.24 is more likely to be wild or
domesticated?
2.3
c. How can you tell from this bar chart

whether a nonpredator or a predator is
more likely to be wild?
E26. Make a grouped bar chart similar to that in
E25 for the hourly and salaried Westvaco
workers (see Display 1.1 on page 5), with
bars showing the frequencies of laid off and
not laid off for the two categories of workers.
E27. Consider the mammals speeds in Display
2.24.
a. Count the number of mammals that have
speeds ending in 0 or 5.
b. How many speeds would you expect to
end in 0 or 5 just by chance?
c. What are some possible explanations for
the fact that your answers in parts a and b
are so different?
E28. Look through newspapers and magazines
to find an example of a graph that is either
misleading or difficult to interpret. Redraw
the graph to make it clear.
Measures of Center and Spread

So far you have relied on visual methods for estimating summary statistics to
measure center and spread. In this section, you will learn how to compute exact
values of those same summary statistics directly from the data.
Measures of Center
The two most commonly used measures of center are the mean and the median.
The mean, , is the same number that many people call the average. To
compute the mean, sum all the values of x and divide by the number of
values, n:
(The symbol , for sum, means to add up all the values of x.)
56
10-03-2009 13:57
Lesson
2 de 18
The mean is the

balance point.
The mean is the balance point of a distribution. To estimate the mean visually
on a dot plot or histogram, find where you would have to place a finger below the
horizontal axis in order to balance the distribution, as if it were a tray of blocks
(see Display 2.43).
Display 2.43 The mean is the balance point of a distribution.

The mean is the
balance point.
The median is the value that divides the data into halves, as shown in
Display 2.44. To find it for an odd number of values, list all the values in order
and select the middle one. If there are n values and n is odd, you will find the
. If n is even, the median is the average of the values
median at position
on either side of position
.
Display 2.44 The median divides the distribution into two

equal areas.
Example: Effect of Round 2 Layoffs on Measures of Center

The ages of the hourly workers involved in Round 2 of the layoffs at Westvaco
were 25, 33, 35, 38, 48, 55, 55*, 55*, 56, and 64* (* indicates laid off in Round 2).
The two dot plots in Display 2.45 show the distributions of hourly workers before
and after the second round of layoffs. What was the effect of Round 2 on the mean
age? On the median age?
2.3
57
10-03-2009 13:57
Lesson
3 de 18
Display 2.45 Ages of Westvaco hourly workers before and after

Round 2, showing the means and medians.
Solution
Means
Before: The sum of the ten ages is 464, so the mean age is , or 46.4 years.
After: There are seven ages and their sum is 290, so the mean age is
, or
41.4 years.
The layoffs reduced the mean age by 5 years.
Medians
Before: Because there are ten ages, n = 10, so
=
or 5.5, and the
median is halfway between the fifth ordered value, 48, and the sixth
ordered value, 55. The median is
, or 51.5 years.
After: There are seven ages, so
=
or 4. The median is the fourth
ordered value, or 38 years.
The layoffs reduced the median age by 13.5 years.
Measures of Center
D12.Find the mean and median of each ordered list, and contrast their behavior.
a. 1, 2, 3
b. 1, 2, 6
c. 1, 2, 9
d. 1, 2, 297
D13.As you saw in D12, typically the mean is affected more than the median by
an outlier.
a. Use the fact that the median is the halfway point and the mean is the
balance point to explain why this is true.
b. For the distributions of mammal speeds in Display 2.31 on page 48,
the means are 43.5 mi/h for predators and 31.5 mi/h for nonpredators.
The medians are 40.5 mi/h and 33.5 mi/h, respectively. What about the
distributions causes the means to be farther apart than the medians?
c. What about the shapes of the plots in Display 2.45 explains why the
means change so much less than the medians?
58
10-03-2009 13:57
Lesson
4 de 18
Measuring Spread Around the Median:

Quartiles and IQR
Pair a measure of center
with a measure of
spread.
Use IQR as a measure of

spread with the median.
You can locate the median of a distribution by dividing your data into a lower
and upper half. You can use the same idea to measure spread: Find the values that
divide each half in half again. These two values, the lower quartile, Q1, and the
upper quartile, Q3, together with the median, divide your data into four quarters.
The distance between the upper and lower quartiles, called the interquartile
range, or IQR, is a measure of spread.
IQR = Q3 Q1
San Francisco, California, and Springfield, Missouri, have about the same
median temperature over the year. In San Francisco, half the months of the year
have a normal temperature above 56.5F, half below. In Springfield, half the
months have a normal temperature above 57F, half below. If you judge by these
medians, the difference hardly matters. But if you visit San Francisco, you better
take a jacket, no matter what month you go. If you visit Springfield, take your
shorts and a T-shirt in the summer and a heavy coat in the winter. The difference
in temperatures between the two cities is not in their centers but in their variability.
In San Francisco, the middle 50% of normal monthly temperatures lie in a narrow
9 interval between 52.5F and 61.5F, whereas in Springfield the middle 50% of
normal monthly temperatures range over a 31 interval, varying from 40.5F to
71.5F. In other words, the IQR is 9F for San Francisco and 31F for Springfield.
Finding the Quartiles and IQR
If you have an even number of cases, finding the quartiles is straightforward:
Order your observations, divide them into a lower and upper half, and then divide
each half in half. If you have an odd number of cases, the idea is the same, but
theres a question of what to do with the middle value when you form the upper
and lower halves.
There is no one standard answer. Different statistical software packages use
different procedures that can give slightly different values for the quartiles. In this
book, the procedure is to omit the middle value when you form the two halves.
Example: Finding the Quartiles and IQR for Workers Ages

Find the quartiles and IQR for the ages of the hourly workers at Westvaco before
and after Round 2 of the layoffs.
Solution
Before: There are ten ages: 25, 33, 35, 38, 48, 55, 55, 55, 56, 64. Because n is even,
the median is halfway between the two middle values, 48 and 55, so it is 51.5. The
lower half of the data is made up of the first five ordered values, and the median
of the lower half is the third value, so Q1 is 35. The upper half of the data is the
set of the five largest values, and the median of these is again the third value,
so Q3 is 55. The IQR is 55 35, or 20.
59
10-03-2009 13:57
Lesson
5 de 18
After: After the three workers are laid off in Round 2, there are seven ages: 25, 33,
35, 38, 48, 55, 56. Because n is odd, the median is the middle value, 38. Omit this
one number. The lower half of the data is made up of the three ordered values
to the left of position 4. The median of these is the second value, so Q1 is 33.
The upper half of the data is the set of the three ordered values to the right of
position 4, and the median of these is again the second value, so Q3 is 55. The
IQR is 55 33, or 22.
Finding the Quartiles and IQR

D14.Here are the medians and quartiles for the speeds of the domesticated and
wild mammals:
a. Use the information in Display 2.24 on page 43 to verify these values,

and then use them to summarize and compare the two distributions.
b. Why might the speeds of domesticated mammals be less spread out than
the speeds of wild mammals?
D15.The following quote is from the mystery The List of Adrian Messenger, by
Philip MacDonald (Garden City, NY: Doubleday, 1959, p. 188). Detective
Firth asks Detective Seymour if eyewitness accounts have provided a
description of the murderer:
Descriptions? he said. You mustve collected quite a few. How did they
boil down?
To a no-good norm, sir. Seymour shrugged wearily. They varied so much,
the average was useless.
Explain what Detective Seymour means.
Five-Number Summaries, Outliers, and Boxplots

The visual, verbal, and numerical summaries youve seen so far tell you about
the middle of a distribution but not about the extremes. If you include the
minimum and maximum values along with the median and quartiles, you get
the five-number summary.
60
10-03-2009 13:57
Lesson
6 de 18
The five-number summary for a set of values:

Minimum: the smallest value
Lower or first quartile, Q1: the median of the lower half of the ordered set
of values
Median or second quartile: the value that divides the ordered set of values
into halves
Upper or third quartile, Q3: the median of the upper half of the ordered
set of values
Maximum: the largest value
The difference of the maximum and the minimum is called the range.
Display 2.46 shows the five-number summary for the speeds of the mammals
listed in Display 2.24.
Display 2.46 Five-number summary for the mammal speeds.
Display 2.47 shows a boxplot of the mammal speeds. A boxplot (or

box-and-whiskers plot) is a graphical display of the five-number summary.
The box extends from Q1 to Q3, with a line at the median. The whiskers run
from the quartiles to the extreme values.
Display 2.47 Boxplot of mammal speeds.
The maximum speed of 70 mi/h for the cheetah is 20 mi/h from the next
fastest mammal (the lion) and 28 mi/h from the nearest quartile. It is handy to
have a version of the boxplot that shows isolated casesoutlierssuch as the
cheetah. Informally, outliers are any values that stand apart from the rest. This
rule often is used to identify outliers.
A value is an outlier if it is more than 1.5 times the IQR from the nearest
quartile.
1.5 IQR rule for outliers
Note that more than 1.5 times the IQR from the nearest quartile is another
way of saying either greater than Q3 plus 1.5 times IQR or less than Q1 minus 1.5
times IQR.
2.3
61
10-03-2009 13:57
Lesson
7 de 18
Example: Outliers in the Mammal Speeds

Use the 1.5 IQR rule to identify outliers and the largest and smallest
non-outliers among the mammal speeds.
Solution
From Display 2.46, Q1 = 30 and Q3 = 42, so the IQR is 42 30 or 12, and
1.5 IQR equals 18.
At the low end:
Q1 1.5 IQR = 30 18 = 12
The pig, at 11 mi/h, is an outlier.
The squirrel, at 12 mi/h, is the smallest non-outlier.
At the high end:
Q3 +1.5 IQR = 42 + 18 = 60
The cheetah, at 70 mi/h, is an outlier.
The lion, at 50 mi/h, is the largest non-outlier.
A modified boxplot, shown in Display 2.48, is like the basic boxplot except
that the whiskers extend only as far as the largest and smallest non-outliers
(sometimes called adjacent values) and any outliers appear as individual dots or
other symbols.
Display 2.48 Modified boxplot of mammal speeds.
Boxplots are particularly useful for comparing several distributions.
Example: Boxplots That Show Outliers

Display 2.49 shows side-by-side modified boxplots of average longevity for wild
and domesticated mammals. Compare the two distributions.
Display 2.49 Comparison of average longevity.
62
10-03-2009 13:57
Lesson
8 de 18
Solution
The boxplot for domesticated animals has no median line. So many domesticated
animals had an average longevity of 12 years that it is both the median and the
upper quartile. These plots show that species of domestic mammals typically have
median average longevities of about 12 years, with about the middle half of these
average longevities falling between 8 and 12 years. The average longevity of wild
mammals centers at about the same place, but the wild mammal average longevities
have more variability, with the middle half between about 7.5 and 15.5 years.
Both shapes are roughly symmetric except for some unusually large average
longevitiestwo wild mammals have average longevities of more than 30 years.
[See Calculator Note 2D to learn how to display regular and modified boxplots and
five-number summaries on your calculator.]
When are boxplots most

useful?
Boxplots are useful when you are plotting a single quantitative variable and
you want to compare the shapes, centers, and spreads of two or more
distributions
you dont need to see individual values, even approximately
you dont need to see more than the five-number summary but would like
outliers to be clearly indicated
Five-Number Summaries, Outliers, and Boxplots

D16. Test your ability to interpret boxplots by answering these questions.
a. Approximately what percentage of the values in a data set lie within the
box? Within the lower whisker, if there are no outliers? Within the upper
whisker, if there are no outliers?
b. How would a boxplot look for a data set that is skewed right? Skewed
left? Symmetric?
c. How can you estimate the IQR directly from a boxplot? How can you
estimate the range?
d. Is it possible for a boxplot to be missing a whisker? If so, give an example.
If not, explain why not.
e. Contrast the information you can learn from a boxplot with what you
can learn from a histogram. List the advantages and disadvantages of
each type of plot.
2.3
63
10-03-2009 13:57
Lesson
9 de 18
Measuring Spread Around the Mean:

The Standard Deviation
There are various ways you can measure the spread of a distribution around its
mean. Activity 2.3a gives you a chance to create a measure of your own.
Comparing Hand Spans: How Far Are You from the Mean?
What youll need: a ruler
1. Spread your hand on a ruler and
measure your hand span (the distance
from the tip of your thumb to the tip of
your little finger when you spread your
fingers) to the nearest half centimeter.
2. Find the mean hand span for your
group.
3. Make a dot plot of the results for your
group. Write names or initials above
below the
the dots to identify the cases. Mark the mean with a wedge
number line.
4. Give two sources of variability in the measurements. That is, give two
reasons why all the measurements arent the same.
5. How far is your hand span from the mean hand span of your group? How
far from the mean are the hand spans of the others in your group?
6. Make a plot of differences from the mean. Again label the dots with names
or initials. What is the mean of these differences? Tell how to get the
second plot from the first without computing any diff erences.
7. Using the idea of differences from the mean, invent at least two measures
that give a typical distance from the mean.
8. Compare your measures with those of the other groups in your class.
Discuss the advantages and disadvantages of each groups method.
The differences from the mean, x , are called deviations. The mean is
the balance point of the distribution, so the set of deviations from the mean will
always sum to zero.
Deviations from the mean sum to zero:
Example: Deviations from the Mean

Find the deviations from the mean for the predators speeds and verify that the
sum of these deviations is 0. Which predators speed is farthest from the mean?
64
10-03-2009 13:57
Lesson
10 de 18
Solution
The speeds 30, 30, 39, 42, 50, and 70 mi/h have mean 43.5 mi/h. The deviations
from the mean are
30 43.5 = 13.5
30 43.5 = 13.5
39 43.5 = 4.5
42 43.5 = 1.5
50 43.5 = 6.5
70 43.5 = 26.5
The sum of the deviations is 13.5 + (13.5) + (4.5) + (1.5) + 6.5 + 26.5,
which equals 0. The cheetahs speed, 70 mi/h, is farthest from the mean.
How can you use the deviations from the mean to get a measure of spread?
You cant simply find the average of the deviations, because you will get 0 every
time. As you might have suggested in the activity, you could find the average of
the absolute values of the deviations. That gives a perfectly reasonable measure
of spread, but it does not turn out to be very easy to use or very useful. Think of
how hard it is to deal with an equation that has sums of absolute values in it, for
example, y = | x 1 | + | x 2 | + | x 3 | . On the other hand, if you square the
deviations, which also gets rid of the negative signs, you get a sum of squares.
Such a sum is always quadratic no matter how many terms there are, for example,
y = (x 1 )2 + (x 2 )2 + (x 3 )2 = 3x2 12x + 14.
The measure of spread that incorporates the square of the deviations is the
standard deviation, abbreviated SD or s, that you met in Section 2.1. Because sums
of squares really are easy to work with mathematically, the SD offers important
advantages that other measures of spread dont give you. You will learn more
about these advantages in Chapter 7. The formula for the standard deviation, s, is
given in the box.
Formula for the Standard Deviation, s
The square of the standard deviation, s2 , is called the variance.
Calculators might label

and
the two versions
, or
and s.
It might seem more natural to divide by n to get the average of the squared
deviations. In fact, two versions of the standard deviation formula are used: One
divides by the sample size, n; the other divides by n 1. Dividing by n 1 gives
a slightly larger value. This is useful because otherwise the standard deviation
computed from a sample would tend to be smaller than the standard deviation of
the population from which the sample came. (You will learn more about this in
Chapter 7.) In practice, dividing by n 1 is almost always used for real data even
if they arent a sample from a larger population.
65
10-03-2009 13:57
Lesson
11 de 18
Example: The Standard Deviation of Mammal Longevity

Compute the standard deviation of the average longevity of domesticated
mammals from Display 2.24 on page 43.
Solution
The table in Display 2.50 is a good way to organize the steps. First find the mean
longevity, , and then subtract it from each observed value x to get the deviations,
x . Square each deviation to get (x )2.
Display 2.50 Steps in computing the standard deviation.
To compute the standard deviation, sum the squared deviations, divide the
sum by n 1, and finally take the square root:
[You can organize the steps of calculating the standard deviation on your calculator.
See Calculator Note 2E.]
Your calculator will compute the summary statistics for a set of data. [See
Calculator Note 2F.] Here are the summary statistics for the domesticated mammal
longevity data. Note that the standard deviation calculated in the previous
example is denoted as Sx. Note also that the five-number summary is shown.
66
10-03-2009 13:57
Lesson
12 de 18

D17. Refer to the previous example for mammal longevities.
a. Does 4.67 years seem like a typical distance from the mean of 11 years
for the average longevities in the example?
b. The average longevities are measured in years. What is the unit of
measurement for the mean? For the standard deviation? For the
variance? For the interquartile range? For the median?
D18. The standard deviation, if you look at it the right way, is a generalization
of the usual formula for the distance between two points. How does the
formula for the standard deviation remind you of the formula for the
distance between two points?
D19. What effect does dividing by n 1 rather than by n have on the standard
deviation? Does which one you divide by matter more with a large number
of values or with a small number of values?
Summaries from a Frequency Table

To find the mean of the numbers 5, 5, 5, 5, 5, 5, 8, 8, and 8, you could sum them
and divide their sum by how many numbers there are. However, you could get
the same answer faster by taking advantage of the repetitions:
You can use formulas to find the mean and standard deviation of values
in a frequency table, like the one in Display 2.51 in the example on the
next page.
Formulas for the Mean and Standard Deviation of Values

in a Frequency Table
If each value x occurs with frequency f, the mean of a frequency table is
given by
The standard deviation is given by
where n is the sum of the frequencies, or
67
10-03-2009 13:57
Lesson
13 de 18
Example: Mean and Standard Deviation of Coin Values

Suppose you have five pennies, three nickels, and two dimes. Find the mean value
of the coins and the standard deviation.
Solution
The table in Display 2.51 shows a way to organize the steps in computing the
mean using the formula for the mean of values in a frequency table.
Display 2.51 Steps in computing the mean of a frequency table.
Display 2.52 gives an extended version of the table, designed to organize the
steps in computing both the mean and the standard deviation.
Display 2.52 Steps in computing the standard deviation of

values in a frequency table.
[See Calculator Note 2F to learn how to compute numerical summaries from a

frequency table.]

D20.Explain why the formula for the standard deviation in the box on page 67
gives the same result as the formula on page 65.
10-03-2009 13:57
Lesson
14 de 18
Summary 2.3: Measures of Center and Spread

Your first step in any data analysis should always be to look at a plot of your data,
because the shape of the distribution will help you determine what summary
measures to use for center and spread.
To describe the center of a distribution, the two most common summaries are
the median and the mean. The median, or halfway point, of a set of ordered
values is either the middle value (if n is odd) or halfway between the two
middle values (if n is even). The mean, or balance point, is the sum of the
values divided by the number of values.
To measure spread around the median, use the interquartile range, or IQR,
which is the width of the middle 50% of the data values and equals the
distance from the lower quartile to the upper quartile. The quartiles are the
medians of the lower half and upper half of the ordered list of values.
To measure spread around the mean, use the standard deviation. To compute
the standard deviation for a data set of size n, first find the deviations from
the mean, then square them, sum the squared deviations, divide by n 1, and
take the square root.
A boxplot is a useful way to compare the general shape, center, and spread
of two or more distributions with a large number of values. A modified boxplot
also shows outliers. An outlier is any value more than 1.5 times the IQR from the
nearest quartile.
Practice
Measures of Center
P14.Find the mean and median of these ordered
lists.
a.1 2 3 4
b. 1 2 3 4 5
c.1 2 3 4 5 6
d. 1 2 3 4 5 . . . 97 98
e.1 2 3 4 5 . . . 97 98 99
P15.Five 3rd graders, all about 4 ft tall, are
standing together when their teacher, who
is 6 ft tall, joins the group. What is the new
mean height? The new median height?
P16.The stemplots in Display 2.53 show the life
expectancies (in years) for females in the
countries of Africa and Europe. The means
are 53.6 years for Africa and 79.3 years for
Europe.
a. Find the median life expectancy for each
set of countries.
b. Is the mean or the median smaller for
each distribution? Why is this so?
Display 2.53 Female life expectancies in Africa and
Europe. [Source: Population Reference Bureau,
World Population Data Sheet, 2005.]
69
10-03-2009 13:57
Lesson
15 de 18
Measuring Spread Around the Median:

Quartiles and IQR
P17. Find the quartiles and IQR for these ordered
lists.
a. 1 2 3 4 5 6
b. 1 2 3 4 5 6 7
c. 1 2 3 4 5 6 7 8
d. 1 2 3 4 5 6 7 8 9
P18. Display 2.54 shows a back-to-back stemplot
of the average longevity of predators and
nonpredators.
b. Estimate the median number of viewers,

and use this median in a sentence.
P20. Use the medians and quartiles from D14 on
page 60 and the data in Display 2.24 on page
43 to construct side-by-side boxplots of the
speeds of wild and domesticated mammals.
(Dont show outliers in these plots.)
P21. The stemplot of average mammal longevities
is shown in Display 2.56.
Display 2.56 Average longevity (in years) of

38 mammals.
Display 2.54 Average longevities of predators and
nonpredators.
a. By counting on the plot, find the median

and quartiles for each group of mammals.
b. Write a pair of sentences summarizing
and comparing the shape, center, and
spread of the two distributions.
Five-Number Summaries, Outliers,
and Boxplots
P19. The boxplot in Display 2.55 shows the
number of viewers who watched the 101
prime-time network television shows in the
week that Seinfeld aired its last new episode.
Display 2.55 Modified boxplot of number of viewers

of prime-time television shows.
a. Seinfeld had more viewers than any other

show. About how many viewers did it
have?
70
a. Use the stemplot to find the five-number

summary.
b. Find the IQR.
c. Compute Q1 1.5 IQR. Identify any
outliers (give the animal name and
longevity) at the low end.
d. Identify an outlier at the high end and the
largest non-outlier.
e. Draw a modified boxplot.
P22. Verify that the sum of the deviations from
the mean is 0 for the numbers 1, 2, 4, 6, and
9. Find the standard deviation.
P23. Without computing, match each list of
numbers in the left column with its standard
deviation in the right column. Check any
answers you arent sure of by computing.
a. 1 1 1 1
i. 0
b. 1 2 2
ii. 0.058
c. 1 2 3 4 5
iii. 0.577
d. 10 20 20
iv. 1.581
e. 0.1 0.2 0.2
v. 3.162
f. 0 2 4 6 8
vi. 3.606
g. 0 0 0 0 5 6 6 8 8
vii. 5.774
10-03-2009 13:57
Lesson
16 de 18

P24. Display 2.57 shows the data on family size
for two representative sets of 100 families,
one set from 1967 and the other from 1997.
Display 2.57 Number of children in a sample of

families, 1967 and 1997. [Source: U.S.
Census Bureau, www.census.gov.]
b. Find the median number of children per

family for 1967.
c. Use the formulas to compute the mean
and standard deviation of the 1967
distribution.
P25. Refer to Display 2.57.
a. Use the formulas for the mean and
standard deviation of values in a frequency
table to compute the mean number of
children per family and the standard
deviation for the 1997 distribution.
b. Find the median number of children
for 1997.
c. What are the positions of the quartiles in
an ordered list of 100 numbers? Find the
quartiles for the 1967 distribution and
compute the IQR. Do the same for the
1997 distribution.
d. Write a comparison of the shape, center,
and spread of the distributions for the
two years.
a. Try to visualize the shapes of the two

distributions. Are they symmetric,
skewed left , or skewed right?
Exercises
E29. The mean of a set of seven values is 25. Six
of the values are 24, 47, 34, 10, 22, and 28.
What is the 7th value?
E30. The sum of a set of values is 84, and the
mean is 6. How many values are there?
E31. Three histograms and three boxplots appear
in Display 2.58. Which boxplot displays the
same information as
a. histogram A?
b. histogram B?
c. histogram C?
Display 2.58
Match the histograms with their

boxplots.
71
10-03-2009 13:57
Lesson
17 de 18
E32.The test scores of 40 students in a first-period

class were used to construct the first
boxplot in Display 2.59, and the test scores
of 40 students in a second-period class were
used to construct the second. Can the third
plot be a boxplot of the combined scores of
the 80 students in the two classes? Why or
why not?
E36. The boxplots below show the average

longevity of mammals, from Display 2.24.
a. Using only the basic boxplot in Display
2.60, show that there must be at least one
outlier in the set of average longevity.
Display 2.60 Boxplot of average longevity of

mammals.
b. How many outliers are there in the

modified boxplot of average longevity in
Display 2.61?
Display 2.59
Boxplots of three sets of test scores.
E33.Make side-by-side boxplots of the speeds of

predators and nonpredators. (The stemplot
in Display 2.31 on page 48 shows the values
ordered.) Are the boxplots or the back-to-back
stemplot in Display 2.31 better for
comparing these speeds? Explain.
E34.The U.S. Supreme Court instituted a
temporary ban on capital punishment
between 1967 and 1976. Between 1977 and
2000, 31 states carried out 683 executions.
(The other 19 states either did not have
a death penalty or executed no one.) The
five states that executed the most prisoners
were Texas (239), Virginia (81), Florida
(50), Missouri (46), and Oklahoma (30).
The remaining 26 states carried out these
numbers of executions: 26, 25, 23, 23, 23, 22,
16, 12, 11, 8, 8, 7, 6, 4, 3, 3, 3, 3, 2, 2, 2, 1, 1,
1, 1, 1. For all 50 states, what was the mean
number of executions per state? The median
number? What were the quartiles? Draw a
boxplot, showing any outliers, of the number
of executions for all 50 states. [Source: U.S.
Department of Justice, Bulletin: Capital Punishment
2000.]
E35.Make a back-to-back stemplot comparing

the ages of those retained and those laid
off among the salaried workers in the
engineering department at Westvaco (see
Display 1.1 on page 5). Find the medians
and quartiles, and use them to write a verbal
comparison of the two distributions.
72
Display 2.61 Modified boxplot of average longevity

of mammals, showing outliers.
c. How many outliers are shown in Display
2.49 on page 62? How can that be,

considering the boxplot in Display 2.61?
E37. No computing should be necessary to
answer these questions.
a. The mean of each of these sets of values is
20, and the range is 40. Which set has the
largest standard deviation? Which has the
smallest?
I. 0 10 20 30 40
II. 0 0 20 40 40
III. 0 19 20 21 40
b. Two of these sets of values have a
standard deviation of about 5. Which
two?
I. 5 5 5 5 5 5
II. 10 10 10 20 20 20
III. 6 8 10 12 14 16 18 20 22
IV. 5 10 15 20 25 30 35 40 45
E38. The standard deviation of the first set
of values listed here is about 32. What
is the standard deviation of the second
10-03-2009 13:57
Lesson
18 de 18
set? Explain. (No computing should be

necessary.)
16 23 34 56 78 92 93
20 27 38 60 82 96 97
E39.
Consider the set of the heights of all female

National Collegiate Athletic Association
(NCAA) athletes and the set of the heights
of all female NCAA basketball players.
Which distribution will have the larger
mean? Which will have the larger standard
deviation? Explain.
E40. Consider the data set 15, 8, 25, 32, 14, 8, 25,
and 2. You can replace any one value with a
number from 1 to 10. How would you make
this replacement
is computed by removing the largest 5% of

values and the smallest 5% of values from
the data set and then computing the mean
of the remaining middle 90% of values. (The
percentage that is cut off at each end can
vary depending on the software.)
a. Find the trimmed mean of the maximum
longevities in Display 2.24 on page 43.
b. Is the trimmed mean affected much by
outliers?
E43.This table shows the weights of the pennies
in Display 2.3 on page 31. For example, the
four pennies in the second interval, 3.0000 g
to 3.0199 g, are grouped at the midpoint of
this interval, 3.01.
a. to make the standard deviation as large as

possible?
b. to make the standard deviation as small
as possible?
c. to create an outlier, if possible?
E41. Another measure of center that sometimes is
used is the midrange. To find the midrange,
compute the mean of the largest value and
the smallest value.
The statistics in this computer output
summarize the number of viewers of
prime-time television shows (in millions)
for the week of the last new Seinfeld episode.
a. Using these summary statistics alone,

compute the midrange both with and
without the value representing the
Seinfeld episode. (Seinfeld had the largest
number of viewers and Seinfeld Clips,
with 58.53 million viewers, the second
largest.) Is the midrange affected much
by outliers? Explain.
b. Compute the mean of the ratings
without the Seinfeld episode, using only
the summary statistics in the computer
output.
E42. In computer output like that in E41,
TrMean is the trimmed mean. It typically
a. Find the mean weight of the pennies.

b. Find the standard deviation.
c. Does the standard deviation appear to
represent a typical deviation from the
mean?
E44.Suppose you have five pennies, six nickels,
four dimes, and five quarters.
a. Sketch a dot plot of the values of the
20 coins, and use it to estimate the mean.
b. Compute the mean using the formula for
the mean of values in a frequency table.
c. Estimate the SD from your plot: Is it
closest to 0, 5, 10, 15, or 20?
d. Compute the standard deviation using
the formula for the standard deviation of
values in a frequency table.
73
10-03-2009 13:57
Lesson
1 de 9
E45.On the first test of the semester, the scores

of the first-period class of 30 students had a
mean of 75 and a median of 70. The scores of
the second-period class of 22 students had a
mean of 70 and a median of 68.
a. To the nearest tenth, what is the mean
test score of all 52 students? If you cannot
calculate the mean of the two classes
combined, explain why.
b. What is the median test score of all 52
students? If you cannot find the median
of the two classes combined, explain why.
E46.The National Council on Public Polls
rebuked the press for its coverage of a Gallup
poll of Islamic countries. According to the
Council:
News stories based on the Gallup poll
reported results in the aggregate without
regard to the population of the countries
they represent. Kuwait, with less than
2.4
2 million Muslims, was treated the same

as Indonesia, which has over 200 million
Muslims. The aggregate quoted in the
media was actually the average for the
countries surveyed regardless of the size
of their populations.
The percentage of people in Kuwait who
thought the September 11 terrorist attacks
were morally justified was 36%, while the
percentage in Indonesia was 4%. [Source:
www.ncpp.org.]
a. Suppose that the poll covered only

these two countries and that the people
surveyed were representative of the
entire country. What percentage of all the
people in these two countries thought
that the terrorist attacks were morally
justified?
b. What percentage would have been
reported by the press?
Working with Summary Statistics

Summary statistics are very useful, but only when they are used with good
judgment. This section will teach you how to tell which summary statistic to
use, how changing units of measurement and the presence of outliers affect your
summary statistics, and how to interpret percentiles.
Which Summary Statistic?
Plot first, then look for

summaries.
74
Which summary statistics should you use to describe a distribution? Should

you use mean and standard deviation? Median and quartiles? Something else?
The right choice can depend on the shape of your distribution, so you should
always start with a plot. For normal distributions, the mean and standard
deviation are nearly always the most suitable. For skewed distributions, the
median and quartiles are often the most useful, in part because they have a simple
interpretation based on dividing a data set into fourths.
Sometimes, however, the mean and standard deviation will be the right choice
even if you have a skewed distribution. For example, if you have a representative
sample of house prices for a town and you want to use your sample to estimate the
total value of all the towns houses, the mean is what you want, not the median.
In Chapter 7, youll see why the mean and standard deviation are the most useful
choices when doing statistical inference.
10-03-2009 13:59
Lesson
2 de 9
Choosing the right summary statistics is something you will get better at
as you build your intuition about the properties of these statistics and how they
behave in various situations.
D21. Explain how to determine the total amount of property taxes if you know
the number of houses, the mean value, and the tax rate. In what sense is
knowing the mean equivalent to knowing the total?
D22. When a measure of center for the income of a communitys residents is
given, that number is usually the median. Why do you think that is the case?
The Effects of Changing Units

This discussion illustrates some important properties of summary statistics. It will
also help you develop your intuition about how the geometry and the arithmetic
of working with data are related.
The lowest temperature on record for Washington, D.C., is 15F. How does
that temperature compare with the lowest recorded temperatures for capitals of
other countries? Display 2.62 gives data for the few capitals whose record low
temperatures turn out to be whole numbers on both the Fahrenheit and Celsius
scales.
Display 2.62 Record low temperatures for seven capitals.

[Source: National Climatic Data Center, 2002.]
The dot plot in Display 2.63 shows that the temperatures are centered at about
32F, with an outlier at 22F. The spread and shape are hard to determine with
only seven values.
Display 2.63 Dot plot for record low temperatures in degrees

Fahrenheit for seven capitals.
2.4
Working with Summary Statistics
75
10-03-2009 13:59
Lesson
3 de 9
What happens to the shape, center, and spread of this distribution if you
convert each temperature to the number of degrees above or below freezing, 32F?
To find out, subtract 32 from each value, and plot the new values. Display 2.64
shows that the center of the dot plot is now at 0 rather than 32 but the spread and
shape are unchanged.
Display 2.64
Dot plot of the number of degrees Fahrenheit

above or below freezing for record low
temperatures for the seven capitals.
Adding (or subtracting) a constant to each value in a set of data doesnt

change the spread or the shape of a distribution but slides the entire distribution
a distance equivalent to the constant. Thus, the transformation amounts to a
recentering of the distribution.
What happens to the shape and spread of this distribution if you convert each
temperature to degrees Celsius? The Celsius scale measures temperature based on
the number of degrees above or below freezing, but it takes 1.8F to make 1C. To
),
convert, divide each value in Display 2.64 by 1.8 (or, equivalently, multiply by
and plot the new values. Display 2.65 shows that the center of the new dot plot
is still at 0 and the shape is the same. However, the spread has shrunk by a
factor of .
Display 2.65
Dot plot for record low temperatures in degrees

Celsius for the seven capitals.
Multiplying each value in a set of data by a positive constant doesnt change

the basic shape of the distribution. Both the mean and the spread are multiplied
by that number. This transformation amounts to a rescaling of the distribution.
[See Calculator Note 2G to explore on your calculator the effects of changing
units.]
Recentering and Rescaling a Data Set

Recentering a data setadding the same number c to all the values in the
setdoesnt change the shape or spread but slides the entire distribution by
the amount c, adding c to the median and the mean.
Rescaling a data setmultiplying all the values in the set by the same
positive number ddoesnt change the basic shape but stretches or shrinks
the distribution, multiplying the spread (IQR or standard deviation) by d and
multiplying the center (median or mean) by d.
76
10-03-2009 13:59
Lesson
4 de 9

D23. Suppose a U.S. dollar is worth 9.4 Mexican pesos.
a. A set of prices, in U.S. dollars, has mean $20 and standard deviation $5.
Find the mean and standard deviation of the prices expressed in pesos.
b. Another set of prices, in Mexican pesos, has a median of 94 pesos and
quartiles of 47 pesos and 188 pesos. Find the median and quartiles of the
same prices expressed in U.S. dollars.
The Influence of Outliers

A summary statistic is resistant to outliers if the summary statistic is not changed
very much when an outlier is removed from the set of data. If the summary
statistic tends to be affected by the removal of outliers, it is sensitive to outliers.
Display 2.66 shows a dot plot of the number of viewers of prime-time
television shows (in millions) in a particular week. (A boxplot of these data is
shown in Display 2.55 on page 70.) The three highest valuesthe three shows
with the largest numbers of viewersare outliers.
Display 2.66 Number of viewers of prime-time television shows

in a particular week.
The printout in Display 2.67 gives the summary statistics for all 101 shows.
Display 2.67 Printout of summary statistics for number of viewers.
The printout in Display 2.68 gives the summary statistics for the number of
viewers when the three outliers are removed from the set of data. Compare this
printout with the one in Display 2.67 and notice which summary statistics are
most sensitive to the outliers.
Display 2.68 Summary statistics for number of viewers without

outliers.
77
10-03-2009 13:59
Lesson
5 de 9

D24. Are these measures of center for the number of television viewers affected
much by the three outliers? (Refer to Displays 2.662.68.) Explain.
a. mean
b. median
D25. Are these measures of spread for the number of television viewers affected
much by the three outliers? Explain why or why not.
a. range
b. standard deviation
c. interquartile range
Percentiles and Cumulative Relative Frequency Plots

Percentiles measure position within a data set. The first quartile, Q1, of a
distribution is the 25th percentilethe value that separates the lowest 25% of the
ordered values from the rest. The median is the 50th percentile, and Q3 is the 75th
percentile. You can define other percentiles in the same way. The 10th percentile,
for example, is the value that separates the lowest 10% of ordered values in a
distribution from the rest. In general, a value is at the kth percentile if k% of all
values are less than or equal to it.
For large data sets, you might see data listed in a table or plotted in a graph,
like those for the SAT I critical reading scores in Display 2.69. Such a plot
is sometimes called a cumulative percentage plot or a cumulative relative
frequency plot. The table shows that, for example, 28% of the students received
a score of 450 or lower and about 13% received a score between 400 and 450.
Display 2.69
Cumulative relative frequency plot of SAT I critical

reading scores and percentiles, 20042005. [Source:
The College Board, www.collegeboard.org.]
10-03-2009 13:59
Lesson
6 de 9
[See Calculator Note 2H to learn how to construct cumulative relative frequency plots
on your calculator.]
Percentiles and Cumulative Relative Frequency Plots

D26. Refer to Display 2.69.
a. Use the plot to estimate the percentile for an SAT I critical reading score
of 425.
b. What two values enclose the middle 90% of SAT scores? The middle 95%?
c. Use the table to estimate the score that falls at the 40th percentile.
D27. What proportion of cases lie between the 5th and 95th percentiles of a
distribution? What percentiles enclose the middle 95% of the cases in a
distribution?
Summary 2.4: Working with Summary Statistics

Knowing which summary statistic to use depends on what use you have for that
summary statistic.
If a summary statistic doesnt change much whether you include or exclude
outliers from your data set, it is said to be resistant to outliers.
The median and quartiles are resistant to outliers.
The mean and standard deviation are sensitive to outliers.
Recentering a data setadding the same number c to all the valuesslides
the entire distribution. It doesnt change the shape or spread but adds c to the
median and the mean. Rescaling a data setmultiplying all the values by the same
nonzero number dis like stretching or squeezing the distribution. It doesnt
change the basic shape but multiplies the spread (IQR or standard deviation) by
| d | and multiplies the measure of center (median or mean) by d.
The percentile of a value tells you what percentage of all values lie at or below
the given value. The 30th percentile, for example, is the value that separates the
distribution into the lowest 30% of values and the highest 70% of values.
Practice
P26. A community in Nevada has 9751
households, with a median house price of
$320,000 and a mean price of $392,059.
a. Why is the mean larger than the median?
b. The property tax rate is about 1.15%.
What total amount of taxes will be
assessed on these houses?
c. What is the average amount of taxes per
house?
P27. A news release at www.polk.com stated that

the median age of cars being driven in 2004
was 8.9 years, the oldest to date. The median
was 8.3 years in 2000 and 7.7 years in 1995
a. Why were medians used in this news
story?
b. What reasons might there be for the
increase in the median age of cars? (The
median age in 1970 was only 4.9 years!)
79
10-03-2009 13:59
Lesson
7 de 9

P28. The mean height of a class of 15 children
is 48 in., the median is 45 in., the standard
deviation is 2.4 in., and the interquartile
range is 3 in. Find the mean, standard
deviation, median, and interquartile
range if
a. you convert each height to feet
b. each child grows 2 in.
c. each child grows 4 in. and you convert
the heights to feet
P29. Compute the means and standard deviations
(use the formula for s) of these sets of
numbers. Use recentering and rescaling
wherever you can to avoid or simplify the
arithmetic.
a. 1 2 3
b. 11 12 13
c. 10 20 30
d. 105 110 115
e. 800 900 1000
P30. The histogram and boxplot in Display 2.70
and the summary statistics in Display 2.71
show the record low temperatures for the
50 states.
a. Hawaii has a lowest recorded temperature
of 12F. The boxplot shows Hawaii as an
outlier. Verify that this is justified.
b. Suppose you exclude Hawaii from the
data set. Copy the table in Display 2.71,
substituting the value (or your best
estimate if you dont have enough
information to compute the value) of each
summary statistic with Hawaii excluded.
Exercises
E47.Discuss whether you would use the mean
or the median to measure the center of each
set of data and why you prefer the one you
chose.
a. the prices of single-family homes in your
neighborhood
80
Display 2.70 Record low temperatures for the

50 states. [Source: National Climatic Data
Center, 2002, www.ncdc.noaa.gov.]
Display 2.71 Summary statistics for lowest

temperatures for the 50 states.
Percentiles and Cumulative Relative

Frequency Plots
P31.Estimate the quartiles and the median of the
SAT I critical reading scores in Display 2.69
on page 78, and then use these values to
draw a boxplot of the distribution. What
is the IQR?
b. the yield of corn (bushels per acre) for a

sample of farms in Iowa
c. the survival time, following diagnosis, of
a sample of cancer patients
10-03-2009 13:59
Lesson
8 de 9
E48. Mean versus median.

a. You are tracing your family tree and
would like to go back to the year 1700.
To estimate how many generations back
you will have to trace, would you need to
know the median length of a generation
or the mean length of a generation?
b. If a car trip takes 3 h, do you need to
know the mean speed or the median
speed in order to find the total distance?
c. Suppose all trees in a forest are right
circular cylinders with radius 3 ft. The
heights vary, but the mean height is 45
ft, the median is 43 ft, the IQR is 3 ft, and
the standard deviation is 3.5 ft. From this
information, can you compute the total
volume of wood in all the trees?
E49. The histogram in Display 2.72 shows record
high temperatures for the 50 states.
Display 2.73 Summary statistics for record high

temperatures for the 50 U.S. states.
c. Are there any outliers in the data in C?

E50. Tell how you could use recentering and
rescaling to simplify the computation of the
mean and standard deviation for this list of
numbers:
5478.1 5478.3 5478.3 5478.9 5478.4 547
E51. Suppose a constant c is added to each value
in a set of data, x1, x2, x3, x4, and x5. Prove
that the mean increases by c by comparing
the formula for the mean of the original
data to the formula for the mean of the
recentered data.
E52. Suppose a constant c is added to each value
in a set of data, x1, x2, x3, x4, and x5. Prove
that the standard deviation is unchanged
by comparing the formula for the standard
deviation of the original data to the formula
for the standard deviation of the recentered
data.
E53. The cumulative relative frequency plot in
Display 2.74 shows the amount of change
carried by a group of 200 students. For
example, about 80% of the students had
$0.75 or less.
Display 2.72 Record high temperatures for the 50

U.S. states. [Source: National Climatic Data
Center, 2002, www.ncdc.noaa.gov.]
a. Suppose each temperature is converted

from degrees Fahrenheit, F, to degrees
Celsius, C, using the formula
If you make a histogram of the

temperatures in degrees Celsius, how
will it differ from the one in Display 2.72?
b. The summary statistics in Display 2.73
are for record high temperatures in
degrees Fahrenheit. Make a similar table
for the temperatures in degrees Celsius.
Display 2.74
Cumulative percentage plot of amount

of change.
a. From this plot, estimate the median

amount of change.
81
10-03-2009 13:59
Lesson
9 de 9
b. Estimate the quartiles and the

interquartile range.
c. Is the original set of amounts of change
skewed right, skewed left, or symmetric?
d. Does the data set look as if it should
be modeled by a normal distribution?
Explain your reasoning.
E54. Use Display 2.74 to make a boxplot of the
amounts of change carried by the students.
E55. Did you ever wonder how speed limits on
roadways are determined? Most government
jurisdictions set speed limits by this standard
practice, described on the website of the
Michigan State Police.
Speed studies are taken during times that
represent normal free-flow traffic. Since
modified speed limits are the maximum
allowable speeds, roadway conditions
must be close to ideal. The primary basis
for establishing a proper, realistic speed
limit is the nationally recognized method
of using the 85th percentile speed. This
is the speed at or below which 85% of the
traffic moves. [Source: www.michigan.gov.]
The 85th percentile speed typically is rounded
down to the nearest 5 miles per hour. The
table and histogram in Display 2.75 give the
measurements of the speeds of 1000 cars
on a stretch of road in Mellowville with no
curviness or other additional factors. At what
speed would the speed limit be set if the
guidelines described were followed?
Display 2.75 Speed of 1000 cars in Mellowville.
E56. Refer to the distribution of speeds in E55.

Make a cumulative relative frequency plot of
these speeds.
E57. The cumulative relative frequency plot in
Display 2.76 gives the ages of the CEOs
(Chief Executive Officers) of the 500 largest
U.S. companies. Does A, B, or C give its
median and quartiles? Using the diagram,
explain why your choice is correct.
A. Q1 51; median 56; Q3 60
B. Q1 50; median 60; Q3 70
C. Q1 25; median 50; Q3 75
82
10-03-2009 13:59
Lesson
1 de 12
E58. Refer to the distribution of ages in E57. Can

you give the median and quartiles of the
distribution of ages in months? If so, do it. If
not, explain why not.
Display 2.76 Cumulative relative frequency plot of

CEO ages. [Source: www.forbes.com]
2.5
These are both the

same normal curve.
The Normal Distribution

You have seen several reasons why the normal distribution is so important:
It tells you how variability in repeated measurements often behaves
(diameters of tennis balls).
It tells you how variability in populations often behaves (weights of pennies,
SAT scores).
It tells you how means (and some other summary statistics) computed from
random samples behave (the Westvaco case, Activity 1.2a).
In this section, you will learn that if you know that a distribution is normal
(shape), then the mean (center) and standard deviation (spread) tell you
everything else about the distribution. The reason is that, whereas skewed
distributions come in many different shapes, there is only one normal shape. Its
true that one normal distribution might appear tall and thin while another looks
short and fat. However, the x-axis of the tall, thin distribution can be stretched
out so that it looks exactly the same as the short, fat one.
The Standard Normal Distribution

Because all normal distributions have the same basic shape, you can use
recentering and rescaling to change any normal distribution to the one with
mean 0 and standard deviation 1. Solving problems involving normal
distributions depends on this important property.
The normal distribution with mean 0 and standard deviation 1 is called the
standard normal distribution. In this distribution, the variable along the
horizontal axis is called a z-score.
83
10-03-2009 14:00
Lesson
2 de 12
The standard normal distribution is symmetric, with total area under the
curve equal to 1, or 100%. To find the percentage, P, that describes the area to the
left of the corresponding z-score, you can use the z-table or your calculator.
The next two examples show you how to use the z-table, Table A on pages
824825.
Example: Finding the Percentage When You Know the z-Score

Find the percentage, P, of values less than z = 1.23, the shaded area in Display
2.77. Find the percentage greater than z = 1.23.
Display 2.77 The percentage of values less than z = 1.23.
Solution
Think of 1.23 as 1.2 +0.03. In Table A on pages 824825, find the row labeled
1.2 and the column headed .03. Where this row and column intersect, you find
the number .8907. That means that 89.07% of standard normal scores are less
than 1.23.
The total area under the curve is 1, so the proportion of values greater than
z = 1.23 is 1 0.8907, or 0.1093, which is 10.93%.
A graphing calculator will give you greater accuracy in finding the proportion
of values that lie between two specified values in a standard normal distribution.
For example, you can find the proportion of values that are less than 1.23 in a
standard normal distribution like this:
[To learn more about calculating the proportion of values between two
z-scores, see Calculator Note 2I.]
Example: Finding the z-Score When You Know the Percentage

Find the z-score that falls at the 75th percentile of the standard normal distribution,
that is, the z-score that divides the bottom 75% of values from the rest.
10-03-2009 14:00
Lesson
3 de 12
Solution
First make a sketch of the situation, as in Display 2.78.
Display 2.78 The z-score that corresponds to the 75th percentile.
Look for .7500 in the body of Table A. No value in the table is exactly equal to
.7500. The closest value is .7486. The value .7486 sits at the intersection of the row
labeled .60 and the column headed .07, so the corresponding z-score is roughly
0.60 + 0.07, or 0.67.
You can use a graphing calculator to find the 75th percentile of a standard
normal distribution like this:
[To learn more about finding the z-score that has a specified proportion of values
below it, see Calculator Note 2J.]

D28. For the standard normal distribution,
a.
b.
c.
d.
what is the median?

what is the lower quartile?
what z-score falls at the 95th percentile?
what is the IQR?
Standard Units: How Many Standard Deviations Is It

from Here to the Mean?
Converting to standard units, or standardizing, is the two-step process of
recentering and rescaling that turns any normal distribution into the standard
normal distribution.
First you recenter all the values of the normal distribution by subtracting the
mean from each. This gives you a distribution with mean 0. Then you rescale by
85
10-03-2009 14:00
Lesson
4 de 12
dividing all the values by the standard deviation. This gives you a distribution
with standard deviation 1. You now have a standard normal distribution. You
can also think of the two-step process of standardizing as answering two
questions: How far above or below the mean is my score? How many standard
deviations is that?
The standard units or z-score is the number of standard deviations that a
given x-value lies above or below the mean.
How far and which way to the mean?
x mean
How many standard deviations is that?
Example: Computing a z-Score

In a recent year, the distribution of SAT I math scores for the incoming class at the
University of Georgia was roughly normal, with mean 610 and standard deviation
69. What is the z-score for a University of Georgia student who got 560 on the
math SAT?
Solution
A score of 560 is 50 points below the mean of 610. This is
deviation below the mean. Alternatively, using the formula,
or 0.725 standard
So the students z-score is 0.725.
To unstandardize, think in reverse. Alternatively, you can solve the z-score

formula for x and get
x = mean + z SD
Example: Finding the Value When You Know the z-Score

What was a University of Georgia students SAT I math score if his or her score
was 1.6 standard deviations above the mean?
Solution
The score that is 1.6 standard deviations above the mean is
x = mean + z SD = 610 + 1.6(69) 720
86
10-03-2009 14:00
Lesson
5 de 12
Example: Using z-Scores to Make a Comparison

In the United States, heart disease kills roughly one-and-a-quarter times as many
people as cancer. If you look at the death rate per 100,000 residents by state, the
distributions for the two diseases are roughly normal, provided you leave out
Alaska and Utah, which are outliers because of their unusually young populations.
The means and standard deviations for all 50 states are given here.
Alaska had 88 deaths per 100,000 residents from heart disease, and 111 from
cancer. Explain which death rate is more extreme compared to other states. [Source:
Centers for Disease Control, National Vital Statistics Report, vol. 53, no. 5, October 12, 2004.]
Solution
Alaskas death rate for heart disease is 2.88 standard deviations below the mean.
The death rate for cancer is 2.74 standard deviations below the mean. These rates
are about equally extreme, but the death rate for heart disease is slightly more
extreme.
Standard Units
D29.Standardizing is a process that is similar to other processes you have seen
already.
a. Youre driving at 60 mi/h on the interstate and are now passing the
marker for mile 200, and your exit is at mile 80. How many hours from
your exit are you?
b. What two arithmetic operations did you do to get the answer in part a?
Which operation corresponds to recentering? Which corresponds to
rescaling?
Solving the Unknown Percentage Problem

and the Unknown Value Problem
Now you know all you need to know to analyze situations involving two related
problems concerning a normal distribution: finding a percentage when you know
the value, and finding the value when you know the percentage.
87
10-03-2009 14:00
Lesson
6 de 12
Example: Percentage of Males Taller Than 74 Inches

For groups of similar individuals, heights often are approximately normal in
their distribution. For example, the heights of 18- to 24-year-old males in
the United States are approximately normal, with mean 70.1 in. and standard
deviation 2.7 in. What percentage of these males are more than 74 in. tall?
[Source: U.S. Census Bureau, Statistical Abstract of the United States, 1991.]
Solution
First make a sketch of the situation, as in Display 2.79. Draw a normal shape
above a horizontal axis. Place the mean in the middle on the axis. Then mark and
label the points that are two standard deviations either side of the mean, 64.7 and
75.5, so that about 95% of the values lie between them. Next, mark and label the
points that are one and three standard deviations either side of the mean (67.4
and 72.8, and 62 and 78.2). Finally, estimate the location of the given value of x
and mark it on the axis.
Display 2.79 The percentage of heights greater than 74 in.
Standardize:
Look up the proportion: The area to the left of the z-score 1.44 is 0.9251, so the
proportion of males taller than 74 in. is 1 0.9251, or 0.0749 or 7.49%.
Example: Percentage of Males Between 72 and 74 Inches Tall

The heights of 18- to 24-year-old
males in the United States are
approximately normal, with mean
70.1 in. and standard deviation 2.7 in.
What percentage of these males are
between 72 and 74 in. tall?
88
10-03-2009 14:00
Lesson
7 de 12
Solution
First make a sketch, as in Display 2.80.
Display 2.80 The percentage of male heights between 72 and

74 in.
Standardize: From the previous example, a height of 74 in. has a z-score of 1.44.
For a height of 72 in.,
Look up the proportion: The area to the left of the z-score 1.44 is 0.9251. The
area to the left of the z-score 0.70 is 0.7580. The area you want is the area between
these two z-scores, which is 0.9251 0.7580, or 0.1671. So the percentage of
18- to 24-year-old males between 72 and 74 in. tall is about 16.71%.
You can also use a graphing calculator to find this value. [See Calculator
Note 2I for more details.]
Example: 75th Percentile of Female Heights

The heights of females in the United States
who are between the ages of 18 and 24 are
approximately normally distributed, with
mean 64.8 in. and standard deviation
2.5 in. What height separates the shortest
75% from the tallest 25%?
89
10-03-2009 14:00
Lesson
8 de 12
Solution
First make a sketch, as in Display 2.81.
Display 2.81 The 75th percentile in height for women age 18

to 24.
Look up the z-score: If the proportion P is 0.75, then from Table A, you find that z
is approximately 0.67.
Unstandardize:
For an unknown percentage problem:

First standardize by converting the given value to a z-score:
Then look up the percentage.
For an unknown value problem, reverse the process:
First look up the z-score corresponding to the given percentage. Then
unstandardize:
Solving the Unknown Percentage Problem and the

Unknown Value Problem
D30. Age of cars. The cars in Clunkerville have a mean age of 12 years and a
standard deviation of 8 years. What percentage of cars are more than 4 years
old? (Warning: This is a trick question.)
Central Intervals for Normal Distributions

You learned in Section 2.1 that if a distribution is roughly normal, about 68% of
the values lie within one standard deviation of the mean. It is helpful to memorize
this fact as well as the others in the box on the next page.
90
10-03-2009 14:00
Lesson
9 de 12
Example: Middle 90% of Death Rates from Cancer

According to the table on page 87, the death rates per 100,000 residents from
cancer are approximately normal, with mean 196 and SD 31. The middle 90% of
death rates are between what two numbers?
Solution
The middle 90% of values in this distribution lie within 1.645 standard
deviations of the mean, 196. That is, about 90% of the values lie in the interval
196 1.645(31), or between about 145 and 247.
91
10-03-2009 14:00
Lesson
10 de 12
You can confirm this result using your calculator: Shade the area under
this normal curve between 145 and 247 and calculate the area. [See Calculator
Note 2K.]

D31. Use Table A on pages 824825 to verify that 99.7% of the values in a normal
distribution lie within three standard deviations of the mean.
Summary 2.5: The Normal Distribution

The standard normal distribution has mean 0 and standard deviation 1. All
normal distributions can be converted to the standard normal distribution by
converting to standard units:
First, recenter by subtracting the mean.
Then rescale by dividing by the standard deviation:
Standard units z tell how far a value x is from the mean, measured in standard
deviations. If you know z, you can find x by using the formula x = mean + z SD.
If your population is approximately normal, you can compute z and then
use Table A or your calculator to find the corresponding proportion. Be sure to
make a sketch so that you know whether to use the proportion in the table or to
subtract that proportion from 1.
For any normal distribution,
68% of the values lie within 1 standard deviation of the mean
90% of the values lie within 1.645 standard deviations of the mean
95% of the values lie within 1.960 (or about 2) standard deviations of the mean
99.7% (or almost all) of the values lie within 3 standard deviations of the mean
Practice
P32. Find the percentage of values below
each given z-score in a standard normal
distribution.
a. 2.23 b. 1.67 c. 0.40 d. 0.80
92
P33. Find the z-score that has the given

percentage of values below it in a standard
a. 32%
b. 41%
c. 87%
d. 94%
10-03-2009 14:00
Lesson
11 de 12
P34. What percentage of values in a standard

a. 1.46 and 1.46?
b. 3 and 3?
P35. For a standard normal distribution, what
interval contains
a. the middle 90% of z-scores?
b. the middle 95% of z-scores?
Standard Units
P36. Refer to the table in the example on page 87.
a. California had 196 deaths from heart
disease and 154 deaths from cancer per
100,000 residents. Which rate is more
extreme compared to other states? Why?
b. Florida had 295 deaths from heart disease
and 234 deaths from cancer per 100,000
residents. Which rate is more extreme?
c. Colorado had an unusually low rate of
heart disease, 143 deaths per 100,000
residents. Hawaii had an unusually low
rate of cancer, 156 deaths per 100,000
residents. Which is more extreme?
Solving the Unknown Percentage Problem
and the Unknown Value Problem
2.7 in. The heights of 18- to 24-year-old

females are also approximately normally
distributed and have mean 64.8 in. and
standard deviation 2.5 in.
a. Estimate the percentage of U.S. males
between 18 and 24 who are 6 ft tall or
taller.
b. How tall does a U.S. woman between 18
and 24 have to be in order to be at the
35th percentile of heights?
P38. Refer to the table in the example on page 87.
a. The middle 90% of the states death rates
from heart disease fall between what two
numbers?
b. The middle 68% of death rates from heart
disease fall between what two numbers?
P39. Refer to the information in P37. Which of
the following heights are outside the middle
95% of the distribution? Which are outside
the middle 99%?
A. a male who is 79 in. tall
B. a female who is 68 in. tall
C. a male who is 65 in. tall
D. a female who is 65 in. tall
P37. The heights of 18- to 24-year-old males in

the United States are approximately normal,
with mean 70.1 in. and standard deviation
Exercises
E59. What percentage of values in a standard
normal distribution fall
a. below a z-score of 1.00? 2.53?
b. below a z-score of 1.00? 2.53?
c. above a z-score of 1.5?
d. between z-scores of 1 and 1?
E60. On the same set of axes, draw two normal
curves with mean 50, one having standard
deviation 5 and the other having standard
deviation 10.
E61. Standardizing. Convert each of these values
to standard units, z. (Do not use a calculator.
These are meant to be done in your head.)
a.
b.
c.
d.
e.
f.
x = 12, mean 10, SD 1

x = 12, mean 10, SD 2
x = 12, mean 9, SD 2
x = 12, mean 9, SD 1
x = 7, mean 10, SD 3
x = 5, mean 10, SD 2
E62. Unstandardizing. Find the value of x that was

converted to the given z-score.
a. z = 2, mean 20, SD 5
b. z = 1, mean 25, SD 3
c. z = 1.5, mean 100, SD 10
d. z = 2.5, mean 10, SD 0.2
93
10-03-2009 14:00
Lesson
12 de 12
E63. SAT I critical reading scores are scaled so that

they are approximately normal, with mean
about 505 and standard deviation about 111.
a. Find the probability that a randomly
selected student has an SAT I critical
reading score
i. between 400 and 600
ii. over 700
iii. below 450
b. What SAT I critical reading scores fall in
the middle 95% of the distribution?
E64. SAT I math scores are scaled so that they are
approximately normal, with mean about 511
and standard deviation about 112. A college
wants to send letters to students scoring in
the top 20% on the exam. What SAT I math
score should the college use as the dividing
line between those who get letters and those
who do not?
E65. Height limitations for flight attendants.
To work as a flight attendant for United
Airlines, you must be between 5 ft 2 in. and
6 ft tall. [Source: www.ual.com.] The mean height
of 18- to 24-year-old males in the United
States is about 70.1 in., with a standard
deviation of 2.7 in. The mean height of
18- to 24-year-old females is about 64.8 in.,
with a standard deviation of 2.5 in. Both
distributions are approximately normal.
What percentage of men this age meet
the height limitation? What percentage of
women this age meet the height limitation?
E66. Where is the next generation of male
professional basketball players coming from?
a. The mean height of 18- to 24-year-old
males in the United States is approximately
normally distributed, with mean 70.1 in.
and standard deviation 2.7 in. Use this
information to approximate the percentage
of men in the United States between the
ages of 18 and 24 who are as tall as or taller
than each basketball player listed here.
Then, using the fact that there are about
13 million men between the ages of 18 and
24 in the United States, estimate how many
are as tall as or taller than each player.
94
i. Shawn Marion, 6 ft 7 in.

ii. Allen Iverson, 6 ft 0 in.
iii. Shaquille ONeal, 7 ft 1 in.
b. Distributions of real data that are
approximately normal tend to have
heavier tails than the ideal normal curve.
Does this mean your estimates in part a
are too small, too big, or just right?
E67. Puzzle problems. Problems that involve
computations with the normal distribution
have four quantities: mean, standard
deviation, value x, and proportion P below
value x. Any three of these values are enough
to determine the fourth. Think of each row
in this table as little puzzles, and find the
missing value in each case. This isnt the sort
of thing you are likely to run into in practice,
but solving the puzzles can help you become
more skilled at working with the normal
distribution.
E68. More puzzle problems. In each row of this

table, assume the distribution is normal.
Knowing any two of the mean, standard
deviation, Q1, and Q3 is enough to determine
the other two. Complete the table.
E69. ACT scores are approximately normally

distributed, with mean 18 and standard
deviation 6. Without using your calculator,
roughly what percentage of scores are
between 12 and 24? Between 6 and 30?
Above 24? Below 24? Above 6? Below 6?
E70. A group of subjects tested a certain brand of
foam earplug. The number of decibels (dB)
that noise was reduced for these subjects was
10-03-2009 14:00
Lesson
1 de 9
c. Use your work in part b to suggest a

rough rule for using the mean and
standard deviation of a set of positive
values to check whether it is possible that
a distribution might be approximately
normal.
approximately normally distributed, with

mean 30 dB and standard deviation 3.6 dB.
The middle 95% of noise reductions were
between what two values?
E71.The heights of 18- to 24-year-old males in
the United States are approximately normally
distributed with mean 70.1 in. and standard
deviation 2.7 in.
a. If you select a U.S. male between ages
18 and 24 at random, what is the
approximate probability that he is less
than 68 in. tall?
b. There are roughly 13 million 18- to
24-year-old males in the United States.
About how many are between 67 and
68 in. tall?
c. Find the height of 18- to 24-year-old
males that falls at the 90th percentile.
E72.If the measurements of height are
transformed from inches into feet, will that
change the shape of the distribution in E71?
Describe the distribution of male heights in
terms of feet rather than inches.
E73.The British monarchy. Over the 1200 years
of the British monarchy, the average reign of
kings and queens has lasted 18.5 years, with
a standard deviation of 15.4 years.
a. What can you say about the shape of the
distribution based on the information
given?
b. Suppose you made the mistake of
assuming a normal distribution. What
fraction of the reigns would you estimate
lasted a negative number of years?
E74.NCAA scores. The histogram in Display 2.82

was constructed from the total of the scores
of both teams in all NCAA basketball
play-off games over a 57-year period.
Display 2.82 Total points scored in NCAA play-off

games. [Source: www.ncaa.com.]
a. Approximate the mean of this distribution.

b. Approximate the standard deviation of
this distribution.
c. Between what two values do the middle
95% of total points scored lie?
d. Suppose you choose a game at random
from next years NCAA play-offs. What is
the approximate probability that the total
points scored in this game will exceed
150? 190? Do you see any potential
weaknesses in your approximations?
Chapter Summary
Distributions come in various shapes, and the appropriate summary statistics (for
center and spread) usually depend on the shape, so you should always start with a
plot of your data.
Common symmetric shapes include the uniform (rectangular) distribution
and the normal distribution. There are also various skewed distributions. Bimodal
distributions often result from mixing cases of two kinds.
Chapter Summary
95
10-03-2009 14:09
Lesson
2 de 9
Dot plots, stemplots, and histograms show distributions graphically and let
you estimate center and spread visually from the plot.
For approximately normal distributions, you ordinarily use the mean (balance
point) and standard deviation as the measure of center and spread. If you know
the mean and standard deviation of a normal distribution, you can use z-scores
and Table A or your calculator to find the percentage of values in any interval.
The mean and standard deviation are not resistanttheir values are sensitive
to outliers. For a description of a skewed distribution, you should consider using
the median (halfway point) and quartiles (medians of the lower and upper halves
of the data) as summary statistics.
Later on, when you make inferences about the entire population from a
sample taken from that population, the sample mean and standard deviation will
be the most useful summary statistics, even if the population is skewed.
E75. The map in Display 2.83, from the U.S.

National Weather Service, gives the number of
tornadoes by state, including the District of
Columbia.
f. Describe the shape, center, and spread

of the distribution of the number of
tornadoes.
E76. Display 2.84 shows some results of the Third
International Mathematics and Science study
for various countries. Each case is a school.
Display 2.83 The number of tornadoes per state in a

recent year. [Source: www.ncdc.noaa.gov.]
a. Make a stemplot of the number of

tornadoes.
b. Write the five-number summary.
Display 2.84 Boxplots of mathematics instruction

time by country for 9-year-olds. [Source:
Report #8, April 1998, of the Third International
Mathematics and Science Study (TIMSS), p. 6.]
c. Identify any outliers.

d. Draw a boxplot.
e. Compare the information in your
stemplot with the information in your
boxplot. Which plot is more informative?
96
a. Estimate the median for the United

States. Use this median value in a
sentence that makes it clear what the
median represents in this context.
10-03-2009 14:09
Lesson
3 de 9
b. Why are there only lines and no boxes for

Norway and Singapore?
c. Describe how the distribution of values
for the United States compares to the
distributions of values for the other
countries.
E77.A university reports that the middle 50% of
the SAT I math scores of its students were
between 585 and 670, with half the scores
above 605 and half below.
a. What SAT I math scores would be
considered outliers for that university?
b. What can you say about the shape of this
distribution?
E78.These statistics summarize a set of television
ratings from a week without any special
programming. Are there any outliers among
the 113 ratings?
a. From your knowledge of the world,

match the boxplots to the correct region.
b. Match the summary statistics (for
Groups AC) to the correct boxplot (for
Regions 13).
E80. The National Climatic Data Center records
high and low temperatures by state since
1890. Stem-and-leaf plots of the years each
state had its lowest temperature and the years
each state had its highest temperature are
shown in Display 2.86. What do the stems
represent? What do the leaves represent?
Compare the two distributions with respect
to shape, center, spread, and any interesting
features.
E79.The boxplots in Display 2.85 show the life

expectancies for the countries of Africa,
Europe, and the Middle East. The table
shows a few of the summary statistics for
each of the three data sets.
Display 2.86
Display 2.85 Life expectancies for the countries

of Africa, Europe, and the Middle East.
[Source: Population Reference Bureau, World
Population Data Sheet, 2005.]
Stem-and-leaf plots of record low and

high temperatures of states. [Source: National
Climatic Data Center, 2002.]
E81. A distribution is symmetric with

approximately equal mean and median. Is
it necessarily the case that about 68% of the
values are within one standard deviation of
the mean? If yes, explain why. If not, give an
example.
Chapter Summary
97
10-03-2009 14:09
Lesson
4 de 9
E82. Display 2.87 shows two sets of graphs. The

first set shows smoothed histograms IIV for
four distributions. The second set shows the
corresponding cumulative relative frequency
plots, in scrambled order AD. Match each
plot in the first set with its counterpart in the
second set.
Distributions
Display 2.87 Four distributions with different

shapes and their cumulative relative
frequency plots.
98
E83. The average number of pedestrian deaths

annually for 41 metropolitan areas is given
in Display 2.88.
Display 2.88 Average annual pedestrian deaths.

[Source: Environmental Working Group and
the Surface Transportation Policy Project.
Compiled from National Highway Traffic Safety
Administration and U.S. Census data. USA Today,
April 9, 1997.]
10-03-2009 14:09
Lesson
5 de 9
a. What is the median number of deaths?

Write a sentence explaining the meaning
of this median.
Display 2.90 shows dot plots of the same

data.
b. Is any city an outlier in terms of the

number of deaths? If so, what is the city,
and what are some possible explanations?
c. Make a plot of the data that you think
will show the distribution in a useful
way. Describe why you chose that plot
and what information it gives you about
average annual pedestrian deaths.
d. In which situations might giving the
death rate be more meaningful than
giving the number of deaths?
E84. The side-by-side boxplots in Display 2.89
give the percentage of 4th-grade-age
children who are still in school on various
continents according to the United Nations.
Each case is a country. The four regions
marked 1, 2, 3, and 4 are Africa, Asia,
Europe, and South/Central America, not
necessarily in that order.
a. Which region do you think corresponds
to which number?
b. Is the distribution of values for any
region skewed left ? Skewed right?
Symmetrical?
Display 2.89 Boxplots of the percentage of

4th-grade-age children still in school
in countries of the world, by continent.
[Source: 1993 Information Please Almanac.]
Display 2.90 Dot plots of the percentage of 4th

grade- age children still in school in
countries of the world, by continent.
c. Match each dot plot to the corresponding

boxplot.
d. In what ways do the boxplots and
dot plots give diff erent impressions?
Why does this happen? Which type
of plot gives a better impression of the
distributions?
E85.The first AP Statistics Exam was given in
1997. The distribution of scores received
by the 7667 students who took the exam is
given in Display 2.91. Compute the mean
and standard deviation of the scores.
Display 2.89 Scores on the first AP Statistics Exam.

[Source:The College Board.]
Chapter Summary
99
10-03-2009 14:09
Lesson
6 de 9
E86. For the countries of Europe, many average

life expectancies are approximately the
same, as you can see from the stemplot in
Display 2.53 on page 69. Use the formulas
for the summary statistics of values in a
frequency table to compute the mean and
standard deviation of the life expectancies
for the countries of Europe.
E87. Construct a set of data in which all values
are larger than 0, but one standard deviation
below the mean is less than 0.
E88. Without computing, what can you say about
the standard deviation of this set of values:
4, 4, 4, 4, 4, 4, 4, 4?
E89. In this exercise, you will compare how
dividing by n versus n 1 affects the SD for
various values of n. So that you dont have to
compute the sum of the squared deviations
each time, assume that this sum is 400.
a. Compare the standard deviation that
would result from
i. dividing by 10 versus dividing by 9
ii. dividing by 100 versus dividing by 99
iii. dividing by 1000 versus dividing
by 999
b. Does the decision to use n or n 1 in the
formula for the standard deviation matter
very much if the sample size is large?
E90. If two sets of test scores arent normally
distributed, its possible to have a larger
z-score on Test II than on Test I yet be in
a lower percentile on Test II than on Test I.
The computations in this exercise will
illustrate this point.
a. On Test I, a class got these scores: 11, 12,
13, 14, 15, 16, 17, 18, 19, 20. Compute the
z-score and the percentile for the student
who got a score of 19.
b. On Test II, the class got these scores:
1, 1, 1, 1, 1, 1, 1, 18, 19, 20. Compute the
z-score and the percentile for the student
who got a score of 18.
c. Do you think the student who got a score
of 19 on Test I or the student who got a
score of 18 on Test II did better relative to
the rest of the class?
100
E91. The average income, in dollars, of people in

each of the 50 states was computed for 1980
and for 2000. Summary statistics for these
two distributions are given in Display 2.92.
Display 2.84 Summary statistics of the average

income, in dollars, for the 50 states
for 1980 and 2000. [Source: U.S. Census
Bureau, Statistical Abstract of the United States,
20042005.]
a. Explain the meaning of $7,007 for the

minimum in 1980.
b. Are any states outliers for either year?
c. In 2000 the average personal income
in Alabama was $23,768, and in 1980 it
was $7,836. Did the income in Alabama
change much in relation to the other
states? Explain your reasoning.
E92. For these comparisons, you will either use
the SAT I critical reading scores in Display
2.69 on page 78 or assume that the scores
have a normal distribution with mean 505
and standard deviation 111.
a. Estimate the percentile for an SAT I
critical reading score of 425 using the
cumulative relative frequency plot. Then
find the percentile for a score of 425 using
a z-score. Are the two values close?
b. Estimate the SAT I critical reading score
that falls at the 40th percentile, using the
table in Display 2.69. Then find the 40th
percentile using a z-score. Are the two
values close?
c. Estimate the median from the cumulative
relative frequency plot. Is this value close
to the median you would get by assuming
a normal distribution of scores?
10-03-2009 14:09
Lesson
7 de 9
d. Estimate the quartiles and the

interquartile range using the plot. Find
the quartiles and interquartile range
assuming a normal distribution of scores.
E93. For 17-year-olds in the United States, blood
cholesterol levels in milligrams per deciliter
have an approximately normal distribution
with mean 176 mg/dL and standard deviation
30 mg/dL. The middle 90% of the cholesterol
levels are between what two values?
E94. Display 2.93 shows the distribution of
batting averages for all 187 American League
baseball players who batted 100 times or
more in a recent season. (A batting average
is the fraction of times that a player hits
safelythat is, the hit results in a player
advancing to a baseusually reported to
three decimal places.)
c. Use your mean and SD from part b to

compute an estimate of the percentage of
players who batted over .300 (or 300).
d. Now use the histogram to estimate the
percentage of players who batted over .300.
Compare to your estimate from part c.
E95. How good are batters in the National
League? Display 2.94 shows the distribution
of batting averages for all 223 National
Leaguers who batted 100 times or more in a
recent season.
Display 2.94 National League batting averages.

[Source: CBS SportsLine.com, www.sportsline
.com.]
Display 2.93 American League batting averages.

[Source: CBS SportsLine.com, www.sportsline
.com.]
a. Do the batting averages appear to be

approximately normally distributed?
b. Approximate the mean and standard
deviation of the batting averages from the
histogram.
a. Approximate the mean and standard

deviation of the batting averages from the
histogram.
b. Compare the distributions of batting
averages for the two leagues. (See E94
for the American League.) What are
the main differences between the two
distributions?
c. A batter hitting .300 in the National
League is traded to a team in the
American League. What batting average
could be expected of him in his new
league if he maintains about the same
position in the distribution relative to his
peers?
Chapter Summary
101
10-03-2009 14:09
Lesson
8 de 9
AP1. These summary statistics are for the

distribution of the populations of the major
cities in Brazil.
Which of the following best describes the

shape of this distribution?
skewed right without outliers
skewed right with at least one outlier
roughly normal, without outliers
skewed left without outliers
skewed left with at least one outlier
AP2. Which of these lists contains only summary
statistics that are sensitive to outliers?
mean, median, and mode
standard deviation, IQR, and range
mean and standard deviation
median and IQR
five-number summary
AP3. This stem-and-leaf plot shows the ages of
CEOs of 60 corporations whose annual sales
were between $5 million and $350 million.
Which of the following is not a correct
statement about this distribution?
The distribution is skewed left (towards
smaller numbers).
The oldest of the 60 CEOs is 74 years old.
The distribution has no outliers.
The range of the distribution is 42.
The median of the distribution is 50.
AP4. A traveler visits Europe and stays thirty

days in thirty different hotels, paying each
102
day with her credit card. The hotels charged

a mean price of 50 euros, with a standard
deviation of 10 euros. When the charges
appear on her credit card statement in
the United States, she finds that her bank
charged her $1.20 per euro, plus a $5 fee
for each transaction. What is the mean and
standard deviation of the thirty daily hotel
charges in dollars, including the fee?
mean $50, standard deviation $17
AP5. The scores on a nationally administered test
are approximately normally distributed with
mean 47.3 and standard deviation 17.3.
Approximately what must a student have
scored to be in the 95th percentile nationally?
AP6. A particular brand of cereal boxes is labeled

16 oz. This dot plot shows the actual
weights of 100 randomly selected boxes.
Which of the following is the best estimate
of the standard deviation of these weights?
AP7. The distribution of the number of points

earned by the thousands of contestants in
the Game of Pig World Championship has
mean 20 and standard deviation 6. What
proportion of the contestants earned more
than 26 points?
10-03-2009 14:09
Lesson
9 de 9
Less than 1%
16%
32%
84%
This proportion cannot be determined
from the information given.
AP8. Anya scored 70 on a statistics test for which
the mean was 60 and the standard deviation
was 10. She also scored 60 on a chemistry
test for which the mean was 50 and the
standard deviation was 5. If the scores for
both tests were approximately normally
distributed, which best describes how Anya
did relative to her classmates?
Anya did better on the statistics test than
she did on the chemistry test because she
scored 10 points higher on the statistics
test than on the chemistry test.
Anya did equally well relative to her
classmates on each test, because she
scored 10 points above the mean on each.
Anya did better on the chemistry test
than she did on the statistics test because
she scored 2 standard deviations above
the mean on the chemistry test and only
1 standard deviation above the mean on
the statistics test.
Its impossible to tell without knowing the
number of points possible on each test.
Its impossible to tell without knowing
the number of students in each class.
Investigative Task
AP9. A game invented by three college students
involves giving the name of an actor and
then trying to connect that actor with
actor Kevin Bacon, counting the number
of steps needed. For example, Sarah Jessica
Parker has a Bacon number of 1 because
she appeared in the same movie as Kevin
Bacon, Footloose (1984). Will Smith has a
Bacon number of 2. He has never appeared
in a movie with Kevin Bacon; however, he
was in Bad Boys II (2003) with Michael
Shannon, who was in The Woodsman (2004)
with Kevin Bacon. Display 2.95 gives the

number of links required to connect each
of the 645,957 actors in the Internet Movie
Database to Kevin Bacon.
Display 2.95 Bacon numbers. [Source: www.cs.virginia

.edu.]
a. How many people have appeared in a

movie with Kevin Bacon?
b. Who is the person with Bacon number 0?
It has been questioned whether Kevin Bacon
was the best choice for the center of the
Hollywood universe. A possible challenger
is Sean Connery. See Display 2.96.
Display 2.96 Connery numbers.
c. Do you think Kevin Bacon or Sean

Connery better deserves the title
Hollywood center? Make your case
using statistical evidence (as always).
d. (For movie fans.) What is Bacons
Connery number? What is Connerys
Bacon number?
AP Sample Test
103
10-03-2009 14:09
Lesson
1 de 2
Relationships Between Two

Quantitative Variables
What variables
contribute to a
college having a
high graduation
rate? Scatterplots,
correlation, and
regression are
the basic tools used
to describe relationships between two
quantitative variables.
10-03-2009 14:13
Lesson
2 de 2
In Chapter 2, you compared the speeds of predators and nonpredators. Not

surprisingly, among mammals meat eaters were usually faster than vegetarians.
Some nonpredators, however, such as the horse (48 mi/h) and the elk (45 mi/h),
were faster than some predators, such as the dog (39) and the grizzly (30). Because
of this variability, comparing the two groups was a matter for statistics; that is, you
needed suitable plots and summaries.
The comparison involved a relationship between two variables, one
quantitative (speed) and one categorical (predator or not). In this chapter, youll
learn how to explore and summarize relationships in which both variables are
quantitative. The data set on mammals in Display 2.24 on page 43 raises many
questions of this sort: Do mammals with longer average longevity also have
longer maximum longevity? Is there a relationship between speed and longevity?
The approach to describing distributions in Chapter 2 boiled down to finding
shape, center, and spread. For distributions that are approximately normal, two
numerical summariesthe mean for center, the standard deviation for spread
tell you basically all you need to know. When comparing two quantitative
variables, you can see the shape of the distribution by making a scatterplot. For
scatterplots with points that lie in an oval cloud, it turns out once again that two
summaries tell you pretty much all you need to know: the regression line and the
correlation. The regression line tells about center: What is the equation of the line
that best fits the cloud of points? The correlation tells about spread: How spread
out are the points around the line?
In this chapter, you will learn to

describe the pattern in a scatterplot, and decide what its shape tells you
about the relationship between the two variables
find a regression line through the center of a cloud of points to summarize
the relationship
use the correlation as a measure of how spread out the points are from
this line
use diagnostic tools to check for information the summaries dont tell you,
and decide what to do with that information
make shape-changing transformations to re-express a curved relationship so
that you can use a line as a summary
10-03-2009 14:13
Lesson
1 de 10
3.1
Scatterplots
In Chapter 1, you explored the relationship between the ages of employees at
Westvaco and whether the employees were laid off when the company downsized.
There is more to see. In the scatterplot in Display 3.1, for example, each employee
is represented by a dot that shows the year the employee was born plotted against
the year the employee was hired.
A scatterplot shows the

relationship between
two quantitative
variables.
Display 3.1 Year of birth versus year of hire for the 50 employees
in Westvaco Corporations engineering department.
In this scatterplot, you can see a moderate positive association: Employees

hired in an earlier year generally were born in an earlier year, and employees hired
in a later year generally were born in a later year. This trend is fairly linear. You
can visualize a summary line going through the center of the data from lower left
to upper right. As you move to the right along this line, the points fan out and
cluster less closely around the line.
Sometimes its easier to think about peoples ages than about the years
they were born. The scatterplot in Display 3.2 shows the ages of the Westvaco
employees at the time layoffs began plotted against the year they were hired. This
scatterplot shows a moderate negative association: Those people hired in later
years generally were younger at the time of the layoffs than people hired in
earlier years.
Display 3.2 Age at layoffs versus year of hire for the 50 employees
in Westvaco Corporations engineering department.
106 Chapter 3 Relationships Between Two Quantitative Variables
10-03-2009 14:20
Lesson
2 de 10
[You can use a graphing calculator to create scatterplots. See Calculator Note 3A.]
Interpreting Scatterplots
D1. You will examine Displays 3.1 and 3.2 more closely in these questions and in
E7 on page 114.
a. Why should the two variables plotted in Display 3.1 show a positive
association and the two variables plotted in Display 3.2 show a negative
association?
b. Why do all but one of the points in Display 3.1 lie on or below a diagonal
line running from the lower left to the upper right?
c. Is this sentence a reasonable interpretation of Display 3.2? As time
passed, Westvaco tended to hire younger and younger people.
Describing the Pattern in a Scatterplot

Shape, trend, and
strength
For the distribution of a single quantitative variable, shape, center, and spread
is a useful summary. For bivariate (two-variable) quantitative data, the summary
becomes shape, trend, and strength.
You might find it helpful to follow this set of steps as you practice describing
scatterplots.
1. Identify the variables and cases. On a scatterplot, each point represents
a case, with the x-coordinate equal to the value of one variable and the
y-coordinate equal to the value of the other variable. You should describe
the scale (units of measurement) and range of each variable.
2. Describe the overall shape of the relationship, paying attention to
linearity: Is the pattern linear (scattered about a line) or curved?
clusters: Is there just one cluster, or is there more than one?
outliers: Are there any striking exceptions to the overall pattern?
3. Describe the trend. If as x gets larger y tends to get larger, there is a positive
trend. (The cloud of points tends to slope upward as you go from left to
right.) If as x gets larger y tends to get smaller, there is a negative trend. (The
cloud of points tends to slope downward as you go from left to right.)
4. Describe the strength of the relationship. If the points cluster closely around
an imaginary line or curve, the association is strong. If the points are
scattered farther from the line, the association is weak.
If, as in Display 3.1, the points tend to fan out at one end (a tendency called
heteroscedasticity), the relationship varies in strength. If not, it has constant
strength.
5. Does the pattern generalize to other cases, or is the relationship an instance
of what you see is all there is?
6. Are there plausible explanations for the pattern? Is it reasonable to conclude
that one variable causes the other? Is there a third or lurking variable that
might be causing both?
3.1 Scatterplots
107
10-03-2009 14:20
Lesson
3 de 10
A plot of variable A
against (or versus)
variable B shows A on
the y-axis and B on the
x-axis.
Example: Dormitory Populations

The plot in Display 3.3 shows, for the 50 states in the United States, the number of
people living in college dormitories versus the number of people living in cities,
in thousands. Describe the pattern in the plot.
Display 3.3
Number of people living in college dormitories

versus number of people living in cities for the
50 states in the United States. [Source: U.S. Census Bureau,
2000 Census of Population and Housing.]
Solution
1. Variables and cases. The scatterplot plots dormitory population against urban
population, in thousands, for the 50 U.S. states. Dormitory population
ranges from near 0 to a high of more than 174,000 in New York. The
urban population ranges from near 0 to about 17 million in Texas and
New York and 32 million in California.
2. Shape. While most states follow a linear trend, the three states with the largest
urban population suggest curvature in the plot because, for those states,
the number of people living in dormitories is proportionately lower than in
the smaller states. California can be considered an outlier with respect to its
urban population, which is much larger than that of other states. It is also
an outlier with respect to the overall pattern, because it lies far below the
generally linear trend.
3. Trend. The trend is positivestates with larger urban populations tend to
have larger dormitory populations, and states with smaller urban populations
tend to have smaller dormitory populations.
4. Strength. The relationship varies in strength. For the states with the smallest
urban populations, the points cluster rather closely around a line. For the
states with the largest urban populations, the points are scattered farther from
the line. Overall, the strength of the relationship is moderate.
5. Generalization. The 50 states arent a sample from a larger population of
cases, so the relationship here does not generalize to other cases. Because
both variables tend to change rather slowly, however, we can expect the
relationship in Display 3.3 to be similar to that of other years.
108
Chapter 3
10-03-2009 14:20
Lesson
4 de 10
6. Explanation. It is tempting to attribute the positive relationship to the idea

that cities attract colleges. (Just pick a large city nearby and see how many
colleges you can name that are located there.) The main reason for the
positive relationship, however, is not nearly so interesting: Both variables ar
related to a states population. The more people in a state, the more people
live in dormitories and the more people live in cities. (Theres a moral here:
Interpreting association can be tricky, in part because the two variables you s
in a plot often will be related to some lurking variable that you dont see.)

D2. Display 3.4 is derived from the data for Display 3.3 by converting
the variables to the proportion of a states population living in college
dormitories (given as the number living in dorms per 1000 state residents)
and the proportion of the states population living in cities.
a. Follow steps 16 in the previous example to describe what you see in thi
new plot.
b. When you go from totals (Display 3.3) to proportion of total population
(Display 3.4), the relationship changes from positive to negative and
becomes weaker. Give an explanation for the di erences in these
two plots.
Display 3.4
The proportion of people living in college

dormitories versus the proportion of people living
in cities for the 50 U.S. states.
Summary 3.1: Scatterplots

A scatterplot shows the relationship between two quantitative variables. Each
case is a point, with the x-coordinate equal to the value of one variable and the
y-coordinate equal to the value of the other variable.
In describing a scatterplot, be sure to cover all of the following:
cases and variables (What exactly does each point represent?)
shape (linear or curved, clusters, outliers)
3.1 Scatterplots
109
10-03-2009 14:20
Lesson
5 de 10
trend (positive, negative, or none)

strength (strong, moderate, or weak; constant or varying)
generalization (Does the pattern generalize?)
explanation (Is there an explanation for the pattern?)
When asked to compare two scatterplots, dont just describe each separately.
Be sure to describe how their shapes, trends, and strengths are similar and
how they differ.
Practice
P1. Growing kids. This table gives median heights
of boys at ages 2, 3, 4, 5, 6, and 7 yr.
a. Scatterplot. Plot height versus age; that

is, put height on the y-axis and age on
the x-axis.
b. Shape, trend, and strength. Describe
the shape, trend, and strength of the
relationship.
c. Generalization. Would you expect
these data to allow you to make good
predictions of the median height of
8-year-olds? Of 50-year-olds?
d. Explanation. It doesnt quite fit to say that
age causes height, but there is still an
underlying cause-and-effect relationship.
How would you describe it?
P2. Late planes and lost bags. A great way to cap
off a long day of travel is to have your plane
arrive late and then find that the airline has
lost your luggage. As Display 3.5 shows, some
airlines handle baggage better than others.
a. Which airline has the worst record for
mishandled baggage? For being on time?
b. Where on the plot would you find the
airline with the best on-time record and
the best mishandled-baggage rate? Which
airlines are best in both categories?
c. Determine whether this statement is

true or false, and explain your answer:
American had a mishandled-baggage
rate that was more than twice the rate of
Southwest.
Display 3.5 On-time arrivals versus mishandled

baggage. [Source: U.S. Department of
Transportation, Air Travel Consumer Report,
October 2005.]
d. Is there a positive or a negative

relationship between the on-time
percentage and the rate of mishandled
baggage? Is it strong or weak?
e. Would you expect the relationship in
this plot to generalize to some larger
population of commercial airlines?
Why or why not? Would you expect
the relationship in this plot to be roughly
the same for data from 10 years ago? For
next year?
10-03-2009 14:20
Lesson
6 de 10
Exercises
E1. For each of the lettered scatterplots in
Display 3.6, give the trend (positive or
negative), strength (strong, moderate, or
weak), and shape (linear or curved). Which
plots show varying strength?
a.
b.
c.
d.
e.
f.
g.
h.
Display 3.6 Eight scatterplots with various

distributions.
E2. For each set of cases and variables, tell

whether you expect the relationship to
be (i) positive or negative and (ii) strong,
moderate, or weak.
LaTasha Colander crosses the finish line of the

womens 100-meter dash final at the 2004 U.S.
Olympic Team Track and Field Trials.
3.1 Scatterplots
111
10-03-2009 14:20
Lesson
7 de 10
E3. Match each set of cases and variables (AD)

with the short summary (IIV) of its
scatterplot.
A.
B.
C.
D.
b. Describe the shape of the plot. Do you

see any clusters? Are there any outliers?
Is the relationship linear or curved? Is the
overall trend positive or negative? What
is the strength of the relationship?
c. Is the distribution of the percentage
of students taking the SAT I bimodal?
Explain how the scatterplot shows this.
Is the distribution of SAT I math scores
bimodal?
I. strong negative relationship, somewhat

curved
II. strong, curved positive relationship
III. moderate, roughly linear, positive
relationship
IV. moderate negative relationship
E4. SAT I math scores. In 2005, the average

SAT I math score across the United States
was 520. North Dakota students averaged
605, Illinois students averaged 606, and
students from the nearby state of Iowa
did even better, averaging 608. Why do
states from the Midwest do so well? It is
easy to jump to a false conclusion, but the
scatterplot in Display 3.7 can help you find
a reasonable explanation.
a. Estimate the percentage of students in
Iowa and in Illinois who took the SAT I.
New York had the highest percentage of
students who took the SAT I. Estimate
that percentage and the average SAT I
math score for students in that state.
112
Chapter 3
Display 3.7 Average SAT I math scores by state

versus the percentage of high school
graduates who took the exam. [Source:
College Board, www.collegeboard.com.]
d. The cases used in this plot are the 50 U.S.

states in 2005. Would you expect the
pattern to generalize to some other set of
cases? Why or why not?
e. Suggest an explanation for the trend.
(Hint: The SAT is administered from
Princeton, New Jersey. An alternative
exam, the ACT, is administered from
Iowa. Many colleges and universities in
the Midwest either prefer the ACT or at
least accept it in place of the SAT, whereas
colleges in the eastern states tend to
prefer the SAT.) Is there anything in the
data that you can use to help you decide
whether your explanation is correct?
10-03-2009 14:20
Lesson
8 de 10
E5. Each of the 51 cases plotted on the

scatterplots in Display 3.8 is a top-rated
university. The y-coordinate of a point tells
the graduation rate, and the x-coordinate
tells the value of some other quantitative
variablethe percentage of alumni who
gave that year, the student/faculty ratio, the
75th percentile of the SAT scores (math plus
critical reading) for a recent entering class,
and the percentage of incoming students
who ranked in the top 10% of their high
school graduating class.
rate? Which variable is almost useless for

predicting graduation rate?
d. Generalization. The cases in these plots
are the 51 universities that happened
to come out at the top of one particular
rating scheme. Do you think the complete
set of all U.S. universities would show
pretty much the same relationships? Why
or why not?
e. Explanation. Consider the two variables
with the strongest relationship to
graduation rates. Offer an explanation
for the strength of these particular
relationships. In what ways, if any, can
you use the data to help you decide
whether your explanation is in fact
correct?
E6. Hat size. What does hat size really measure?
A group of students investigated this
question by collecting a sample of hats.
They recorded the size of the hat and then
measured the circumference, the major
axis (the length across the opening in the
long direction), and the minor axis. (See
Display 3.9 on the next page. Hat sizes
have been changed to decimals; all other
measurements are in inches.) Is hat size
most closely related to circumference, major
axis, or minor axis? Answer this question by
making appropriate plots and describing the
patterns in those plots.
Display 3.8 Scatterplots showing the relationship

between graduation rate and four other
variables for 51 top-rated universities.
[Source: U.S. News and World Report, 2000.]
a. Compare the shapes of the four plots.

i. Which plots show a linear shape?
Which show a curved shape?
ii. Which plots show just one cluster?
Which show more than one?
iii. Which plots have outliers?
b. Compare the trends of the relationships:
Which plots show a positive trend? A
negative trend? No trend?
c. Compare the strengths of the
relationships: Which variables give more
precise predictions of the graduation
3.1 Scatterplots
113
10-03-2009 14:20
Lesson
9 de 10
Display 3.10 Age at hire versus year of hire for

the 50 employees in Westvaco
Corporations engineering department.
c. Display 3.11 shows the year of birth of

the Westvaco employees plotted against
the year they were hired. Open circles
represent employees laid off, and solid
circles represent employees kept. Does
this scatterplot suggest a reason why
older employees tended to be laid off
more frequently?
Display 3.9 Hat sizes, with circumference and axes

in inches. [Source: Roger Johnson, Carleton
College, data from student project.]
Display 3.11 Year of birth versus year of hire for

Westvaco employees.
E8. Passenger aircraft. Airplanes vary in their

E7. Westvaco, revisited. To determine
whether Westvaco discriminated by age in
size, speed, average flight length, and cost
laying off employees, you could investigate
of operation. You can probably guess that
whether it might have discriminated in
larger planes use more fuel per hour and
hiring. Display 3.10 shows the age at hire
cost more to operate than smaller planes,
plotted against the year the person was
but the shapes of the relationships are less
obvious. Display 3.12 lists data on the 33
hired.
most commonly used passenger airplanes
a. Describe the pattern in the plot, following
in the United States. The variables are the
the six-step model.
number of seats, average cargo payload in
b. Does this plot provide evidence that
tons, airborne speed in miles per hour, flight
Westvaco discriminated by age in hiring?
length in miles, fuel consumption in gallons
per hour, and operating cost per hour in
dollars.
10-03-2009 14:20
Lesson
10 de 10
Display 3.12 Data on passenger aircraft. [Source: Air Transport Association of America, 2005,
www.air-transport.org.]
a. cost per hour

i. Make scatterplots with cost per hour
on the y-axis to explore this variables
dependence on the other variables.
Report your most interesting findings.
Here are examples of some questions
you could investigate: For which

variable is the relationship to the
cost per hour strongest? Is there any
one airplane whose cost per hour, in
relation to other variables, makes it
an outlier?
3.1 Scatterplots
115
10-03-2009 14:20
Lesson
1 de 24
ii. Do your results mean that larger

planes are less efficient? Define your
own variable, and plot it against
other variables to judge the relative
efficiency of the larger planes.
b. flight length
i. Make scatterplots with length of flight
on the x-axis to explore this variables
relationship to the other variables.
Report your most interesting
findings. Here is an example of a
question you could investigate:
Which variable, cargo or number of
seats, shows a stronger relationship
to flight length? Propose a reasonable
explanation for why this should be so.
ii. Do planes with a longer flight length
tend to use less fuel per mile than
planes with a shorter flight length?
3.2
c. speed, seats, and cargo

i. Make scatterplots to explore the
relationships between the variables
speed, seats, and cargo. Report your
most interesting findings. Here are
some examples of questions you
could investigate: For which variable,
cargo or number of seats, is the
relationship to speed more obviously
curved? Explain why that should be
the case. Which plane is unusually
slow for the amount of cargo it
carries? Which plane is unusually
slow for the number of seats it has?
ii. The plot of cargo against seats has two
parts: a flat stretch on the left and a fan
on the right. Explain, in the language
of airplanes, seats, and cargo, what
each of the two patterns tells you.
Getting a Line on the Pattern

In this section, you will learn how to use a regression line to summarize the
relationship between two quantitative variables. This section deals first with
the simple situation in which all the data points lie close to a line. In practice,
however, data points are often more scattered. The second part of this section
shows how to choose a summary line when your data points form an oval cloud.
To begin, you will review the properties of a linear equation.
Lines as Summaries
Youve seen the equation of a line, y = slope x + y-intercept, so the review here
will be brief. Linear relationships have the important property that for any two
points (x1, y1) and (x2, y2) on the line, the ratio
is a constant. This ratio is the slope of the line. The rise and run are illustrated
in Display 3.13, where the slope is the ratio of the two sides of the right triangle.
This ratio is the same for any two points on the line because all the triangles
formed are similar.
23-03-2009 21:33
Lesson
2 de 24
The slope of a line is the

change in y divided by
the change in x.
How thick is a single sheet of your book? One sheet alone is too thin to
measure directly with a ruler, but you could measure the thickness of 100 sheets
together, then divide by 100. This method would give you an estimate of the
thickness but no information about how much your estimate is likely to vary from
the true thickness. The approach in the next activity lets you judge precision as
well as thickness.
Pinch 50 sheets of
paper, not up to
page 50.
Pinching Pages
What youll need: a ruler with a millimeter scale, a copy of your textbook
1. Pinch together the front cover and first
50 sheets of your textbook. Then
measure and record the thickness to
the nearest millimeter.
2. Repeat for the front cover plus 100,
150, 200, and 250 sheets.
3. Plot your data on a scatterplot, with
number of sheets on the horizontal
scale and total thickness on the vertical
scale.
4. Does the plot look linear? Should it?
Discuss why or why not, and make
your measurements again if necessary. On the plot, place a straight line
that best fits the cloud of points.
5. Find the slope and y-intercept of your line. What does the y-intercept tell
you? What does the slope tell you? What is your estimate of the thickness
of a sheet?
6. Use the information in your graph to discuss how much your estimate in
step 5 is likely to vary from the true thickness.
7. How would your line have changed if you hadnt included the front cover?
117
23-03-2009 21:33
Lesson
3 de 24
Interpreting Slope
D3. Decide what the variables y and x represent in these situations.
a. Suppose regular unleaded gasoline costs $2.60 per gallon. The number
2.60 is the slope of the line you get if you plot y versus x.
b. Suppose your car averages 30 mi/gal. The number 30 is the slope of a line
fitted to a scatterplot of y versus x.
c. A pints a pound the world around. This slogan summarizes the slope of
a line fitted to a scatterplot of y versus x for various quantities of various
kinds of liquids.
The next example illustrates how to find the equation of a line when you know
two points that fall on the line.
Example: Minimum Wage

In an effort to keep wages of hourly workers at a level that allows some possibility
of making a decent living, the United States government establishes a minimum
hourly wage rate. The scatterplot in Display 3.14 shows the minimum wage (in
dollars) for every five years from 1960 through 2005. The line on the plot is the
least squares regression line, which you will learn about later in this section.
Estimate the slope of the line. What does the slope tell you? Estimate the equation
of the line.
Display 3.14 Minimum wages at five-year intervals, 1960

through 2005.
Solution
In theory, you can find the slope from any two points on the line. Here, however,
you have to estimate the coordinates from the graph. In such cases, you usually
can produce a better estimate of the slope by choosing two points that are far
apart. For this plot, choosing the points on the line for the years 1960 and 2000
works well. Approximate points are (1960, 0.80) for (x1, y1) and (2000, 4.80) for
(x2, y2). The estimated slope is
118
Chapter 3 Relationships Between Two Quantitative Variables
23-03-2009 21:33
Lesson
4 de 24
Be sure to qualify your

interpretations of the
slope. The minimum
wage didnt go up
exactly $0.10 per year.
It tended to go up this
amount or went up
about this amount.
The slope tells you that the minimum wage increased by about $0.10 per year over
the 45-yr period 1960 through 2005.
You can write the equation of a line in terms of its slope and y-intercept.
y = slope x + y-intercept
You have the slope, but you cannot read the y-intercept from the plot. To find it,
use the slope and a point such as (1960, 0.80) to solve for the y-intercept.
0.80 = 0.10(1960) + y-intercept
y-intercept = 0.80 196 = 195.20
The equation of the line using these approximate values is
y = 0.10x 195.20
In statistics this equation usually is written with the intercept first, becoming
y = 195.20 + 0.10x
Lines as Summaries
D4. The Consumer Price Index (CPI) is a measure of the prices paid by urban
consumers for a selected group of goods and services (called a market
basket) thought to be typical of urban households. The CPI often is used to
adjust salaries, rents, and other segments of the economy. The CPI is, itself,
a statistical estimate based on a number of large-scale surveys conducted by
agencies of the federal government. Display 3.15 shows the CPI for every five
years from 1970 to 2005. Fit a line to these data by eye, and use two points
to estimate its slope. What is the annual rate of increase for the CPI over this
35-yr period? What is the equation of your line?
Display 3.15 CPI at five-year intervals, 19702005.
119
23-03-2009 21:33
Lesson
5 de 24
Using Lines for Prediction

There are two main reasons why you would want to fit a line to a set of data:
to find a summary, or model, that describes the relationship between the two
variables
to use the line to predict the value of y when you know the value of x. In
Lines are summaries and
cases where it makes sense to do this, the variable on the x-axis is called the
can be used to predict.
predictor or explanatory variable, and the variable on the y-axis is called the
predicted or response variable.
In the previous example, the equation
y = 195.20 + 0.10x
models the rise in the minimum wage for the years 1960 through 2005. Knowing
this equation enables you to make a general statement about the minimum wage
throughout these years: The minimum wage went up roughly $0.10 per year.
You might instead want to use the line to predict the minimum wage in one of
the years for which no amount is given or for years before 1960 or after 2005.
Example: Predicting the Minimum Wage

Use the equation y = 195.20 + 0.10x to predict the minimum wage in the years
2003 and 1950.
Solution
The predicted minimum wage for 2003 is
y = 195.20 + 0.10x = 195.20 + 0.10(2003) = 5.10
Assuming the linear trend continues back to earlier years, the predicted minimum
wage for 1950 is
y = 195.20 + 0.10x = 195.20 + 0.10(1950) = 0.20
The predicted minimum wage for 2003 is very close to the actual minimum
wage of $5.15 per hour. But the actual minimum wage in 1950 was $0.75 per hour,
not a negative number! As you can see, making the assumption that the linear
trend continues can be risky. This type of prediction, making a prediction when
the value of x falls outside the range of the actual data, is called extrapolation.
Interpolationmaking a prediction when the value of x falls inside the range of
the data, as does 2003is safer.
Suppose you know the value of x and use a line to predict the corresponding
value of y. You know that your prediction for y wont be exact, but you hope
that the error will be small. The prediction error is the difference between the
is read y-hat and may observed value of y and the predicted value of y, or . You usually dont know
what that error is. If you did, you wouldnt need to use the line to predict the value
be called the predicted
value or the fitted value. of y. You do, however, know the errors for the points used to construct the line.
These differences are called residuals:
residual = observed value of y predicted value of y = y
23-03-2009 21:33
Lesson
6 de 24
The residual is the

signed vertical distance
of the data point from
the line.
The geometric interpretation of the residual is shown in Display 3.16. A residual

is the signed vertical distance from an observed data point to the regression line.
The residual is positive if the point is above the line and negative if the point is
below the line.
Display 3.16 Residual y
Example: Finding Residuals

Display 3.17 shows the mean net income (after expenses and before taxes, in
thousands of dollars) for doctors who were board-certified in family practice and
working during the years 19901998 and 2001.
Display 3.17 Mean net income for family practitioners,

19902001. [Source: U.S. Census Bureau, Statistical Abstract of
the United States, 20042005.]
The equation of the fitted line is = 8300.6 + 4.2248x, where x is the year
is the income in thousands of dollars.
Graph the fitted line with the data points. What is the residual for the year
1996?
and
Solution
You can use a graphing calculator to graph a scatterplot with a summary line.
[See Calculator Note 3B.]
121
23-03-2009 21:33
Lesson
7 de 24
The actual net income value for 1996 was $139,000. Using the equation of the
fitted line, the prediction for 1996 is
= 8300.6 + 4.2248x = 8300.6 + 4.2248(1996) = 132.1008, or $132,101
You also can use your calculator to calculate a predicted value quickly. [See
Calculator Note 3C.]
To find the residual, subtract the predicted value from the observed value:
y = 139 132.1008 = 6.8992
or about $6899. The residual is positive because the observed value is higher than
the predicted value. That is, the point lies above the line.
You can use a calculator to calculate residuals for all points in a data set
simultaneously. [See Calculator Note 3D.]

D5. Test how well you understand residuals.
a. If a residual is large and negative, where is the point located with respect
to the line? Draw a diagram to illustrate. What does it mean if the
residual is 0?
b. If someone said that they had fit a line to a set of data points and all their
residuals were positive, what would you say to them?
c. Interpret the y-intercept of the regression line in the previous example.
Does this make sense?
23-03-2009 21:33
Lesson
8 de 24
D6. What do you think of the arithmetic and the reasoning in this passage from
Mark Twains Life on the Mississippi?
In the space of one hundred and seventy-six years the Lower Mississippi
has shortened itself two hundred and forty-two miles. That is an average
of a trifle over one mile and a third per year. Therefore, any calm person,
who is not blind or idiotic, can see that in the Old Olitic Silurian
Period, just a million years ago next November, the Lower Mississippi
River was upwards of one million three hundred thousand miles long,
and stuck out over the Gulf of Mexico like a fishing rod.
And by the same token any person can see that seven hundred and
forty-two years from now the Lower Mississippi will be only a mile and
three quarters long, and Cairo and New Orleans will have joined their
streets together, and be plodding comfortably along under a single mayor
and a mutual board of aldermen. There is something fascinating about
science. One gets such wholesale returns of conjecture out of such a
trifling investment of fact. [Source: James R. Osgood and Company, 1883, p. 208.]
Given that the Mississippi/Missouri river system was about 3710 mi long in
the year 2000, write an equation that Twain would say gives the length of the
river in terms of the year.
Least Squares Regression Lines

The general approach to fitting lines to data is called the method of least
squares. The method was invented about 200 years ago by Carl Friedrich
Gauss (17771855), Adrien-Marie Legendre (17521833), and Robert Adrain
(17751843), who were working independently of one another in Germany,
France, and Ireland, respectively.
The least squares regression line, also called least squares line or regression
line, for a set of data points (x, y) is the line for which the sum of squared errors
(residuals), or SSE, is as small as possible.
Example: Regression Line for the Passenger Jets

This table shows cost per hour versus number of seats for three models of
passenger jets from the data in Display 3.12 on page 115. (Some of the values
have been rounded.)
3.2 Getting a Line on the Pattern 123
23-03-2009 21:33
Lesson
9 de 24
Which of these two equations gives the least squares regression line for predicting
cost from number of seats?
= 367 + 16x
= 300 + 16x
Solution
The least squares regression line minimizes the sum of the squared errors,
SSE, so the equation with the smaller SSE must be the equation of the
regression line.
For the equation = 367 + 16x:
For the equation
= 300 + 16x:
[You can use your calculator to calculate the SSE quickly. See Calculator Note 3E.]
The first equation has the smaller SSE, so it must be the equation of the least
squares regression line. Note that for this line, except for rounding error, the sum
, is equal to 0. This is always the case for the least squares
of the residuals,
regression line, but it can be true for other lines, too.
[You can use a calculator program to visually explore the least squares regression line
and SSE. See Calculator Note 3F.]
In addition to making the sum of the squared errors as small as possible, the
least squares regression line has some other properties, given in the box on the
next page.
23-03-2009 21:33
Lesson
10 de 24
Properties of the Least Squares Regression Line

The fact that the sum of squared errors, or SSE, is as small as possible means
that, for the least squares regression line, these properties also hold:
The sum (and mean) of the residuals is 0.
The line contains the point of averages,
The standard deviation of the residuals is smaller than for any other line
that goes through the point
The line has slope b1, where
There are some appealing mathematical relationships among these properties,

which, taken together, show that the line through the point of averages
having slope b1 does, in fact, minimize the sum of the squared errors.This gives
you a way to find the equation of the least squares line:
First,
compute the slope b1 using the formula in the box. Then find the y-intercept, b0,
and solving the equation
using the point
Example: Least Squares Line for the Passenger Jets

Find the least squares line for the passenger jets data given in the previous
example.
Solution
Finding the line requires three main steps: Find the point of averages
, find
the slope b1, and use the point and slope to find the y-intercept b0.
A convenient way to organize the computations is to work from a table.
Point of averages: The point of averages

squares regression line passes through this point.
is
, and the least
23-03-2009 21:33
Lesson
11 de 24
Slope: To compute the slope, first create two new columns for deviations from
the mean, one for x . and the other for y :
Now create two more columns, one for
and the other for
The ratio of the sums of the last two columns gives the slope
y-intercept: Now that you have a point on the line,

slope, 16, you can find the y-intercept from the equation
and the
This agrees with what you found in the previous example. That is, the
equation of the least squares regression line (with rounded y-intercept) is
[You also can use your calculator to find the equation of the least squares line. See
Calculator Note 3G.]
Least Squares Regression Lines

D7. You might have wondered why statisticians dont fit a regression line by
rather
minimizing the sum of the absolute values of the residuals,
than the sum of the squares of the residuals. Here you will learn one
reason why.
23-03-2009 21:33
Lesson
12 de 24
a. Plot the points in the table and the line y = 1 + x. Explain why this is the
line that best fits these points. Compute the sum of the absolute values of
the residuals.
b. Draw another line that passes between the two points at x = 0 and also
passes between the two points at x = 2. Compute the sum of the absolute
values of the residuals for this line and compare it to your sum from
part a.
c. Draw yet another line that passes between the two points at x = 0 and
also passes between the two points at x = 2. Find the sum of the absolute
values of the residuals for this line and compare it to your sums from
parts a and b.
d. Draw a line that does not pass between the two points at x = 0. Find the
sum of the absolute values of the residuals for this line and compare it to
your sums from parts a and b.
e. Now find the least squares regression line and compute the sum of the
squared residuals. Compute the sum of the squared residuals for your
lines in parts b, c, and d. What can you conclude?
f. Find the standard deviation of the residuals for the least squares
regression line and for the lines in parts b, c, and d. What can you
conclude?
Reading Computer Output

When you are working with real data, the best way to get the least squares line
is by computer or calculator. Display 3.18 shows typical computer output for the
minimum wage data in Display 3.14 on page 118.
Display 3.18 Data Desk output giving the equation of the least
squares line for the minimum wage data.
23-03-2009 21:33
Lesson
13 de 24
You can ignore most of the output for now. You will learn how to interpret it
in Chapter 11. For the time being, focus on the first two columns in the last three
rows, which are reproduced in Display 3.19.
Display 3.19 The lower-left corner of the computer output gives

the y-intercept and slope.
The y-intercept is the coefficient in the row labeled Constant and is 196.977.
The slope is the coefficient of the predictor variable Year and is 0.100909.
The SSE for the regression line is found in the Residual row and is 0.354545.
D8. Doctors incomes. Display 3.20 shows the mean net income y of family
practitioners versus year x (from page 121), with Data Desk computer
output for the least squares line.
Display 3.20 Scatterplot of mean net income (in thousands

of dollars) of doctors board-certifi ed in family
practice, 19902001, and Data Desk output for the
regression.
23-03-2009 21:33
Lesson
14 de 24
a. What is the equation of the least squares line? Estimate the SSE from the
scatterplot in Display 3.20, and then find it in the computer output.
b. The Minitab software output for this regression is shown in Display 3.21.
How is it different from the Data Desk output?
Display 3.21 Minitab output for the regression of family

practitioners income versus year.
Summary 3.2: Getting a Line on the Pattern

For many quantitative relationships, it makes sense to use one variable, x, called
the predictor or explanatory variable, to predict values of the other variable, y,
called the predicted or response variable. When the data are roughly linear, you
can use a fitted line, called the least squares regression line, as a summary or
model that describes the relationship between the two variables. You might also
use it to predict the value of an unknown value y when you know the value of x.
Interpolationusing a fitted relationship to predict a response value when
the predictor value falls within the range of the datagenerally is much more
trustworthy than extrapolationpredicting response values based on the
assumption that a fitted relationship applies outside the range of the observed data.
Each residual from a fitted line measures the vertical distance from a data
point to the line:
residual = observed value predicted value = y y
The least squares regression line for a set of pairs (x, y) is the line for which
the sum of squared errors, or SSE, is as small as possible. For this line, these
properties hold:
The sum (and mean) of the residuals is 0.
The line contains the point of averages,
The variation in the residuals is as small as possible.
The line has slope b1, where
23-03-2009 21:33
Lesson
15 de 24
To find the equation of the regression line,

compute and .
find the slope using the formula for b1
compute the y-intercept: b0 = b1 .
The equation is
value of y.
= b0 + b1x. Remember to use a hat,
Practice
Lines as Summaries
P3. Display 3.22 shows the weight of a students
pink eraser, in grams, plotted against
the number of days into the school year.
Estimate the slope of the line drawn on the
graph. Interpret the slope in the context
of the situation. [Source: Zachs Eraser, CMC
ComMuniCator, 28 (June 2004): 28.]
a.
b.
c.
d.
, to indicate a predicted
Estimate the slope of the line.

What does the slope tell you?
Estimate the equation of the line.
Students were instructed to measure their
hand width with their fingers spread
apart as far as possible. The scatterplot
shows a smaller cloud of points below
the main one. Why do you think that
is the case? What would happen to the
regression line if those points were
removed?

P5. If you attend a university where class sizes
tend to be small, are you more likely to give
to your alumni fund after you graduate than
if you graduate from a university with large
classes? Display 3.24 shows a scatterplot of
a sample of 40 universities. Each university
Display 3.22 Weight of pink eraser.
appears as a point. The vertical coordinate,
P4. Display 3.23 shows the hand width of 383
y, tells the percentage of alumni who gave
students plotted against hand length. The line
money. Each x-coordinate tells the student/
drawn on the plot is the least squares line.
faculty ratio (number of students per faculty
member). The equation of the fitted line is
approximately = 55 2x.
a. Which is the explanatory variable and
which is the response variable?
b. Explain how you can see from the
graph that an increase of five students
per faculty member corresponds to a
decrease of about 10 percentage points in
the giving rate. Explain how you can see
this from the equation of the fitted line.
c.
Does the y-intercept have a useful
Display 3.23 Hand width and hand length, in
interpretation
in this situation?
inches, for 383 students.
23-03-2009 21:33
Lesson
16 de 24
d. Use the regression line to predict the

giving rate for a university with a student/
faculty ratio of 16. When you use the
regression line to predict the giving rate,
would you expect a rather large error or a
relatively small error in your prediction?
e. Use the plot to estimate the residual for
the university with the highest student/
faculty ratio and for the university with
the highest giving rate.
f. The university with the lowest student/
faculty ratio, 6 to 1, had a giving rate of
32%. Use the equation of the fitted line to
find the residual for that university.
g. Suppose the Alumni Association at
Piranha State University boasts a giving
rate of 80%. Without knowing the
student/faculty ratio at PSU, can you
tell whether the prediction error will be
positive or negative?
Display 3.24 Percentage of alumni giving to the

alumni fund versus the student/
faculty ratio for 40 highly rated U.S.
universities.
c. Interpret the slope and y-intercept in the

context of this situation.
d. Verify that the least squares regression
line goes through the point of averages.
e. Verify that the sum of the residuals is 0.
P7. Use the statistical functions of your
calculator to make a scatterplot, find the
regression equation for predicting percentage
on-time arrivals from mishandled baggage,
and compute residuals for the airline data
from P2 on page 110. [See Calculator Notes
3A, 3G, and 3D.] The data values are given in
Display 3.25.
Display 3.25 Comparison, by airline, of mishandled

baggage and on-time arrival rate.
[Source: U.S. Department of Transportation, Air
Travel Consumer Report, October 2005.]
Least Squares Regression Line

P6. The fat and calorie contents of 5 oz of three
kinds of pizza are represented by the data
points (9, 305), (11, 309), and (13, 316).
a. Plot the points.
b. Compute the equation of the least squares
regression line by hand, and draw the line
on your plot.
23-03-2009 21:33
Lesson
17 de 24

P8. The JMP-IN computer output in Display
3.26 is for the pizza data in P6. Does it give
the same results that you computed by hand?
Where in the output is the SSE found?
Display 3.26 JMP-IN computer output for pizza data.
Exercises
E9. Display 3.27 shows cost in dollars per hour
versus number of seats for three aircraft
models. Five lines, labeled AE, are shown
on the plot. Their equations, listed below, are
labeled IV.
a. Match each line (AE) with its equation
(IV).
I. cost = 290 + 15.8 seats
II. cost = 400 + 15.8 seats
III. cost = 1000 + 15.8 seats
IV. cost = 370 + 25 seats
V. cost = 900 + 10 seats
b. Match each line (AE) with the
appropriate verbal description (IV):
I. This line overestimates cost.
II. This line underestimates cost.
III. This line overestimates cost for the
smallest plane and underestimates
cost for the largest plane.
IV. This line underestimates cost for the

smallest plane and overestimates cost
for the largest plane.
V. On balance, this line gives a better fit
than the other lines.
Display 3.27 Cost in dollars per hour versus number

of seats for three aircraft models.
23-03-2009 21:33
Lesson
18 de 24
E10. Examine the scatterplot in Display 3.28.
c. Consider the possible summary lines in

Display 3.29.
i. Which line gives predicted values
for calorie content that are too high?
How can you tell this from the plot?
ii. Which line tends to give predicted
calorie values that are too low?
iii. Which line tends to overestimate
calorie content for lower-fat pizzas
and underestimate calorie content for
higher-fat pizzas?
iv. Which line has the opposite problem,
underestimating calorie content
Display 3.28 Calories versus fat, per 5-oz serving,
when fat content is lower and
for seven kinds of pizza. [Source: Consumer
overestimating calorie content when
Reports, July 2003.]
fat content is higher?
a. Which two kinds of pizza in Display 3.28
v. Which line fits the data best overall?
have the fewest calories? Which two have
E11. Heights of boys. The scatterplot in Display
the least fat? Which region of the graph
3.30 shows the median height, in inches, for
has the pizzas with the most fat?
boys ages 2 through 14 years.
b. Display 3.29 shows the data again, with
five possible summary lines. Match each
equation (IV) with the appropriate
line (AE).
I. calories 70 15 fat
II. calories 10 25 fat
III. calories 150 15 fat
IV. calories 110 15 fat
V. calories 170 10 fat
Display 3.30 Median height versus age for boys.

[Source: National Health and Nutrition
Examination Survey (NHANES), 2002,
www.cdc.gov.]
Display 3.29 Five possible fitted lines for the

pizza data.
a. Estimate the slope of the line that

summarizes the relationship between age
and median height.
b. Explain the meaning of the slope with
respect to boys and their median height.
c. Write the equation of the line using the
slope from part a and a point on the line.
d. Interpret the y-intercept. Does the
interpretation make sense in this context?
23-03-2009 21:33
Lesson
19 de 24
E12. Pizza again. Display 3.31 shows the calorie

and fat content of 5 oz of various kinds of
pizza.
Display 3.32 Reaction distance at various speeds.
Display 3.31 Calories and fat content per 5-oz

serving, for seven kinds of pizza. [Source:
Consumer Reports, January 2002.]
a. Use the line on the scatterplot to predict

the calorie content of a pizza with 10.5 g
of fat. Often use the line to predict the
calorie content of a pizza with 15 g of fat.
b. Use the two predictions in part a to
estimate the slope of the line. Write the
equation of the line using this slope and a
point on the line.
c. There are 9 calories in a gram of fat. How
is your estimated slope related to this
number?
E13. Stopping on a dime? In an emergency, the
typical driver requires about 0.75 second to
get his or her foot onto the brake pedal. The
distance the car travels during this reaction
time is called the reaction distance. Display
3.32 shows the reaction distances for cars
traveling at various speeds.
a. Plot reaction distance versus speed, with

speed on the horizontal axis. Describe the
shape of the plot.
b. What should the y-intercept be?
c. Find the slope of the line of best fit
by calculating the change in y per
unit change in x. What does the slope
represent in this situation?
d. Write the equation of the line that fits
these data.
e. Use the equation of the line in part d to
predict the reaction distance for a car
traveling at a speed of 55 mi/h and at
75 mi/h.
f. How would the equation change if it
actually took 1 second, instead of 0.75
second, for drivers to react?
E14. The scatterplot in Display 3.33 shows
operating cost (in dollars per hour) versus
fuel consumption (in gallons per hour) for
a sample of commercial aircraft.
Display 3.33 Operating cost versus fuel

consumption for commercial aircraft.
23-03-2009 21:33
Lesson
20 de 24
a. Which is the explanatory variable and

which is the response variable?
b. Estimate the slope of the regression line
from the graph, and interpret it in the
context of this situation.
c. The y-intercept is 470. Does this value
have a reasonable interpretation in this
situation?
d. Use the line to predict the cost per hour for
a plane that consumes 1500 gal/h of fuel.
E15. Arsenic is a potent poison sometimes found in
groundwater. Long-term exposure to arsenic
in drinking water can cause cancer. How
much arsenic a person has absorbed can be
measured from a toenail clipping. The plot in
Display 3.34 shows the arsenic concentrations
in the toenails of 21 people who used water
from their private wells plotted against the
arsenic concentration in their well water. Both
measurements are in parts per million.
in drinking water should be less than

0.01 mg/L. (1 mg/L = 1 ppm.) Is this
standard exceeded in any of these wells?
[Source: www.who.int.]
E16. More pizza. Refer to the pizza data in E12.

a. The least squares residuals for the pizza
data are, in order from smallest to largest,
40.58, 17.66, 15.95, 1.03, 14.28,
26.44, and 34.50. Match each residual
with its pizza.
b. What does the residual for Pizza Huts
Pan pizza tell you about the pizzas
number of calories versus fat content?
c. For Pizza Huts Hand Tossed and
Dominos Deep Dish, are the residuals
positive or negative? How can you tell
this from the scatterplot in Display 3.31?
E17. The level of air pollution is indicated by a
measure called the air quality index (AQI).
An AQI greater than 100 means the air
quality is unhealthy for sensitive groups such
as children. The table and plot in Display
3.35 show the number of days in Detroit that
the AQI was greater than 100 for the years
2001, 2002, and 2003.
Display 3.34 Arsenic concentrations. [Source: M. R.

Karagas et al. Toenail Samples as an Indicator
of Drinking Water Arsenic Exposure, Cancer
Epidemiology, Biomarkers and Prevention 5 (1996):
84952.]
a. What is the predictor variable, and what

is the response variable?
b. Describe the relationship.
c. Estimate the residual for the person with
the highest concentration of arsenic in
the well water.
d. Find the person on the plot with
the largest residual. What was the
concentration of arsenic in that persons
toenails?
e. The World Health Organization has set a
standard that the concentration of arsenic
Display 3.35 Air quality index for 20012003.

[Source: U.S. Environmental Protection Agency,
www.epa.gov.]
a. By hand, compute the equation of the

least squares line.
23-03-2009 21:33
Lesson
21 de 24
b. Interpret the slope in the context of this

situation.
c. Which year has the largest residual? What
is this residual?
d. Compute the SSE for this line.
e. Verify that the sum of the residuals is 0.
f. Find the SSE for the line that has the
same slope as the least squares line but
passes through the point for 2002. Is
this SSE larger or smaller than the SSE
for the least squares line? According to
the least squares approach, which line
fits better?
g. Find the slope of the line that passes
through the points for 2001 and 2003.
Then find the fitted value for 2002.
Finally, find all three residuals and the
Display 3.36 Median height for girls ages 214.
value of the SSE for this line.
b Judging from the plot, is the residual
h. The least squares line doesnt pass
for 11-year-olds positive or negative?
through any of the points, and yet
Compute this residual to check your
judging by the SSE that line fits better
answer.
than the one in part g. Do you agree that
c. Verify that the line contains the point of
the least squares line fits better than the
averages,
lines in parts f and g? Explain why or
d. How does the regression line for girls
why not.
compare to the line for boys in E11?
E18. Even more pizza. Refer again to the table and
E20. Sum of residuals. In this exercise, you will
scatterplot in Display 3.31 on page 134.
show that the sum of the residuals is equal
a. By hand, compute the equation of the
to 0 if and only if the regression line passes
least squares regression line for using fat
through the point of averages,
to predict calories. How close was your
a. Show that for a horizontal line the sum
estimate of the equation in E12?
of the residuals will be 0 if and only if
b. Which of these values must be the SSE for
the line passes through the point of
this regression? Explain your answer.
averages.
0 29.3 861.4 4307
b. Show that no matter what the slope of the
E19. Heights of girls. Display 3.36 gives the
line is, the sum of the residuals will be 0
median height in inches for girls ages 214.
if and only if the line passes through the
a. Practice using your calculator by making
point of averages.
a scatterplot, finding the equation of
c. Why isnt it good enough to define the
the least squares line for median height
regression line as the line that makes the
versus age, and graphing the equation on
sum of the residuals equal 0?
the plot.
23-03-2009 21:33
Lesson
22 de 24
E21. Height versus age. Display 3.37 shows a

standard computer printout for the median
height versus age data of E11.
Display 3.37 Computer output of median height

versus age data.
a. Write the equation of the regression line.

How does it compare to your estimate of
the equation in E11?
b. What is the SSE for this least squares line?
Does its value seem reasonable given the
scatterplot in Display 3.30 on page 133?
E22. Part of a printout for the percentage of
alumni who give to their colleges versus the
student/faculty ratio is shown in Display
3.38. (These are the data in the scatterplot
shown in Display 3.24 on page 131.)
Display 3.38 Computer output: regression analysis

of percentage giving to alumni fund
versus student/faculty ratio.
a. What equation is given in the printout for

the least squares regression line?
b. Examine the table of unusual
observations. What is the student/faculty
ratio at the college with the largest
residual (in absolute value)? Find this
college in Display 3.24 on page 131.
c. Verify that the fit and the value of the
largest residual were computed correctly.
d. Locate the SSE on the printout. Why is
this value so large?
E23. For the least squares regression line you
found in E19, calculate the residuals for
girls ages 2, 8, and 14. What does this
suggest about the pattern of growth beyond
what is summarized in the equation of the
regression line?
E24. More about slope.
a. You and three friends, one right after
the other, each buy the same kind of
gas at the same pump. Then you make a
scatterplot of your data, with one point
per person, plotting the number of
gallons on the x-axis and the total price
paid on the y-axis. Will all four points lie
on the same line? Explain.
b. You and the same three friends each
drive 80 mi but at different average
speeds. Afterward, you plot your data
twice, first as a set of four points with
coordinates average speed, x, and elapsed
time, y, and then as a set of points with
coordinates average speed, x, and y*
1
defined as _________
. Which plot will give
elapsed time
a straight line? Explain your reasoning.
Will the other plot be a curve opening up,
a curve opening down, or neither?
E25. The data set in Display 3.39 is the pizza data
of E12 augmented by other brands of cheese
pizza typically sold in supermarkets.
a. Plot calories versus fat. Does there appear
to be a linear association between calories
and fat? If so, fit a least squares line to the
data, and interpret the slope of the line.
23-03-2009 21:33
Lesson
23 de 24
b. Plot fat versus cost. Does there appear to

be a linear association between cost and
fat? If so, fit a least squares line to the
data, and interpret the slope of the line.
c. Plot calories versus cost. Does there
appear to be a linear association between
cost and calories?
d. Write a summary of your findings.
E26. Poverty. What variables are most closely

associated with poverty? Display 3.40
provides information on population
characteristics of the 50 U.S. states plus
the District of Columbia. Each variable
is measured as a percentage of the states
population, as described here:
Percentage living in metropolitan areas
Percentage white
Percentage of adults who have graduated
from high school
Percentage of families with incomes
below the poverty line
Percentage of families headed by a single
parent
Construct scatterplots to determine which
variables are most strongly associated with
poverty.
Write a letter to your representative in
Congress about poverty in America, relying
only on what you find in these data. Point
out the variables that appear to be most
strongly associated with poverty and those
that appear to have little or no association
with poverty.
Display 3.39 Food values and cost per 5-oz serving

of pizza. [Source: Consumer Reports,
January 2002.]
23-03-2009 21:33
Lesson
24 de 24
Display 3.40 Characteristics of state populations, as percentage of population.

[Source: U.S. Census Bureau, www.census.gov.]
23-03-2009 21:33
Lesson
1 de 22
3.3
The correlation
coefficient measures the
amount of variation from
the regression line.
Correlation: The Strength of a Linear Trend

Some of the linear relationships youve seen in this chapter have been extremely
strong, with points packed tightly around the regression line. Other linear
relationships have been quite weakalthough there was a general linear trend, a
lot of variation was present in the values of y associated with a given value of x.
Still other linear relationships have been in between.
In this section, youll learn how to measure the strength of a linear
relationship numerically by using the correlation coefficient, r (which, from this
point on, will be referred to simply as the correlation ), where
Just as
in the last section, youll start by working intuitively and visually and then move
on to a computational approach.
Estimating the Correlation

Examine the scatterplots and their correlations in Display 3.41. To get a rough
idea of the size of a correlation, it is helpful to sketch an ellipse around the cloud
of points in the scatterplot. If the ellipse has points scattered throughout and the
points appear to follow a linear trend, then the correlation is a reasonable measure
of the strength of the association. If the ellipse slants upward as you go from left to
right, the correlation is positive. If the ellipse slants downward as you go from le
to right, the correlation is negative. If the ellipse is fat, the correlation is weak and
the absolute value of r is close to 0. If the ellipse is skinny, the correlation is strong
and the absolute value of r is close to 1.
Display 3.41 Scatterplots with ellipses and their correlations.

[Source: George Cobb, Electronic Companion to Statistics (Cogito
Learning Media, Inc., 1997), p. 114.]
In Activity 3.3a, you will learn more about correlation and youll practice
finding the value of r using your calculator.
22-03-2009 20:55
Lesson
2 de 22
Was Leonardo Correct?

What youll need: a measuring tape, yardstick, or meterstick
Leonardo da Vinci was a scientist and an artist who combined these skills to
draft extensive instructions for other artists on how to proportion the human
body in painting and sculpture. Three of Leonardos rules were
height is equal to the span of the outstretched arms
kneeling height is three-fourths of the standing height
the length of the hand is one-ninth of the height
1. Work with a partner to measure your height, kneeling height, arm span,
and hand length. Combine your data with the rest of your class.
2. Check Leonardos three rules visually by plotting the data on three scatterplots.
3. For the plots that have a linear trend, use your calculator to find the
equation of the regression line and the value of r (the correlation). [See
Calculator Notes 3G and 3H.]
4. Interpret the slopes of the regression lines. Interpret the correlations.

5. Do the three relationships described by Leonardo appear to hold? Do they
hold strongly?

D9. Match each of the four scatterplots with its correlation, choosing from the
values 0.783, 0.783, 0.908, and 0.999.
a. year of birth versus year of hire (Display 3.1 on page 106)
b. age at layoffs versus year of hire (Display 3.2 on page 106)
c. median height of boys versus age (Display 3.30 on page 133)
d. calories in pizza versus fat (Display 3.31 on page 134)
D10. Four relationships are described here.
I. For a random sample of students from the senior class, x represents
the day of the month of the persons birthday and y represents the
cost of the persons most recent haircut.
II. For a random collection of U.S. coins, x represents the diameter and
y represents the circumference.
III. For a random sample of bags of white socks, x represents the
number of socks and y represents the price per bag.
IV. For a random sample of bags of white socks, x represents the
number of socks and y represents the price per sock.
a. Which of these relationships have a positive correlation, and which a
negative correlation? Which has the strongest relationship? The weakest?
b. For each of the four relationships, discuss the connection between your
ability to precisely predict a value of y for a given value of x and the
strength of the correlation.
141
22-03-2009 20:55
Lesson
3 de 22
A Formula for the Correlation, r

A formula for the correlation, r, follows. It looks impressive, but the basic idea is
simpleyou convert x and y to standardized values (z-scores), and then find their
average product (dividing by n 1).
You can think of the
correlation, r, as the
average product of the
z-scores.
In this formula, sx is the standard deviation of the xs, and sy is the standard
deviation of the ys. Remember that the z-score tells you how many standard
deviations the value lies above or below the mean.
Example: Computing r for the Airline Data

Compute the correlation for the relationship between the number of mishandled
bags per thousand passengers and the percentage of on-time arrivals for the
airline data in Display 3.25 on page 131.
Solution
When you are on a desert island or taking a test where you must compute r by hand,
it is easiest to organize your work as in Display 3.42. First, compute the average
values of x and y, and .. Then compute their standard deviations, sx and sy .
Display 3.42 Calculations for the correlation for airline data.
22-03-2009 20:55
Lesson
4 de 22
Then convert each value of x and y to standard units, or z-scores. The correlation,
r, is the average product of the z-scoresthe sum of the products in the last
column divided by n 1.
[You can use a calculator to find the value of r. See Calculator Note 3H.]
A way to visualize the computations in the example is to look at the four

quadrants formed on the scatterplot by dividing it vertically at the mean value of
x and horizontally at the mean value of y. Such a plot is shown in Display 3.43.
Points in Quadrant I, such as the point for US Airways, have positive z-scores for
both x and y, so their product contributes a positive amount to the calculation
of the correlation. Points in Quadrant III, such as the point for Northwest, have
negative z-scores for both xand y, so their product also contributes a positive
amount to the calculation of the correlation. Points in Quadrants II and IV, such
as America West and Delta, contribute negative amounts to the calculation of the
correlation because one z-score is positive and the other is negative.
The points in Quadrants
II and IV have negative
products zx zy.
The correlation, r, is a
quantity without units.
Display 3.43 Scatterplot divided into quadrants at
Because r is the average of the products zx zy and z-scores have no units,

r has no units. In fact, r does not depend on the units of measurement in the
original data. In Activity 3.3a, if you measure arm spans and heights in inches and
your friend measures them in centimeters, you will both compute the same value
for the correlation.
143
22-03-2009 20:55
Lesson
5 de 22

D11. Look at Display 3.42, showing the calculations for the correlation, r.
a. Confirm the calculations for the first row, America West.
b. Which point makes the largest contribution to the correlation? Where is
this point on the scatterplot?
c. Which point makes the smallest contribution to the correlation? Where
is this point on the scatterplot?
D12. Understanding r.
a. Explain in your own words what the correlation measures.
b. Explain in your own words why r has no units.
c. When computing the correlation between two variables, does it matter
which variable you select as y and which you select as x? Explain.
a. What can you say about r if there are many points in Quadrants I and III
and few in Quadrants II and IV?
b. What can you say about r if there are many points in Quadrants II and
IV and few in Quadrants I and III?
c. What can you say about r if points are scattered randomly in all four
quadrants?
Correlation and the Appropriateness

of a Linear Model
It is tempting to believe that a high correlation (either positive or negative) is
evidence that a linear model is appropriate for your data. Alas, the real world
is not so simple. For example, Display 3.44 shows the number of blogs
(Web-based periodic postings of a persons thoughts) for the first few years
after 2003, when bloggings popularity took off. The growth is exponential, yet
the (linear) correlation is very strong, r = 0.91. The points do cluster fairly
tightly about the linear regression line, but that line is not the best model for
the data. The plot on the right shows the points and the graph of the best-fitting
exponential equation, y = 0.353 1.140x. (You will learn how to fit exponential
equations to data in Section 3.5.)
22-03-2009 20:55
Lesson
6 de 22
Display 3.44 The number of blogs, in millions, versus the

number of months after March 2003. [Source: State of
the Blogosphere, October 2005, Part 1: Blogosphere Growth, posted
by Dave Sifry, October 17, 2005. Technorati News, www.technorati
.com. Table numbers estimated from graph.]
The moral: Always
plot your data before
computing summary
statistics!
The quiz scores for 22 students in Display 3.45, on the other hand, have a
correlation, r, of only 0.48. There is quite a bit of scatter, partly because the quizzes
covered very different topics. Quiz 2 covered exponential growth, and Quiz 3
covered probability. In spite of the scatter, a line is the most appropriate model
because there is no curvature in the pattern of data points.
Display 3.45 Scores on two consecutive 30-point quizzes.
Correlation and the Appropriateness of a Linear Model

D14. When the correlation is small in absolute value, what does it mean for the
prediction error? Why would anyone want to fit a line to data in a case in
which the correlation is small, as in Display 3.45 (quiz scores)?
D15. Provide a real-life scenario involving two variables for each situation.
Assume r is positive in each case.
a. r is small and you do not want to fit a line.
b. r is small and you do want to fit a line.
c. r is large and you do not want to fit a line.
d. r is large and you do want to fit a line.
3.3 Correlation: The Strength of a Linear Trend 145
22-03-2009 20:55
Lesson
7 de 22
D16. Its common in situations similar to the growth in blogs for the numbers
to increase exponentially for the first few years. What do you think would
happen if the table in Display 3.44 were continued to include numbers for
months up to the current year?
The Relationship Between the Correlation

and the Slope
By now you might have observed that the slope of the regression line, b1, and the
correlation, r, always have the same sign. But these have a more specific relationship.
The slope varies

directly as the
correlation.
Finding the Slope from the Correlation and the SDs

The slope of a least squares regression line, b1, and the correlation, r, are related
by the equation
where sx is the standard deviation of the xs and sy is the standard deviation of

the ys. This means that if you standardize the data so that sx = 1 and sy = 1,
then the slope of the regression line is equal to the correlation.
Example: Critical Reading and Math SAT Scores

In 2005, the mean critical reading score for all SAT I test takers was 508, with a
standard deviation of 113. For math scores, the mean was 520, with a standard
deviation of 115. The correlation between the two scores was not given but is
known to be quite high. If you can estimate this correlation as, say, 0.7, you can
find the equation of the regression line and use it to estimate the math score from
a students critical reading score. [Source: The College Board, 2005 College Bound Seniors:
A Profile of SAT Program Test Takers.]
Solution
The formula gives an estimate of the slope of
To find the y-intercept, use the fact that the point ( ,

regression line:
) = (508, 520) is on the
520 = 0.71 (508) + y-intercept

y-intercept = 159.32
The equation is
= 0.71x + 159.32.
22-03-2009 20:55
Lesson
8 de 22
The Relationship Between the Correlation and the Slope

D17. Find the equation of the regression line for predicting an SAT I critical
reading score given the students SAT I math score.
Correlation Does Not Imply Causation

In a sample of elementary school students, there is a strong positive relationship
between shoe size and scores on a standardized test of ability to do arithmetic.
Does this mean that studying arithmetic makes your feet bigger? No. Shoe size
and skill at arithmetic are related to each other because both increase as a child
gets older. Age is an example of a lurking variable.
Beware the Lurking Variable
A lurking variable is a variable that you didnt include in your analysis but
that might explain the relationship between the variables you did include.
That is, when variables x and y are correlated, it might be because both are
consequences of a third variable, z, that is lurking in the background.
Two variables might
be highly correlated
without one causing
the other.
Even if you cant identify a lurking variable, you should be careful to avoid
jumping to a conclusion about cause and effect when you observe a strong
relationship. The value of r does not tell you anything about why two variables
are related. The statement Correlation does not imply causation can help you
remember this. To conclude that one thing causes another, you need data from a
randomized experiment, as youll learn in the next chapter.
D18. Display 3.43 on page 143 shows a negative association between the
percentage of on-time arrivals and the number of mishandled bags per
thousand passengers. Discuss whether you think one of these variables
might cause the other, or whether a lurking variable might account for both.
D19. For the sample of 50 top-rated universities in E5 on page 113, theres a
very strong positive relationship between acceptance rate (percentage of
applicants who are offered admission) and SAT scores (the 75th percentile
for an entering class). Explain why these two variables have such a strong
relationship. Does one cause the other? If not, how might you account for
the strong relationship?
D20. People who argue about politics and public policy often point to relationships
between quantitative variables and then offer a cause-and-effect explanation
to support their points of view. For each of these relationships, first give a
possible explanation by assuming a causal relationship and then give
another possible explanation based on a lurking variable.
a. Faculty positions in academic subjects with a higher percentage of
male faculty tend to pay higher salaries. (For example, engineering and
geology have high percentages of male faculty and high average salaries;
22-03-2009 20:55
Lesson
9 de 22
journalism, music, and social work have much lower percentages of male
faculty and much lower academic salaries.)
b. States with larger reported numbers of hate groups tend to have more
people on death row. (Here, also tell how you could adjust for the lurking
variable to uncover a more informative relationship.)
c. States with higher reported rates of gun ownership tend to have lower
reported rates of violent crime.
Interpreting r2
You might have noticed that computer outputs for regression analysis, like that in
Display 3.18 on page 127, give the value of R-squared, or r2, rather than the value
of r. The student in this discussion will show you how to think about r2 as the
fraction of the variation in the values of y that you can eliminate by taking x into
account.
Alexis: Ive invented another way to measure the strength of the
relationship between x and y. Its based on the idea that the
less variation there is from the linear trend, the stronger the
correlation.
Statistician: How does it work?
Alexis: Let me ask the questions for a change. What is the best way to
predict the values of y?
Statistician: Id fit a least squares line and use the equation y = b0 + b1x.
Alexis: And how would you measure your total error?
Statistician: Well, Id use the sum of the squares of the residuals, just like we
did in the last section:
Alexis: Thats what I hoped you would say.

Statistician: We consultants try to be helpful.
Alexis: Dont get cocky. Im about to change the rules. Pretend that you
cant use the information about x. You dont have the x-values, and
you want a single fitted value for y. What value would you choose?
Statistician: Id use
Alexis: And could you again use the sum of the squared errors to measure
your total error?
Statistician: Sure, its almost like the standard deviation. Id find the sum of the
squares of the deviations from the mean, which Ill call the total
sum of squared error, SST:
22-03-2009 20:55
Lesson
10 de 22
Okay, now for my new way to measure the strength of the

relationship. If you have a strong relationship, the SSE will be
small compared to the SST, right?
Statistician: Right.
Alexis:
Alexis:
On the other hand, if you have a weak relationship, x isnt much

use for predicting y, and the SSE will be almost as big as the SST.
Right?
Statistician: I see where youre going with this. You can use a ratio to measure
the strength of the relationship.
Alexis:
Exactly. Except now I have a problem. My ratio is near 0 when the

relationship is strong, and its near 1 when the relationship is weak.
Thats backward! Oh, I see how to fix it. Just subtract my ratio
from 1.
Statistician: Good. Your new ratio is near 1 when the relationship is strong and
near 0 when the relationship is weak. Your old ratio, SSE/SST, gave
the proportion of error still there after the regression, so your new
ratio . . .
Alexis:
I can handle it from here. SST is the total error I started with.
SST minus SSE is the amount of error I get rid of by using the
relationship of y with x. So my new ratio is the proportion of error
I eliminate by using the regression.
Statistician: Right!
Alexis:
But now I have two measures of strengththe correlation, r, and

my new ratio. Which one should I use?
Statistician: Lucky for youwith a little algebra, they turn out to be equivalent.
Alexis: Cool!
Statistician: We statisticians call r2 the coeffcient of determination. It tells us
the proportion of variation in the ys that is explained by x.
Alexis: I like it. Anythings better than those z-scores!
Predicting Pizzas
Display 3.46 compares two sets of predictions of the calorie content in the seven
kinds of pizza from E12 on page 134.
149
22-03-2009 20:55
Lesson
11 de 22
Display 3.46 Two ways of predicting calorie content from fat

content for seven kinds of pizza.
If you had to pick a single number as your predicted calorie content, you
might choose the mean, 307.14 calories per serving. The fourth column in the
table and the plot in Display 3.47 show the resulting errors.
Display 3.47 Squared deviations around the mean of y.
If you use the regression equation, calories = 112 + 14.9 fat, to predict
calories, the resulting errors are much smaller in most cases and are given in the
second to last column of Display 3.46 and shown on the plot in Display 3.48.
Display 3.48 Squared deviations around the least squares line.
22-03-2009 20:55
Lesson
12 de 22
The coeffcient of
determination tells the
proportion of the total
variation in y that can be
explained by x.
Using Alexiss formula,
About 82% of the variation in calories among these brands of pizza can be
attributed to fat content.
Taking the square root gives r = 0.908.
Interpreting r2
D21. The scatterplot in Display 3.49 shows IQ plotted against head circumference,
in centimeters, for a sample of 20 people. The mean IQ was 101, and the
mean head circumference was 56.125 cm. The correlation is 0.138.
Display 3.49 [Source: M. J. Tramo, W. C. Loftus, R. L. Green, T. A. Stukel, J. B.

Weaver, and M. S. Gazzaniga, Brain Size, Head Size, and IQ in
Monozygotic Twins, Neurology 50 (1998): 124652.]
a. If you knew nothing about any possible relationship between head

circumference and IQ, what IQ would you predict for a person with a
head circumference of 54 cm?
b. The regression equation is IQ = 0.997 head circumference + 45. By
about how much does this equation predict IQ will change with a
1-cm increase in head circumference?
c. What IQ does this equation predict for a person with a head
circumference of 54 cm? How much faith do you have in this prediction?
d. How much of the variability in IQ is accounted for by the regression?
Does the regression equation help you predict IQ in any practical sense?
e. If more people were added to the plot, how do you think the regression
equation and correlation would change?
D22. Use Alexiss formula for r2 to explain why
D23. In a study of the effect of temperature on household heating bills, an
investigator said, Our research shows that about 70% of the variability in
the number of heating units used by a particular house over the years can be
explained by outside temperature. Explain what the investigator might have
meant by this statement.
22-03-2009 20:55
Lesson
13 de 22
Regression Toward the Mean
The regression line is a

line of means.
Regression toward the

mean is another term
for regression effect.
Display 3.50 shows a hypothetical data set with the height of younger sisters
plotted against the height of their older sisters. There is a moderate positive
association: r = 0.337. For both younger and older sisters, the mean height is
65 in. and the standard deviation is 2.5 in. The line drawn on the first plot,
y = x, indicates the location of points representing the same height for both
sisters. If you rotate your book and sight down the line, you can see that the
points are scattered symmetrically about it.
In the second plot, look at the vertical strip for the older sisters with heights
between 62 in. and 63 in. The X is at the mean height of the younger sisters with
older sisters in this height range. It falls at about 64 in., not between 62 in. and
63 in. as you would expect. Looking at the vertical strip on the right, the mean
height of younger sisters with older sisters between 68 in. and 69 in. is only
about 66 in. If you were to use the line y = x to predict the height of the younger
sister, you would tend to predict a height that is too small if the older sister is
shorter than average and a height that is too large if the older sister is taller than
average.
The flatter line through the third scatterplot in Display 3.50 is the least
squares regression line. Notice that this line gets as close as it can to the center
of each vertical strip. Thus, the least squares line is sometimes called the line of
means. The predicted value of y at a given value of x, using the regression line
as the model, is the estimated mean of all responses that can be produced at that
particular value of x.
Notice that the regression line has a smaller slope than the major axis ( y = x )
of the ellipse. This means that the predicted values are closer to the mean than
you might expect, which will always be the case for positively correlated data
following a linear trend. The difference between these two lines is sometimes
called the regression effect. If the correlation is near +1 or 1, the two lines will
be nearly on top of each other and the regression effect will be minimal. For a
moderate correlation such as that for the sisters heights, the regression effect
will be quite large.
The regression effect was first noticed by British scientist Francis Galton
around 1877. Galton noticed that the largest sweet-pea seeds tended to produce
daughter seeds that were large but smaller than their parent. The smallest
sweet-pea seeds tended to produce daughter seeds that were also small but larger
than their parent. There was, in Galtons words, a regression toward the mean.
This is the origin of the term regression line. [Source: D. W. Forrest, Francis Galton: The Life and
Work of a Victorian Genius (Taplinger, 1974).]
The regression effect is with us in everyday life whenever some element of

chance is involved in a persons score. For example, athletes are said to experience
a sophomore slump. That is, athletes who have the best rookie seasons do not
tend to be the same athletes who have the best second year. The top students on
the second exam in your class probably did not do as well, relative to the rest of
the class, on the first exam. The children of extremely tall or short parents do not
tend to be as extreme in height as their parents. There does, indeed, seem to be a
phenomenon at work that pulls us back toward the average. As Galton noticed,
this prevents the spread in human height, for example, from increasing. Look for
this effect as you work on regression analyses of data.
22-03-2009 20:55
Lesson
14 de 22
Display 3.50
Scatterplots showing the regression effect.
153
22-03-2009 20:55
Lesson
15 de 22

D24. Why is the regression line sometimes called the line of means?
D25. The equation of the regression line for the scatterplot in Display 3.50
is y = 43.102 + 0.337x. Interpret the slope of this line in the context
of the situation and compare it to the interpretation of the slope of the
line y = x.
Summary 3.3: Correlation

In your study of normal distributions in Chapter 2, you used the mean to tell
the center and then used the standard deviation as the overall measure of how
much the values deviated from that center. For well-behaved quantitative
relationshipsthat is, those whose scatterplots look ellipticalyou use the
regression line as the center and then measure the overall amount of variation
from the line using the correlation, r. You can think of the correlation, r, as the
average product of the z-scores.
Geometrically, the correlation measures how tightly packed the points of the
scatterplot are about the regression line.
The correlation has no units and ranges from 1 to +1. It is unchanged if you
interchange x and y or if you make a linear change of scale in x or y, such as
from feet to inches or from pounds to kilograms.
In assessing correlation, begin by making a scatterplot and then follow these
steps:
1. Shape: Is the plot linear, shaped roughly like an elliptical cloud, rather
than curved, fan-shaped, or formed of separate clusters? If so, draw
an ellipse to enclose the cloud of points. The data should be spread
throughout the ellipse; otherwise, the pattern might not be linear or
might have unusual features that require special handling. You should
not calculate the correlation for patterns that are not linear.
2. Trend: If your ellipse tilts upward to the right, the correlation is
positive; if it tilts downward to the right, the correlation is negative. The
relationship between the correlation and the slope, b1, of the regression
line is given by
3. Strength: If your ellipse is almost a circle or is horizontal, the relationship

is weak and the correlation is near zero. If your ellipse is so thin that it
looks like a line, the relationship is very strong and the correlation is
near +1 or 1.
22-03-2009 20:55
Lesson
16 de 22
Correlation is not the same as causation. Two variables may be highly

correlated without one having any causal relationship with the other. The
value of r tells nothing about why x and y are related. In particular, a strong
relationship between x and y might be due to a lurking variable.
You can interpret the value r2 as the proportion of the total variation in y that
can be accounted for by using x in the prediction model:
The regression effect (or regression toward the mean) is the tendency of
y-values to be closer to their mean than you might expect. That is, the regression
line is flatter than the major axis of the ellipse surrounding the data.
Practice
P9. By comparing to the plots in Display 3.41 on
page 140, match each of the five scatterplots
in Display 3.51 with its correlation, choosing
from 0.95, 0.5, 0, 0.5, and 0.95.
P10. The table in E12 ( Display 3.31 on page

134 ) gives the amount of fat and number of
calories in various pizzas.
a. Guess a value for the correlation, r.
b. Calculate r using your calculator.
a.

P11. Eight artificial data sets are shown here.
For each one, find the value of r, without
computing if possible. Drawing a quick
sketch might be helpful.
b.
c.
a.
b.
c.
d.
e.
f.
g.
h.
d.
e.
Display 3.51
Five scatterplots.
155
22-03-2009 20:55
Lesson
17 de 22
P12. The table in E12 ( Display 3.31 on page 134 )

gives the amount of fat and number of calories
in various pizzas. In P10, you used your
calculator to find the correlation, r. This
time, make a table like that in Display 3.42 on
page 142, and use the formula to find r. What
do you notice about the products
P13. The scatterplot in Display 3.52 is divided into
quadrants by vertical and horizontal lines that
pass through the point of averages, ( x, y )
Display 3.52 Scatterplot divided into quadrants at

the point of averages, ( x, y ).
a. Is the correlation positive or negative?

b. Give the coordinates of the point that will
contribute the most to the correlation, r.
c. Consider the product
Where are the points that have a positive

product? How many of the 30 points have
a positive product?
d. Where are the points that have a negative
product? How many of the 30 points have
a negative product?
Correlation and the Appropriateness
of a Linear Model
P14. Both plots in Display 3.53 have a correlation
of 0.26. For each plot, is fitting a regression
line (as shown on the plot) an appropriate
thing to do? Why or why not?
Display 3.53 Two scatterplots with the same

correlation.
The Relationship Between the Correlation

and the Slope
P15. Imagine a scatterplot of two sets of exam
scores for students in a statistics class. The
score for a student on Exam 1 is graphed
on the x-axis, and his or her score on Exam
2 is graphed on the y-axis. The slope of the
regression line is 0.368. The mean of the
Exam 1 scores is 72.99, and the standard
deviation is 12.37. The mean of the Exam 2
scores is 75.80, and the standard deviation
is 7.00.
a. Find the correlation of these scores.
b. Find the equation of the regression line for
predicting an Exam 2 score from an Exam
1 score. Predict the Exam 2 score for a
student who got a score of 80 on Exam 1.
22-03-2009 20:55
Lesson
18 de 22
c. Find the equation of the regression line

for predicting an Exam 1 score from an
Exam 2 score.
d. Sketch a scatterplot that could represent
the situation described.
P16. If you take a random sample of U.S. cities
and measure the number of fast-food
franchises in each city and the number of
cases of stomach cancer per year in the city,
you find a high correlation.
a. What is the lurking variable?
b. How would you adjust the data for the
lurking variable to get a more meaningful
comparison?
P17. If you take a random sample of public school
students in grades K12 and measure weekly
allowance and size of vocabulary, you will
find a strong relationship. Explain in terms
of a lurking variable why you should not
conclude that raising a students allowance
will tend to increase his or her vocabulary.
P18. For the countries of the United Nations,
there is a strong negative relationship
between the number of TV sets per
thousand people and the birthrate. What
would be a careless conclusion about
cause and effect? What is the lurking
variable?
Interpreting r2
P19. Data on the association between high
school graduation rates and the percentage
of families living in poverty for the 50 U.S.
states were presented in E26. Display 3.54
contains the scatterplot and a standard
computer output of the regression analysis.
Display 3.54
Poverty rates versus high school

graduation rates.
a. Under SOURCE, the Total variation

is the SST, and the Error variation is the
SSE. From this information, find r, the
correlation.
b. Write an interpretation for r2 in the
context of these data.
c. Does the presence of a linear relationship
here imply that a state that raises its
graduation rate will cause its poverty rate
to go down? Explain your reasoning.
d. What are the units for each of the values
x, y, b1, and r?
157
22-03-2009 20:55
Lesson
19 de 22

P20. The plot in Display 3.55 shows the heights
of older sisters plotted against the heights
of their younger sisters. On a copy of this
scatterplot, draw vertical lines to divide the
points into six groups. Mark the approximate
location of the mean of the y-values of each
vertical strip. Sketch the regression line,
y = 43 + 0.337x. Note that the regression
line comes as close as possible to the mean

of each vertical strip. Now draw an ellipse
around the data and connect the two ends
of the ellipse. Is the regression line flatter
than this line? Does this plot show the
regression effect?
P21. Display 3.56 shows the first two exam
scores for 29 college students enrolled in an
introductory statistics course. Do you see
any evidence of regression to the mean? If so,
explain the nature of the evidence.
Display 3.55 The heights of older sisters versus the

heights of their younger sisters.
Display 3.56
Exam scores.
Exercises
E27. Each scatterplot in Display 3.57 was made on
the same set of axes. Match each scatterplot
with its correlation, choosing from 0.06,
0.25, 0.40, 0.52, 0.66, 0.74, 0.85, and 0.90.
a.
b.
c.
d.
e.
f.
g.
h.
Display 3.57 Eight scatterplots with various

correlations.
E28. Estimate the correlation between the

variables in these scatterplots.
a. The proportion of the state population
living in dorms versus the proportion
living in cities in Display 3.4 on page 109.
b. The graduation rate versus the 75th

percentile of SAT scores in E5 on
page 113.
c. The college graduation rate versus the
percentage of students in the top 10%
of their high school graduating class
in E5 on page 113.
E29. For each set of pairs, ( x, y ), compute the
correlation by hand, standardizing and
finding the average product.
a. ( 2, 1 ), ( 1, 1 ), ( 0, 0 ), ( 1, 1 ), ( 2, 1 )
b. ( 2, 2 ), ( 0, 2 ), ( 0, 3 ), ( 0, 4 ), ( 2, 4 )
E30. For each artificial data set in P11 on
page 155, compute the correlation by hand,
standardizing and finding the average
product.
E31. The scatterplot in Display 3.58 shows part
of the hat size data of E6 on page 113. The
plot is divided into quadrants by vertical and
horizontal lines that pass through the point
of averages, ( x, y ).
22-03-2009 20:55
Lesson
20 de 22
b. Draw a pair of elliptical scatterplots to

illustrate each comparison.
i. One
is larger than the other, the
are equal, and the correlations are
weak.
ii. One
strong.
Display 3.58
Head circumference, in inches, versus

hat size.
a. Estimate the value of the correlation.

b. Using the idea of standardized scores,
explain why the correlation is positive.
c. Identify the point that contributes the
most to the correlation. Explain why the
contribution it makes is large.
d. Identify a point that contributes little
to the correlation. Explain why the
contribution it makes is small.
E32. The ellipses in Display 3.59 represent
scatterplots that have a basic elliptical shape.
E33. Several biology students are working

together to calculate the correlation for the
relationship between air temperature and
how fast a cricket chirps. They all use the
same crickets and temperatures, but some
measure temperature in degrees Celsius and
others measure it in degrees Fahrenheit.
Some measure chirps per second, and others
measure chirps per minute. Some use x
for temperature and y for chirp rate, while
others have it the other way around.
a. Will all the students get the same value
for the slope of the least squares line?
Explain why or why not.
b. Will they all get the same value for the
correlation? Explain why or why not.
Display 3.59 Three pairs of elliptical scatterplots.
a. Match these conditions with the

corresponding pair of ellipses.
I. One
strong.
II. One of the correlations is stronger
than the other, the
are equal, and
the
are equal.
III. One
weak.
E34. For the sample of top-rated universities

in E5 on page 113, the graduation rate has
mean 82.7% and standard deviation 8.3%.
The student/faculty ratio has mean 11.7
and standard deviation 4.3. The correlation
is 0.5.
a. Find the equation of the least squares
line for predicting graduation rate from
student/faculty ratio.
b. Find the equation of the least squares line
for predicting student/faculty ratio from
graduation rate.
159
22-03-2009 20:55
Lesson
21 de 22
E35. These questions concern the relationship

b. Over the last 30 years, there has been
between the correlation, r, and the slope, b1,
a strong positive correlation between
of the regression line.
the average price of a cheeseburger and
the average tuition at private liberal arts
a. If y is more variable than x, will the slope
colleges.
of the least squares line be greater (in
absolute value) than the correlation?
c. Over the last decade, there has been
Justify your answer.
a strong positive correlation between
the price of an average share of stock,
b. For a list of pairs (x, y), r = 0.8, b1 = 1.6,
as measured by the S&P 500, and the
and the standard deviations of x and y are
number of Web sites on the Internet.
25 and 50. (Not necessarily in that order.)
Which is the standard deviation for x?
E38. Manufacturers of low-fat foods often
Justify your answer.
increase the salt content in order to keep
the flavor acceptable to consumers. For a
c. Students in a statistics class estimated and
sample of different kinds and brands of
then measured their head circumferences
cheeses, Consumer Reports measured several
in inches. The actual circumferences had
variables, including calorie content, fat
SD 0.93, and the estimates had SD 4.12.
content, saturated fat content, and sodium
The equation of the least squares line for
content. Using these four variables, you can
predicting estimated values from actual
form six pairs of variables, so there are six
values was = 11.97 + 0.36x. What was
different correlations. These correlations
the correlation?
turned out to be either about 0.95 or
d. What would be the slope of the least
about 0.5.
squares line for predicting actual head
a. List all six pairs of variables, and for each
circumferences from the estimated
pair decide from the context whether the
values?
correlation is close to 0.95 or to 0.5.
E36. Lost final exam. After teaching the same
b. State a careless conclusion based on
history course for about a hundred years, an
taking the negative correlations as
instructor has found that the correlation, r,
evidence of cause and effect.
between the students total number of points
before the final examination and the number
c. Explain the negative correlation using the
of points scored on their final examination
idea of a lurking variable.
is 0.8. The pre-final-exam point totals for all E39. A study to determine whether ice cream
students in this years course have mean 280
consumption depends on the outside
and SD 30. The points on the final exam have
temperature gave the results shown in
mean 75 and SD 8. The instructors dog ate
Display 3.60.
Julies final exam, but the instructor knows
that her total number of points before the
exam was 300. He decides to predict her final
exam score from her pre-final-exam total.
What value will he get?
E37. Lurking variables. For each scenario, state
a careless conclusion assuming cause and
effect, and then identify a possible lurking
variable.
a. For a large sample of different animal
species, there is a strong positive
correlation between average brain
weight and average life span.
22-03-2009 20:55
Lesson
22 de 22
Display 3.60 Data table, scatterplot, and regression analysis for the effects of outside
temperature on ice cream consumption. [Source: Koteswara Rao Kadiyala, Testing for
the Independence of Regression Disturbances, Econometrica 38 (1970): 97117.]
a. Use the values of SST and SSE in the

regression analysis to compute r, the
correlation for the relationship between
the temperature in degrees Fahrenheit
and the number of pints of ice cream
consumed per person. Check your
answer against R-sq in the analysis.
b. Compute the value of the residual that is
largest in absolute value.
c. Is there a cause-and-effect relationship

between the two variables?
d. What are the units for each of x, y, b1,
and r?
e. The letters MS stand for mean square.
How do you think the MS is computed?
161
22-03-2009 20:55
Lesson
1 de 17
E40. The scatterplot in Display 3.61 shows part of

the aircraft data of Display 3.12 on page 115.
For these data, r2 = 0.83. Should r2 be used
as a statistical measure for these data? If so,
interpret this value of r2 in the context of
the data. If not, explain why not.
E41. Suppose a teacher always praises students

who score exceptionally well on a test
and always scolds students who score
exceptionally poorly. Use the notion of
regression toward the mean to explain
why the results will tend to suggest the
false conclusion that scolding leads to
improvement whereas praise leads to
slacking off.
E42. A few years ago, a school in New Jersey
tested all its 4th graders to select students for
a program for the gifted. Two years later, the
students were retested, and the school was
shocked to find that the scores of the gifted
students had dropped, whereas the scores of
the other students had remained, on average,
the same. What is a likely explanation for
this disappointing development?
Display 3.61 Scatterplot of number of seats versus

fuel consumption (gal/h) for passenger
aircraft.
3.4
Diagnostics: Looking for Features That the

Summaries Miss
As you learned in Chapter 2, summaries simplify. They are useful because they
omit detail in order to emphasize a few general features. This quality also makes
summaries potentially misleading, because sometimes the detail that is ignored
has an important message to convey. Knowing just the mean and the standard
deviation of a distribution doesnt tell you if there are any outliers or whether the
distribution is skewed. The same is true of the regression line and the correlation.
This section is about diagnosticstools for looking beyond the summaries
to see how well they describe the data and what features they leave out. The first
part of this section deals with individual cases that stand apart from the overall
pattern and with how these cases influence the regression line and the correlation.
The second part shows you how to identify systematic patterns that involve many
or all of the casesthe shape of the scatterplot.
Which Points Have the Influence?

Just as among people,
some data points have
more influence than
others.
Not all data points are created equal. You saw in the calculation of the correlation
in Display 3.42 on page 142 that some points make large contributions and some
small. Some make positive contributions and some negative. Your goal is to learn
to recognize the points in a data set that might have an unusually large influence
on where the regression line goes or on the size and sign of the correlation.
22-03-2009 20:56
Lesson
2 de 17
Near and Far

What youll need: an open area in which to step off
distances
In this activity, you compare the actual distance to an
object with what the distance appears to be.
1. Go to an open area, such as the hall or lawn of your
school, and pick a spot as your origin. Choose six
objects at various distances from the origin. Five of
the objects should be within 10 to 20 paces, and the
other should be a long way away (at least 100 paces).
2. For each of the six objects, estimate the number of paces
from the origin to the object. Record your estimates.
3. From your origin, walk to each of your objects and
count the actual number of regular paces it takes you
to get there. Record this number beside your estimate.
4. Plot your data on a scatterplot, with your estimated
value on the x-axis and the actual value on the y-axis.
Does the plot show a linear trend?
5. Determine the equation of the regression line, and calculate the correlation.
6. Delete the point for the object that is farthest away from the origin.
Determine the equation of the regression line and calculate the correlation
for the reduced data set.
7. Did the extreme point have any influence on the regression line? On the
correlation? Explain.
In Chapter 2, you learned about outliers for distributionsvalues that are
separated from the bulk of the data. Outliers are atypical cases, and they can
exert more than their share of influence on the mean and standard deviation. For
scatterplots, as you will soon see, working with two variables together means that
there can be outliers of various kinds. Different kinds of outliers can have different
types of influence on the least squares line and the correlation. Unfortunately,
there is no rule you can use to identify outliers in bivariate data. Just look for
points surrounded by white space.
Judging a Points Influence
Points separated from the bulk of the data by white space are outliers and are
potentially influential. To judge a points influence, compare the regression
equation and correlation computed first with and then without the point
in question.
To see these ideas in action, turn to the data on mammal longevity in
Display 2.24 on page 43 and think about how to summarize the relationship
between maximum and average longevity.
3.4Diagnostics: Looking for Features That the Summaries Miss
163
22-03-2009 20:56
Lesson
3 de 17
Example: Influential Mammals

The average elephant lives 35 years. The oldest elephant on record lived 70 years.
The average hippo lives 41 yearslonger than the average elephantbut the
record-holding hippo lived only 54 years. The oldest-known beaver lived 50 years,
almost as long as the champion hippo, but the average beaver cashes in his wood
chips after only 5 short years of making them. Other mammals, however, are
more predictable. If you look at the entire sample, shown in Display 3.62, it
turns out that the elephant (E), hippo (H), and beaver (B) are the oddballs of
the bunch. For the rest, theres an almost linear relationship between average
longevity and maximum longevity. The least squares line for the entire sample
has the equation
M = 10.53 + 1.58A
where M, or M-hat, stands for predicted maximum longevity and A stands for
observed average longevity. For every increase of 1 year in average longevity, the
model predicts a 1.58-year increase in maximum longevity. The correlation for
the relationship between these two variables is 0.77. How much influence do the
oddballs have on these summaries?
Points surrounded by
white space might have
strong influence.
Display 3.62 Maximum longevity versus average longevity.
Solution
The hippo has the effect of pulling the right end of the regression line downward
( like putting a heavy weight on one end of a seesaw ), as you can see in Display 3.63.
When the hippo is removed, that end of the regression line will spring upward
and the slope will increase. Because one large residual has been removed and many
of the remaining residuals have been reduced in size, the correlation will increase.
The new slope is 1.96, and the new correlation is 0.80. The hippo has considerable
influence on the slope and some influence on the correlation.
Now envision the scatterplot with just the elephant, E, missing. Because E
is close to the straight line fi t to the data, it produces a small residual. Thus, you
would expect that removing E should not change the slope of the regression line
much (not nearly as much as removing H did ) and should reduce the correlation
just a bit. In fact, the correlation does decrease some, to 0.72 from 0.77. However,
the new slope is 1.53. It turns out that removing the elephant gives the hippo even
more influence, and the slope decreases.
22-03-2009 20:56
Lesson
4 de 17
Display 3.63 Regression lines for maximum longevity versus

average longevity, with and without the hippo.
Finally, envision the scatterplot with just the beaver, B, removed. B produces
a large, positive residual close to the left end of the regression line. Thus,
removing B should allow the left end of the line to drop, increasing the slope,
and removing a large residual should increase the correlation. The new slope
is 1.69 (an increase from 1.58), and the new correlation is 0.83 (an increase
from 0.77). The beaver also has considerable infl uence on both slope and
correlation.
165
22-03-2009 20:56
Lesson
5 de 17
With a little practice, you often can anticipate the influence of certain points
in a scatterplot, as in the previous example, but it is difficult to state general rules.
The best rule is the one given in the box on page 163: Fit the line with and without
the questionable point and see what happens. Then report all the results, with
appropriate explanations.
Why the Anscombe Data Sets Are Important
Display 3.64 shows four scatterplots. These plots, known as the Anscombe data
after their inventor, are arguably the most famous set of scatterplots in all of
statistics. The questions that follow invite you to figure out why statistics books
refer to them so often. In the process, youll learn more about what a summary
doesnt tell you about a data set.
Display 3.64
Four regression data sets invented by Francis J.

Anscombe. [Source: Francis J. Anscombe, Graphs in Statistical
Analysis, American Statistician 27 (1973): 1721.]
D26. For each plot in Display 3.64, first give a short verbal description of the
pattern in the plot. Then
a. either fit a line by eye and estimate its slope or tell why you think a line is
not a good summary
b. either estimate the correlation by eye or tell why you think a correlation
is not an appropriate summary
D27. Display 3.65 shows a computer output for one of the four Anscombe data set
plots. Can you tell which one? If so, tell how you know. If not, explain why
you cant tell.
22-03-2009 20:56
Lesson
6 de 17
Display 3.65 Regression analysis for one of the Anscombe data

sets.
D28. Display 3.66 lists values for the Anscombe plots.
Display 3.66
Anscombe plot data values.
a. Which plot has a point that is highly influential both with respect to the
slope of the regression line and with respect to the correlation?
b. Compared to the other points in the plot, does the influential point lie far
from the least squares line or close to it?
c. How would the slope and correlation change if you were to remove this
point? Discuss this first without actually performing the calculations.
Then carry out the calculations to verify your conjectures.
Residual Plots: Putting Your Data Under a Microscope

As you can see from the Anscombe plots, there are many features of the shape
of a scatterplot that you cant learn from the standard set of summary numbers.
Only when the cloud of points is elliptical, as in Display 3.41 on page 140, does
the least squares line, together with the correlation, give a good summary of
the relationship described by the plot. If the cloud of points isnt elliptical, thes
summaries arent appropriate. How can you decide?
167
22-03-2009 20:56
Lesson
7 de 17
Residual plots may

uncover more
detailed patterns.
A special kind of scatterplot, called a residual plot, often can help you see
more clearly whats going on. For some data sets, a residual plot can even show
you patterns you might otherwise have overlooked completely. Statisticians use
residual plots the way a doctor uses a microscope or an X rayto get a better
look at less obvious aspects of a situation. ( Plots you use in this way are called
diagnostic plots because of the parallel with medical diagnosis. ) Push the
analogy just a little. Youre the doctor, and data sets are your patients. Sets with
elliptical clouds of points are the healthy ones; they dont need special attention.
A residual plot is a scatterplot of residuals, yy , versus predictor values, x
(or, sometimes, versus predicted values, y ).
Example: Constructing a Residual Plot

Return to the data on percentage of on-time arrivals versus mishandled
baggage for airlines, introduced in P2 on page 110. Calculate the residuals and
make a plot.
Solution
Visualize each residualthe difference between the observed value of y and the
predicted value, yas a vertical segment on the scatterplot in Display 3.67.
Display 3.67
Scatterplot of airline data.
The calculated residuals are shown in Display 3.68, with the list of carriers
ordered from smallest to largest on the x-scale. This allows the size of the residuals
in the far right column to appear in the same order as in Display 3.67. Alaska
produces a negative residual of modest size, whereas US Airways produces a large
positive residual.
The residual plot, Display 3.69, is simply a scatterplot of the residuals versus
the original x-variable, mishandled baggage. Note that 0 is at the middle of the
residuals on the vertical scale.
22-03-2009 20:56
Lesson
8 de 17
Display 3.68 Table showing residuals for the airline data.
Display 3.69
Residual plot for the airline data.
The residual plot shows nearly random scatter, with no obvious trends. This
is the ideal shape for a residual plot, because it indicates that a straight line is
reasonable model for the trend in the original data. [You can use your calculator to
create residual plots. See Calculator Note 3I.]
Residual Plots
D29. In Display 3.69, identify which residual belongs to Delta and which
to Northwest.
169
22-03-2009 20:56
Lesson
9 de 17
D30. To see how residual plots magnify departures from the regression line,
compare the Anscombe plots in Display 3.64 with Display 3.70, which shows
the four corresponding residual plots in scrambled order.
Display 3.70 Residual plots for the four Anscombe data sets.
a. Match each of the original scatterplots in Display 3.64 with its

corresponding residual plot in Display 3.70.
b. Describe the overall difference between the original scatterplots and
the residual plots. What do the scatterplots show that the residual plots
dont? What do the residual plots show that the scatterplots dont?
What to Look For in a Residual Plot
A careful data analyst always looks at a residual plot.
If the original cloud of points is elliptical, so that a line is an appropriate
summary, the residual plot will look like a random scatter of points.
Residual plots
sometimes yield
surprises.
Use residual plots to check for systematic departures from constant slope
(linear trend) and constant strength (same vertical spread). Look in particular
for plots that are curved or fan-shaped. Its true that for data sets with only one
predictor value (like those in this chapter), you often can get a good idea of what
the residual plot will look like by carefully inspecting the original scatterplot.
Once in a while, however, you get a surprise.
Example: Interpreting a Residual Plot

E19 on page 136 introduced data on median height versus age for young girls.
Display 3.71 shows the scatterplot of these data, with the regression line. The overall
average growth rate for the 12-year period is the slope of the regression line.
22-03-2009 20:56
Lesson
10 de 17
The plot looks nearly linear, but is a line a suitable model?
Display 3.71 Median height versus age for young girls.
Solution
The residual plot, shown in Display 3.72, quite dramatically reveals that the trend
is not as linear as first imagined. The curvature in the residual plot mimics the
curvature in the original scatterplot, which is harder to see. A line is not a good
model for these data.
Display 3.72 Residual plot of median height versus age for

young girls.
Residuals sometimes
are plotted against the
predicted values, y.
Statistical software often plots residuals against the predicted values, y, rather
than against the predictor values, x. For simple linear regression, both plots have
exactly the same shape as long as the slope of the regression line is positive.
Types of Residual Plots
D31. Display 3.73 shows a scatterplot and two residual plots for the data set
consisting of these three ordered pairs (x, y): (0, 1), (1, 0), and (2, 2). One
residual plot plots residuals versus predictor values, x, the sort of plot you
171
22-03-2009 20:56
Lesson
11 de 17
get from graphing calculators. The other plots residuals versus predicted
(fitted) y-values, or y, the sort of plot you get from computer software
packages. Explain how the residual plots were produced and how you can
tell which residual plot is which. The equation of the least squares line is
y = 0.5 + 0.5x.
Display 3.73
A scatterplot and two residual plots.
Summary 3.4: Diagnostics: Looking for Features

That the Summaries Miss
For the simplest clouds of data pointselliptical in shape, with linear trend and
no outliersyou can summarize all the main features of a scatterplot with just a
few numbers, mainly the slope of the fitted line, y-intercept, and correlation. Not
all plots are this simple, however, and a good statistician always does diagnostic
checks for outliers and influential points and for departures from constant slope
or constant strength.
Points separated from the bulk of the data by white space are outliers and
potentially influential.
To judge a points influence, fit a line to the data and compute a regression
equation and a correlation first with and then without the point in question.
If the change in the regression equation and correlation is meaningful in your
situation, report both sets of summary statistics.
For some data sets, a residual plot can show patterns you might otherwise
overlook. A residual plot is a scatterplot of residuals, y y , versus predictor
values, x. A residual plot also can be constructed as a scatterplot of residuals,
y y , versus fitted values, . Use residual plots to check for systematic departures
from linearity and for constant variability in y across the values of x. If the data
arent linear, the residual plot doesnt look random. If the data have nonconstant
variability, the residual plot is fan-shaped.
Practice
Which Points Have the Influence?
P22. The data in Display 3.74 show some
interesting patterns in the relationship
between domestic and international gross
income from the ten movies with the highest
domestic gross ticket sales.
a. Construct a scatterplot suitable for

predicting international sales from
domestic sales. Describe the pattern in
the data.
22-03-2009 20:56
Lesson
12 de 17
b. Find the least squares line and the

correlation for these data.
c. Remove the most influential data point
and recalculate the least squares line and
correlation. Describe the influence of the
removed point.
P23. A data table and scatterplot of one students

results from Activity 3.4a are shown in
Display 3.75.
a. How well did the student do in estimating
the number of paces?
b. Which point appears to be most
influential?
c. Calculate the slope of the regression line
and the correlation with and without
this point. Describe the influence of this
point.
Display 3.74 Ticket sales for the ten highest-grossing

domestic (United States and Canada)
movies of all time. [Source: Internet Movie
Database, us.imdb.com, September 12, 2006.]
Display 3.75
Sample data from Activity 3.4a.
Residual Plots
P24. For the set of (x, y) pairs (0, 0), (0, 1), (1, 1),
and (3, 2), the equation of the least squares
line is y = 0.5 + 0.5x.
a. Plot the data and graph the least squares
line.
b. Next complete a table for the predicted
values and residuals, like the table in
Display 3.68 on page 169.
c. Using the values in your table, plot
residuals versus predictor, x.
d. How does the residual plot differ from the
scatterplot?
173
22-03-2009 20:56
Lesson
13 de 17
P25. Display 3.76 shows four scatterplots

( AD ) for the data from a sample of
commercial aircraft. Display 3.77 shows four
corresponding residual plots (IIV).
a. Match the residual plots to the
scatterplots.
b. Using scatterplots AD as examples,
describe how you can identify each of
these in a scatterplot from the residual
plot.
i. a curve with increasing slope
ii. unequal variation in the responses

iii. a curve with decreasing slope
iv. two linear patterns with di erent
slopes
c. For one of the plots, two line segments
joined together seem to give a better fit
than either a single line or a curve. Which
plot is this? Is this pattern easier to see in
the original scatterplot or in the residual
plot?
Display 3.76 Four scatterplots for the sample of commercial aircraft.
Display 3.77 Four residual plots corresponding to the scatterplots in Display 3.71.
22-03-2009 20:56
Lesson
14 de 17
Exercises
E43. Extreme temperatures. The data in Display
3.78 provide the maximum and minimum
temperatures ever recorded on each continent.
E44. The data and plot in Display 3.79 are from

E15 on page 135. They show the arsenic
concentrations in the toenails of 21 people
who used water from their private wells.
Both measurements are in parts per million.
Display 3.78 Maximum and minimum recorded

temperatures for the continents.
[Source: National Climatic Data Center, 2005,
www.ncdc.noaa.gov .]
a. Construct a scatterplot of the data

suitable for predicting the minimum
temperature from a given maximum
temperature. Is a straight line a good
model for these points? Explain.
b. Fit a least squares line to the points and
calculate the correlation, even if you
thought in part a that a straight line was
not a good model.
c. Explain, in words and numbers, what
influence Antarctica has on the slope of
the regression line and on the correlation.
How could an account of these data be
misleading if it were not accompanied by
a plot?
Display 3.79 Arsenic concentrations.
Two climbers stand on Mount Erebus,

Antarctica, 12,500 ft above sea level.
a. Which point do you think has the most

influence on the slope and correlation?
What would be the effect of removing
3.4Diagnostics: Looking for Features That the Summaries Miss
175
22-03-2009 20:56
Lesson
15 de 17
this point? Perform the calculations to see

if your intuition is correct.
b. Find a point that you think has almost no
influence on the slope and correlation.
Perform the calculations to see if your
intuition is correct.
c. Find a point whose removal you think
would make the correlation increase.
Perform the calculations to see if your
intuition is correct.
E45. How effective is a disinfectant? The data in
Display 3.80 show (coded) bacteria colony
counts on skin samples before and after a
disinfectant is applied.
Display 3.80 Coded bacteria colony counts before

(x) and after (y) treatment. [Source:
Snedecor and Cochran, Statistical Methods (Iowa
State University Press, 1967), p. 422.]
a. Plot the data, fit a regression line to them,

and complete a copy of the table, filling in
the predicted values and residuals.
b. Plot the residuals versus x, the count
before the treatment. Comment on the
pattern.
c. Use the residual plot to determine for
which skin sample the disinfectant was
unusually effective and for which skin
sample it was not very effective.
E46. Textbook prices. Display 3.81 compares
recent prices at a college bookstore to those
of a large online bookstore.
a. The equation of the regression line is
online = 3.57 + 1.03 college. Interpret
this equation in terms of textbook prices.
Display 3.81 Prices for a sample of textbooks at

a college bookstore and an online
bookstore.
b. Construct a residual plot. Interpret it and

point out any interesting features.
c. In comparing the prices of the textbooks,
you might be more interested in a
different line: y = x. Draw this line on a
copy of the scatterplot in Display 3.81.
What does it mean if a point lies above
this line? Below it? On it?
d. A boxplot of the differences
college price online price is shown
in Display 3.82. Interpret this boxplot.
Display 3.82 A boxplot of the differences between

the college price and the online price
for various textbooks.
22-03-2009 20:56
Lesson
16 de 17
E47. Pizzas, again. Display 3.83 shows the

pizza data from E12 on page 134, with its
regression line.
Display 3.83 Calories versus fat, per 5-oz serving, for

seven kinds of pizza.
a. Estimate the residuals from the graph,

and use your estimates to sketch a rough
version of a residual plot for this data set.
b. Which pizza has the largest positive
residual? The largest negative residual?
Are any of the residuals so extreme as
to suggest that those pizzas should be
regarded as exceptions?
c. Is any one of the pizzas a highly
influential data point? If so, specify
which one(s), and describe the effect
on the slope of the fitted line and the
correlation of removing the influential
point or points from the analysis.
E48. Aircraft. Look again at Display 3.76 on page
174, which shows a scatterplot of flight length
versus number of seats.
a. Does the slope of the pattern increase,
decrease, or stay roughly constant as you
move from left to right across the plot?
b. Focusing on the variation (spread) in
flight length, y, for planes with roughly
the same seating capacity, compare
the spreads for planes with few seats, a
moderate number of seats, and a large
number of seats. As you move from left to
right across the plot, how does the spread
change, if at all?
c. Suppose a friend chose a plane from

the sample at random and told you the
approximate number of seats. Could you
guess its flight length to within 500 miles
if the number of seats was between 50
and 150? If it was between 200 and 300?
Explain.
d. What is the relationship between your
answer in part b and residual plot I in
Display 3.77?
e. Give an explanation for why the variation
in flight length shows the pattern it does.
E49. Match each scatterplot ( AD ) in Display
3.84 with its residual plot ( IIV ) in Display
3.85. For which plots is a linear regression
appropriate?
Display 3.84 Four scatterplots.
Display 3.85 Four residual plots.
177
22-03-2009 20:56
Lesson
17 de 17
E50. Can either of the plots in Display 3.86 be a

residual plot? Explain your reasoning.
Display 3.86
E53. Can you recapture the scatterplot from the

residual plot? The residual plot in Display
3.88 was calculated from data showing the
recommended weight (in pounds) for men at
various heights over 64 in. The fitted weights
ranged from 145 lb to 187 lb. Make a rough
sketch of the scatterplot of these data.
Residual Plots?
E51. Display 3.87 gives the data set for the three
passenger jets from the example on page 123,
along with a scatterplot showing the least
squares line. (Values have been rounded.)
a. Use the equation of the line to find
predicted values and residuals to
complete the table in Display 3.87.
b. Use your numbers from part a to construct
two residual plots, one with the predictor,
x, on the horizontal axis and the other with
the predicted value, y, on the horizontal
axis. How do the two plots differ?
Display 3.88 Residuals of recommended weight

versus height for men.
E54. The plot in Display 3.89 shows the residuals

resulting from fitting a line to the data for
female life expectancy (life exp) versus gross
national product (GNP, in thousands of
dollars per capita) for a sample of countries
from around the world. The regression
equation for the sample data was
life exp = 67.00 + 0.63 GNP
Sketch the scatterplot of life exp versus GNP.
Display 3.87 Cost per hour versus number of seats for

three models of the passenger aircraft.
E52. Explain why a residual plot of ( x, residual )

and a plot of ( predicted value, residual ) have
exactly the same shape if the slope of the
regression line is positive. What changes if
the slope is negative?
Display 3.89 Residuals of female life expectancy

versus gross national product.
22-03-2009 20:56
Lesson
1 de 24
3.5
Transforming data is
sometimes called
re-expressing data.
Shape-Changing Transformations
For scatterplots in which the points form an elliptical cloud, the regression line
and correlation tell you pretty much all you need to know. But data dont always
behave so obligingly. For plots in which the points are curved, fan out, or contain
outliers, the usual summaries do not tell you everything and can actually be
misleading. What do you do then? This section shows you one possible remedy:
Transform the data to get the shape you want.
Youre already familiar with linear transformations from Section 2.4things
like changing temperatures from degrees Fahrenheit to degrees Celsius or
changing distances from feet to inches or times from minutes to seconds. These
linear transformationsadding or subtracting a constant or multiplying or
dividing by a constantcan change the center and spread of the distribution
without changing its basic shape.
Nonlinear transformations, such as squaring each value or taking logarithms,
do change the basic shape of the plot. This section shows how a transformation of
a measurement scale can lead to simplified statistical analyses. One of the most
common nonlinear relationships is the exponential, and that is where we begin
our discussion.
Exponential Growth and Decay

Whenever a quantity
changes by an amount
proportional to the
amount present,
consider logarithms.
Data over time will have

more varied patterns,
sometimes diffcult to
see.
Exponential functions often arise when you study how a quantity grows or decays
as time passes. Many such quantities grow by an amount that is proportional to
the amount present. This means that the amount present at one point in time is
multiplied by a fixed constant to get the amount
present at the next point in time,
resulting in a function of the form y = abx. The amount of growth of a population
is proportional to population size, the growth of a bank account is proportional
to the amount of money in the account, the amount by which a radioactive
substance decays is proportional to how much is left, and the amount by which
a cup of coffee cools is proportional to how much hotter it is than the air around
it. For such situations, replacing y with log y often gives a plot that is much more
nearly linear than the original plot.
The examples in this section illustrate exponential growth and decay. At
the same time, they illustrate another important feature of measurements taken
over time: There is often an up-and-down pattern to the residuals. This pattern
cannot be removed by a simple transformation.
Whenever your measurements come in chronological order, what happens
next is likely to depend on what just happened. As a result, the patterns in your
data might be subtler than the ones youve seen up to this point. In addition, the
difference between a meaningful pattern and a quirk in the data might be harder
to detect because the data typically show only one observation for a single point
in time.
An Example of Exponential Decay
In Activity 3.5a, you will study a population that decreases exponentially
over time.
179
22-03-2009 20:57
Lesson
2 de 24
Copper Flippers
What youll need: 200 pennies, a paper cup
Count 200 pennies. Youll use these for an exercise
to see how fast pennies die.
A certain insect, the copper flipper, has a life
span determined by the fact that there is a 5050
chance a particular live flipper will die at the end
of the day. So, on average, half of any population of
copper flippers will die during the first day of life. Of those that survive the first
day, on average half die during the second day, and so on.
By extraordinary coincidence, these bugs behave like a bunch of tossed
pennies. If you toss 200 pennies, about half should come up heads and half
tails. The heads represent the insects that survive the first day, and the tails
represent those that die. You can collect the pennies that came up heads and
toss them again to see how many survive the second day, and so on. This gives
you a physical model for the distribution of the life span of the insects.
1. Place your pennies in a cup, shake them up, and toss them on a table.
Count the number of heads, and record the number.
2. Set aside the pennies that came up tails. Place the pennies that came up
heads back in the cup, and repeat the process.
3. After each toss (day), set the tails aside, collect the heads, count them, and
toss them again. Repeat this process until you have fewer than five pennies
left, but stop before you get to zero heads.
4. Construct a scatterplot of your data, with the number of the toss on the
horizontal axis and the number of heads on the vertical axis. Does the
pattern look linear?
5. How might you find an equation to summarize this pattern?
Exponential Functions and Log Transformations

How do you know if you have an exponential relationship of the form y = abx ?
One clue is that the points cluster about a function similar to those in Display
3.90. Another clue is that you have a variable whose values are mostly clustered
at one end but range over two or more orders of magnitude (powers of 10).
The best test, however, is to take the log of each value of y and see if this will
straighten the points.
Replacing y with
log y will straighten
y = abx.
Exponential relationships have an underlying model of the form y = abx. If

a > 1, a 1 gives the growth rate per time period. If 0 < a < 1, a 1 is
negative and gives the decay rate. The points can be linearized (straightened)
by taking the logarithm (base 10 or base e) of each value of y. The result will be
a linear equation of the form
log y = log a + (log b)x
22-03-2009 20:57
Lesson
3 de 24
Display 3.90 The exponential functions
Analyzing the Copper Flippers

Display 3.91 shows the scatterplot for one students data from Activity 3.5a.
Display 3.91 Number of heads versus number of the toss.
Does a log transformation appear to be appropriate here? The pattern looks

much like the left-hand curve in Display 3.90, and the values for the number of
heads are clustered at the smaller values but range over two orders of magnitude.
In addition, the number of heads remaining after each toss of the coins is
roughly proportional to the number of coins tossed. A log transformation is
worth a try. Display 3.92 shows the natural log (base e) of the number of heads
plotted against the toss number, along with the regression line, for the data of
Display 3.91.
Compare the scatterplot in Display 3.92 to the residual plot in Display 3.93.
(The line segments are added to help your eye follow the time sequence.) Does
the model appear to fit well if you look only at the scatterplot? How, if at all,
does the residual plot alter your judgment of how well the line fits? The cyclical
up-and-down pattern of residuals is common in such time series data.
The equation of the regression line (shown in Display 3.92) for the
= 5.21 0.66x. If you solve this
transformed data is given by the equation
equation for y, you get y =
.The number of copper fippers that are
alive each day is about 52% of the number alive the previous day. In other words,
the decay rate is estimated to be 48%, or 0.48.
3.5
181
22-03-2009 20:57
Lesson
4 de 24
Display 3.92
A plot of ln(heads) versus the number of the toss,
with the regression line.
Display 3.93 Residual plot for the ln(heads) regression.

D32. How would the scatterplot and the least squares line change if the coins had
a probability of 0.6 of coming up heads? What insect death rate does this
situation model?
An Example of Exponential Growth
Display 3.94 shows the population density (people per square mile) of the
United States for all census years through 2000. For the years prior to 1960,
only the 48 contiguous states are included. Alaska and Hawaii were added to
the census in 1960. To find a reasonable model for this situation, start with a
scatterplot.
Plot. Obviously, the pattern here is not linear. A curve of this type can be
straightened by proportionally decreasing the large y-values (population densities,
in this case). For variables like population growth (or the growth of many other
phenomena), the logarithmic transformation works well. This transformation not
only solves the data analysis problem nicely but also gives a neat interpretation to
the resulting model.
22-03-2009 20:57
Lesson
5 de 24
Display 3.94
Population density of the United States over the

census years. [Source: U.S. Census Bureau, Statistical Abstract of
the United States, 20042005, p. 8.]
Transform and Plot Again. You can see the transformed points in Display
3.95. For example, for the year 1800, the point (1800, ln 6.1) or (1800, 1.808) is
plotted.
Fit. Although there is still some curvature, the pattern is much more nearly
linear, and a straight line might be a reasonable model to fit these data. The
regression line and regression analysis also are shown in Display 3.95 (on the
next page).
The equation of the regression line is
= 25.118 + 0.0148x. Solving for
y gives
This means that the population density is growing at about 1.5% per year.
183
22-03-2009 20:57
Lesson
6 de 24
Display 3.95
ln (population density) versus year, with regression

line and computer output.
Residuals. In the case of data over time, it is often advantageous to plot the
residuals over time, as in Display 3.96.
Display 3.96 Residual plot of ln (population density) versus year.
The result is not exactly random scatter! Well, it is about the best you can do.
The problem is that there are subtle patterns in the dataand in the residuals
that no simple model will adequately account for.
D33. Relate the pattern you see in the residual plot in Display 3.96 to the pattern
of the data in the original scatterplot (Display 3.94).
22-03-2009 20:57
Lesson
7 de 24
D34. Why is there a huge jump from a large positive residual to a large negative
residual as you move from 1800 to 1810? What events in U.S. history explain
some of the other features of the residual plot?
D35. If you use a computer to fit a line to the (year, density) data, it will
automatically compute a correlation.
a. Explain why a correlation is not a very useful summary for this data set.
b. In Display 3.95, the computer gave the value of R-sq as 97.3% for the
transformed data. Statisticians ordinarily are not very interested in the
size of this diagnostic measure for time-ordered data. Can you think of
any reasons why?
D36. What is the estimated annual rate of growth of the population density of the
United States? What is the estimated rate of growth over a decade?
Example: Logarithmic Transformation
Do aircraft with a higher typical speed also have a greater average flight length? As
you might expect, the answer is yes, but the relationship is nonlinear. (See Display
3.97.) Is there a simple equation that relates typical speed and flight length? The
solution that follows leads you through one approach to these questions.
Display 3.97 Data table and scatterplot for the flight length and
speed of various aircraft. [Source: Air Transport Association
of America, 2005, www.air-transport.org.]
185
22-03-2009 20:57
Lesson
8 de 24
Solution
The plot is curved much in the manner of exponential growth, so shrinking the
y-scale is in order. Display 3.98 shows a scatterplot of ln( flight length) versus
speed, along with a least squares line and computer output.
Display 3.98 ln(flight length) versus speed.
Although the pattern of points appears much more linear and the fit looks
pretty good, the lack of randomness in the residual plot in Display 3.99 indicates
that a linear model still does not really fit the points. D38 and E58 will give you a
chance to continue the detective work.
Display 3.99 Residual plot for ln(flight length) versus speed.
22-03-2009 20:57
Lesson
9 de 24
Log Transformations
D37. How can you tell from Display 3.97 that the flight length values range over
two orders of magnitude? Show how transforming from flight length to
ln( flight length) will shrink the larger y-values more than the smaller ones
and thus help straighten the plot.
D38. Describe the pattern in the residual plot in Display 3.99, and tell what it
suggests as a next step in the analysis.
D39. In what sense, if any, is the relationship in Display 3.97 one of cause and
effect? What is your evidence?
Log-Log Transformations of Power Functions

Wildlife biologists can estimate the length of an alligator without getting very
close to it. However, to get its weight with any accuracy, they have to wrestle
it onto a scale. This is a procedure that neither the biologist nor the alligator is
happy to participate in.
Perhaps you can help the biologist predict the weight of an alligator spotted in
the swamp from an estimate of its length. One way to do that is to collect data on
both weight and length and then find a regression line that provides a good model
of the relationship. Display 3.100 shows the weights and lengths of 25 alligators as
measured by experts from the Florida Game and Freshwater Fish Commission.
Display 3.100 Alligator weights and lengths. [ Source: Florida Game

and Freshwater Fish Commission.]
A scatterplot of the data (Display 3.101, on the next page) shows that the
relationship is not linear. On thinking carefully about an appropriate model, you
might realize that length is a linear measure while weight is related to volumea
cubic measure. So perhaps weight is related to the cube of length, or some power
close to that. That is, the relationship is of the form weight
where a
is constant.
187
22-03-2009 20:57
Lesson
10 de 24
Display 3.101 Plot of the alligator data.
Power relationships have an equation of the form

y
axb
as the underlying model. The points can be linearized (straightened) by

taking the logarithm (base 10 or base e) of both the values of x and the values
of y. The result will be a linear equation of the form
log y log a b log x
Thus, if ln(weight) is plotted against ln(length), the plot should be fairly linear
and the slope of the least squares line will provide an estimate of the power, b. The
result of this transformation is shown in Display 3.102. The regression equation is
ln(weight) = 3.29 ln(length) 10.2.
The plot does indeed look linear, and the estimate of b is 3.29. (Natural logs
are used here, but logs to base 10 will produce essentially the same results.) The
biologist can use this model to predict ln(weight) from ln(length) and then change
the predicted value back to the original scale, if he or she chooses.
Note that the residual plot still shows a bit of curvature, mainly because the
three largest alligators are somewhat influential. But the offsetting advantage
of the power model is that the residuals are fairly homogeneous; that is, they
dont tend to grow or shrink as length increases. This means that the error in
the prediction of weight will be relatively constant for alligators of all lengths.
The exponential model also fits these data well, but the residuals then lose their
homogeneity with no substantial decrease in their size.
Ultimately, a model should be selected based on its intended use.These
biologists wanted a model that predicts weight well for all reasonable values of
length, not just for large alligators. Furthermore, a model should make sense to
the experts in the field of use. The biologists could understand why weight (or
volume) should have a cubic relationship with length but could see no reason why
weight should grow exponentially with length.
22-03-2009 20:57
Lesson
11 de 24
Display 3.102 Linear regression of ln(weight) versus ln(length),

with residual plot.
Example: Using the Regression Equation

Use the regression equation to predict the weight of an alligator that is 100 inches
in length.
Solution
The prediction is ln(weight) = 3.29ln(100) 10.2
is found by solving
ln(weight)
weight
4.951. The predicted weight
4.951
141.3 lb
Both biologist and alligator can rest more easily.

Log-Log Transformations of Power Functions
D40. For variable x taking on the integer values from 1 through 10, sketch
the graph of a power function with a = 1 and b = 2. Compare it to the
exponential equation with a = 1 and b = 2. Discuss the differences between
exponential models and power models.
D41. Use the alligator data to show that you get the same predicted weight for a
given length whether you use base 10 or base e logarithms.
3.5
Shape-Changing Transformations 189
22-03-2009 20:57
Lesson
12 de 24
Power Transformations
You always can use a log-log transformation to straighten points that follow a power
However, sometimes your knowledge of a situation can allow
relationship,
you, with a little thought, to go directly to an appropriate power transformation
without transforming through logarithms. For example, in exploring the
relationship between time in free-fall and the distance an object falls, it is the square
root of the distance that is linearly related to the time. In relating gas mileage to
size of cars, miles per gallon could just as well be gallons per mile (a reciprocal
transformation). Commonly used power transformations are given in the box.
To use a power transformation, transform y by replacing it with a power of y,
such as
The next example describes this process.
Example: A Power Transformation
The success of sustainable harvesting of timber depends on how fast trees grow,
and one way to measure a trees growth rate is to find the relationship between its
diameter and its age. If you know this relationship and the relationship between the
age and height of a tree, then you can estimate the growth rate for total volume of
timber. Displays 3.103 through 3.106 give data on the age (in years) and diameters
(in inches at chest height) of a sample of oak trees, a scatterplot of diameter versus
age, a residual plot, and numerical summaries in the form of computer output.
Based on these data, find a good model for predicting tree diameter from age.
Display 3.103 Diameters and ages of oak trees. [Source: Herman

H. Chapman and Dwight B. Demeritt, Elements of Forest
Mensuration, 2nd ed. (J. B. Lyon Company, 1936).]
22-03-2009 20:57
Lesson
13 de 24
Display 3.104
Diameter versus age of oak trees, with regression

line.
Display 3.105
Display 3.106
Residual plot for the oak tree data.
Numerical summary in the form of computer

output of the ages and diameters (inches at chest
height) of a sample of oak trees.
Solution
Inspection reveals that the point cloud is roughly elliptical with a slight downward
curvature. Straightening this curve will require expanding the y-scale. You can
formulate a power transformation by thinking carefully about the practical
3.5
191
22-03-2009 20:57
Lesson
14 de 24
situation, rather than jumping immediately to the log-log transformation. If

the diameter does not grow linearly with age, perhaps the cross-sectional area,
does. Try squaring the diameters.
Display 3.107 shows a scatterplot of diameter squared versus age, together with
the equation of the least squares line. Display 3.108 shows the plot of residuals
versus age. Even though the value of
is about the same as before, the scatterplot
and residual plot are now less curved. Taking all the evidence together, a linear
model appears to fit the transformed data better than it does the original data.
Display 3.107 Diameter squared versus age for the oak tree data,
with the regression line.
Display 3.108
Measurement scales are

selected by the user, so
select one that meets
your needs.
Residual plot for diameter squared versus age.
Power transformations like the ones you have just seen can straighten a curved
plot (or change a fan shape to a more nearly oval shape). By choosing the right
power, you often can take a data set for which a fitted straight line and correlation
are not suitable and convert it into one for which those summaries work well.
Is it cheating to change the shape of your data? (You wanted a linear cloud,
but you got a curved wedge. You didnt like that, so you fiddled with the data
until you got what you wanted.) In fact, as youll see, changing scale is a matter
of re-expressing the same data, not replacing the data with entirely new facts.
The intelligent measurer selects a scale that is most useful for the problem the
measurements were taken to solve.
[You can use your calculator to perform all the transformations youve learned
about in this section. See Calculator Note 3J.]
22-03-2009 20:57
Lesson
15 de 24
D42. Give a plausible explanation for why a tree
might grow at a rate that makes the
square of its diameter proportional to
its age.
D43. If you need to predict the diameter of some
oak trees for which you know only the age,
would you rather do the predicting for
10-year-old trees or for 40-year-old trees?
Summary 3.5: Shape-Changing

Transformations
For many scatterplots that show slopes or spreads that change as x changes, you
can find a shape-changing transformation that brings your scatterplot much
closer to the form for which the summaries of this chapter work besta cloud
of points in the shape oftan ellipse with a linear trend. A transformation will not
always make the pattern linear, however.
Transformations should have some basis in reality; they should not be simply
chosen arbitrarily to see what might happen. Ideally, the transformation you use
will be related in a plausible way to the situation that created your data.
The most common transformations are powers, in which you replace y with
and so forth, and logarithms, in which you
replace y with log10y or ln y. You can also transform x.
Here are some other helpful facts to consider when choosing a transformation
involving logarithms:
A log-log transformation replacing x with log x and y with log y will straighten
data modeled by a power function of the form y
Replacing y with log y will straighten data modeled by an exponential
function of the form y
If you have data on a quantity that changes over time by an amount roughly
proportional to the quantity at a given time, then the logarithm of the
quantity will be roughly a linear function of time.
Consider a log transformation whenever you have a variable whose values
are clustered at one end and range over two or more orders of magnitude
(powers of 10).
For data collected over time (or over some other sequential ordering, such as
distance along a path), there generally is one data value for each time, and each
data value usually is correlated quite highly with the values to either side (its close
neighbors). So the data will tend to have a much more intricate pattern than can
be modeled by a straight line. Careful analysis of the pattern in the residuals often
can help you see what is really happening over time.
3.5
193
22-03-2009 20:57
Lesson
16 de 24
Practice
P26. Dying dice. One of the authors gathered data
on dying dice, starting with 200, and used
the rule that a die lives if it lands showing
1, 2, 3, or 4. Here are the results:
200 122 81 58 29 19 11 8 6 4 2 2
a. Construct a scatterplot of the number of
"live" dice versus the roll number.
b. Transform the number of live dice using
natural logs, and construct a scatterplot
of ln(dice) versus roll number.
c. Fit a line by the least squares method, and
use its slope to estimate the rate of dying.
d. Plot residuals versus roll number, and
describe the pattern.
P27. Florida is one of the fastest-growing states in
the United States. The population figures for
each census year from 1830 through 2000 are
given in Display 3.109.
Display 3.109 Population of Florida over the years
18302000. [Source: U.S. Census Bureau,
Statistical Abstract of the United States,
20042005.]
a. Find a transformation that straightens the

relationship.
b. Fit a line to the transformed data, and
use it to estimate the rate of population
growth.
c. Produce a residual plot for your model,
and comment on the pattern in the
residuals.
Exponential Functions and Log

Transformations
P28. The logarithm of a number is the exponent
when you write the number in the form
base raised to a power. Thus,
=3
means the same thing as 1000 = . Use
this fact to transform y into logs (in base 10)
in this set of (x, y) pairs: (2, 1000), (1, 100),
(0, 10), and (1, 1).Then plot log10 y versus
x and check that the points lie in a straight
line. Find the slope and y-intercept of the
line.
P29. Repeat P28 for these two sets of pairs.
a. (6, 1000), (4, 100), (2, 10), (0, 1)
b. (5, 0.0001), (6, 0.01), (8, 100)
22-03-2009 20:57
Lesson
17 de 24
P30. Verify that if log10 y = c + dx, then y = abx,

where a = 10c and b = 10d. Use this fact to
rewrite each of your fitted equations in P28
and P29 in the form y = ab x.
P31. Rewrite the fitted equation in Display 3.98
on page 186 in the form
flight length = abspeed
P32. In setting standards for the consumption
of fish tainted by chemicals in the water
from which they were taken, the U.S.
government commissioned a study of fish
consumption for one such contaminated
area. A part of the study involved interviews
with a sample of noncommercial fishers
that asked, among other things, how often
the person fished in this water over the past
month and how many fish meals his or
her family consumed over the past month.
The number of meals was then converted
to grams of fish consumed (a statistical
process in itself).The number of fishing
trips is fairly easy for people to remember,
but they arent so accurate at reporting the
amount of fish their family ate. A good
model relating the number of fishing trips to
consumption would be helpful in estimating
fish consumption. From the set of data in
Display 3.110,find such a model.
Display 3.110 Fish consumption data.
Log-Log Transformations of Power

Functions
P33. Use the regression equation on page 189 to
predict the weight of an alligator that is
75 inches long.
3.5
195
22-03-2009 20:57
Lesson
18 de 24
P34. Having a good measure of tidal velocity (the

speed at which water depth increases) in
an estuary is critically important, especially
during storms. Tidal velocity is difficult
to measure, but it is related to the depth
of the water. Thus, a good model of this
relationship would allow scientists to predict
the velocity from measurements of water
depth. Display 3.111 shows measurements
of the depth of water (in meters) and tidal
velocities (in meters per second) for certain
locations in an estuary.
P35. Three sets of pairs (x, y) are given here. For
each set, (i) plot y versus x, (ii) find a power
of y to use in place of y itself to get a linear
plot, and (iii) plot that power of y versus x.
a. (0, 0), (1, 1), (2, 8), (3, 27)
b. (1, 10), (2, 5), (5, 2), (10, 1)
c. (100, 10), (64, 8), (25, 5), (9, 3), (1, 1)
P36. In P35, suppose that, instead of plotting
a power of y versus x, you plot y versus a
power of x. For each of the three sets of
pairs (x, y), find the power of x for which the
points lie on a line. What is the relationship
between the powers of y in P35 and the
powers of x in this problem?
P37. For each of these relationships, first write the
equation that relates x and y. Then use this
equation to find a power of y that you could
plot against x in order to get a linear plot.
a. y is the area of a circle, and x is the radius
of the circle.
b. y is the volume of a block whose sides
all have equal lengths, and x is the side
length.
Display 3.111 Tidal velocity versus the water depth.
[Source: Shi et al., International Journal of

Numerical Methods for Fluids, 2003.]
a. Describe the nature of the relationship

between depth and velocity.
b. Fit an appropriate model that would
allow the prediction of velocity from
depth.
c. y is the volume of an 8-ft section of log

with a circular cross section, and x is the
diameter of the logs cross section.
P38. If the square of a trees diameter is roughly
proportional to its age, then you can
expect the diameter itself to be roughly
proportional to the square root of the
trees age. For the data in Display 3.103
on page 190, use a computer or calculator
to make a scatterplot of diameter versus
square root of age, fit a least squares line,
and plot the residuals. Which residual plot
shows more of a fan shape, the plot of
diameter squared versus age or the plot of
diameter versus square root of age? If you
want a plot that shows an elliptical cloud,
which transformation should you choose?
P39. Display 3.112 shows the brain weights and
body weights for a collection of mammals.
The goal is to establish the relationship of
brain weight to body weight.
22-03-2009 20:57
Lesson
19 de 24
a. Assuming that there is a power

relationship here, can you guess what it
is from the scatterplot? If y is written as a
function of x to some power, should the
power be greater than 1 or less than 1?
b. Plot log(brain) versus log(body). Describe

the pattern of the plot.
c. Fit a line to the plot in part b. Write
an equation relating y to x. Does your
equation support your answer to part a?
Display 3.112 The brain weights and body weights of mammals. [Source: T. Allison and
D. V. Cicchetti, Sleep in Mammals: Ecological and Constitutional Correlates, Science 194
(1976): 73234.]
3.5
197
22-03-2009 20:57
Lesson
20 de 24
Exercises
E55. For the data in Display 3.97 on page 185, try
fitting a straight line to the square root
of flight length as a function of speed.
Does this transformation work as well
as the log transformation? Explain your
reasoning.
E56. More dying dice. Follow the same steps as
in P26 on page 194 for these numbers of
surviving dice: 200, 72, 28, 9, 5, 2, and 1. Use
your data to estimate what the probability
of dying was in order to generate these
numbers.
E57. Growing kids. Median heights and weights
of growing boys are presented in Display
3.113. What model would you choose to
predict weight from a boys known height?
Look through the complete listing of the

data (Display 3.12 on page 115) and the
scatterplots from P25 to see whether any
features other than ight length separate the
aircra into the same two groups.
E59. Chimp hunting parties. After Jane Goodall
discovered that chimpanzees are not solely
vegetarian, much research began into the
behavior of chimpanzees as hunters. Some
animals hunt alone or in small groups, while
others hunt in large groups. Where does the
chimp fit in, and what is the success rate of
chimps hunting parties? Not surprisingly,
the success of the hunt depends in part on
the size of the hunting party. Display 3.114
gives some data on the number of chimps
in a hunting party and the success rate of
parties of that size.
Display 3.113 Median heights and weights of

growing boys. [Source: National Health
and Nutrition Examination Survey (NHANES),
2002, www.cdc.gov.]
E58. Cost per seat per mile and flight length,

revisited. As you saw in P25 on page 174,
when cost per seat per mile is plotted against
flight length, the pattern is not linear. The
residual plot in Display 3.77 on page 174
strongly suggests that two line segments
might provide a better model than a single
line. Apparently, there is one relationship
for aircra meant for longer routes and
another for aircra meant for shorter routes.
Display 3.114 Hunting party size and percentage
of success.
[Sources: Mathematics Teacher,

August 2005, p. 13; C. B. Stanford, Chimpanzee
Hunting Behavior and Human Evolution,
American Scientist 83 (1995).]
a. Plot the data in a way that allows the

building of a model to predict success
from size of hunting party. Describe the
pattern you see.
22-03-2009 20:57
Lesson
21 de 24
b. Will a simple linear regression model

of the major trend(s) in U.S. population
growth.
work well here? Why or why not?
c. Make a plot over time of the immigration
c. Look for a transformation that will
by decade. Describe the pattern you see
produce a model with better predicting
here. Can you fit one of the models from
ability than the simple linear one. Fit the
this section to data that look like this?
model to the data.
d. Investigate the residuals from the model E61. Display 3.116 gives data about passengers on
United Airlines flight 815, Chicago-OHare
in part c. Are you happy with the fit of
that model?
to Los Angeles, on October 31, 1997. There
were 186 passengers, but the data concern
E60. The data in Display 3.115 are the population
those 33 passengers who had tickets for
of the United States from 1830 through 2000
the Chicago-to-Los Angeles leg only. The
and the number of immigrants entering the
variables are
country in the decade preceding the given
year.
X: number of days before the fiight that the
ticket was purchased
Y: price of the ticket
Display 3.115 Population and immigration in the

United States, 18302000. [Source: U.S.
Census Bureau, Statistical Abstract of the
United States, 20042005.]
a. Find the population growth for each

decade. Was the increase in population
constant from decade to decade? How
would you describe the pattern?
b. Fit a model to the (year, population) data
and defend your model as representative
Display 3.116
Number of days before the fl ight that

the ticket was purchased and price
of airline ticket. [Source: New York Times,
Weekly Review, April 12, 1998.]
Because it is in the airlines interest to sell

tickets early, you might expect Y to be
negatively associated with X.
It happens that the first four cases in Display
3.116 are for passengers who first class,
and those passengers pay more than other
199
22-03-2009 20:57
Lesson
22 de 24
passengers no matter when they purchase

on oxygen consumption is that the rates
their ticket. So you can justify examining the
of consumption within organ tissue can
data for the 29 economy-class passengers
be explained largely by the relative size of
only.
the organ within the body. Is this theory
supported by the data? Explain your
Finally, one passenger paid $0 because he
reasoning.
or she used frequent-flyer miles. You are
justi ed in deleting this value from the data
set if the goal is to find a model that relates
price to time of purchase.
Can you find a model that relates the cost of
the ticket to the number of days in advance
that the ticket was purchased? Explain the
problems you encounter in doing this.
E62. Different body organs use different amounts
of oxygen, even when you take their mass
into consideration. For example, the brain
uses more oxygen per kilogram of tissue
than the lungs do. Scientists are interested
in how oxygen consumption is related to
Display 3.117 Oxygen use by certain animal organs.
the mass of an animal and whether that
The oxygen measurements are coded
values (original measurements not
relationship differs from organ to organ.
given) but are still comparable.
The data in Display 3.117 show typical body
[Source: K. Schmidt-Nielsen, Why Is Animal
mass, oxygen consumption in brain tissue,
Size So Important? (Cambridge University Press,
and oxygen consumption in lung tissue for a
1984), p. 94.]
selection of animals. (Oxygen consumption
often is measured in milliliters per hour per E63. How is the birthrate of countries related to
their economic output? Do richer countries
gram of tissue, but the actual units were not
have higher birthrates, perhaps because
recorded for these data.)
families can a ord more children? Or do
a. As you can see from the table, as
poorer countries have higher birthrates,
total body mass increases, the oxygen
perhaps due to the need for family workers
consumption in brain tissue tends to go
and a lack of education? Display 3.118
down. Define a function that models this
shows the birthrates (number of births
situation. Then find a way to describe the
per
thousand population) and the GNP
rate of decrease.
(in thousands of dollars per capita) for
b. Repeat part a for lung tissue. How does
a selection of countries from around the
this relationship differ from that of brain
world.
tissue?
a. Construct a scatterplot of these data and
c. It is known that the proportion of
comment on the pattern you observe.
body mass concentrated in the brain
b.
Fit a statistical model to these data and
decreases appreciably as the size of the
interpret
the slope and intercept of the
animal increases, whereas the proportion
model
in
the context of the data.
concentrated in the lungs remains
relatively constant. One possible theory
22-03-2009 20:57
Lesson
23 de 24
b. Study the trend in the percentage of

smokers for the group age 18 to 24. What
model would you use to explain the
relationship between the percentage of
smokers and the year for this age group?
Explain your reasoning. What feature
makes this data set more difficult to
model than the data set in part a?
c. Study the trend in the percentage of
smokers for the group age 65 and over.
Does this group show the same kind of
trend as seen in the two groups studied in
parts a and b? Explain.
Display 3.118
Birthrates and GNP for selected

countries, 2002. [Source: U.S. Census
20042005.]
E64. According to the National Center for Health

Statistics, the percentage of males who
smoke has decreased markedly over the
past 40 years, but there still may be some
interesting trends to observe. Display 3.119
shows the percentage of males who smoke in
selected years for various age groups.
a. Study the trend in the percentage of
smokers for the entire male population
age 18 and over. The points follow the
pattern of exponential decay. How should
you modify the percentages before taking
their logarithms? Fit the model and
interpret the slope.
Display 3.119
Percentage of males who smoke by

age group and year. [Source: National
Center for Health Statistics, 2003.]
E65. Is global warming a reality? One measure

of global warming is the amount of carbon
dioxide (CO2) in the atmosphere. Display
3.120 gives the annual average carbon
dioxide levels ( in parts per million ) in the
atmosphere over Mauna Loa Observatory in
Hawaii for the years 1959 through 2003.
a. Plot the data and describe the trend over
the years.
201
22-03-2009 20:57
Lesson
24 de 24
b. Fit a straight line to the data and look at

the residuals. Describe the pattern you
see.
c. Suggest another model that might fit
these data well. Fit the model and assess
how well it removes the pattern from the
residuals.
d. Use the model you like best to describe
numerically the growth rate in
atmospheric carbon dioxide over Hawaii.
Display 3.120 Carbon dioxide in the atmosphere.

[Source: Mauna Loa Observatory.]
E66. How does the average SAT math score for

students in a state relate to the percentage
of students taking the exam? Display 3.121
shows the average SAT math score for each
state in 2005, along with the percentage of
high school seniors taking the exam. Find
a model that seems like a good predictor
of average SAT math scores based on
knowledge of the percentage of seniors
taking the exam.
Display 3.121 Average SAT math scores by state.

[Source: College Board, www.collegeboard.com.]
22-03-2009 20:57
Lesson
1 de 13
Chapter Summary
In Chapter 2, you worked with univariate data. In Chapter 3, you learned how to
uncover information for bivariate (two-variable) data, using plots and numerical
summaries of center and spread. In Chapter 2, the basic plot was the histogram.
For histograms of distributions that are approximately normal, the fundamental
measures of center and spread are the mean and the standard deviation. For
bivariate data, the basic plot is the scatterplot. For scatterplots that have an
elliptical shape, the fundamental summary measures are the regression line
(which you can think of as the measure of center) and the correlation (which
you can think of as the measure of spread).
For now, correlation and regression merely describe your data set. In
Chapter 11, you will learn to use numerical summaries computed from a sample
to make inferences about a larger population. Using diagnostic tools such as
residual plots and finding transformations that re-express a curved relationship as
a linear one will come in especially handy because you wont be able to make valid
inferences unless the points form an elliptical cloud.
Review Exercises
E67. Leonardos rules. A class of 15 students
recorded the measurements in Display 3.122
for Activity 3.3a.
Display 3.122 Sample measurements, in

centimeters, for Activity 3.3a.
a. Construct scatterplots and fit least

squares lines for each of Leonardos rules
in Activity 3.3a. Do the rules appear to
hold?
b. Interpret the slopes of your regression
lines.
c. If appropriate, find the value of r for
each of the three relationships. Which
correlation is strongest? Which is
weakest?
E68. Space Shuttle Challenger. On January 28,
1986, because two O-rings did not seal
properly, Space Shuttle Challenger exploded
and seven people died. The temperature
predicted for the morning of the flight was
between 26F and 29F. The engineers were
concerned that the cold temperatures would
cause the rubber O-rings to malfunction.
On seven previous flights at least one of the
twelve O-rings had shown some distress.
The NASA officials and engineers who
decided not to delay the flight had available
to them data like those on the scatterplot
in Display 3.123 before they made that
decision.
Chapter Summary
203
22-03-2009 20:58
Lesson
2 de 13
E69. Exam scores. Students scores on two exams

in a statistics course are given in Display
3.125 along with a scatterplot with regression
line and a residual plot. The regression
equation is Exam 2 = 51.0 + 0.430(Exam 1),
and the correlation, r, is 0.756.
Display 3.123 Flights when at least one O-ring

showed some distress.
a. Why did it seem reasonable to launch

despite the low temperature?
b. Display 3.123 contains information only
about flights that had O-ring failures.
Data for all flights were available on
a table like the one in Display 3.124.
Add the missing points to a copy of the
scatterplot in Display 3.123. How do these
data affect any trend in the scatterplot?
Would you have recommended launching
the space shuttle if you had seen the
complete plot? Why or why not?
Display 3.125 Data for exam scores in a statistics

class, with scatterplot and residual
plot.
Display 3.124 Challenger O-ring data. [Source:

Siddhartha R. Dalal et al., Risk Analysis of
the Space Shuttle: Pre-Challenger Prediction
of Failure, Journal of the American Statistical
Association 84 (1989): 94547.]
a. Is there a point that is more influential

than the other points on the slope of the
regression line? How can you tell from
the scatterplot? From the residual plot?
b. How will the slope change if the
scores for this one influential point are
removed from the data set? How will the
correlation change? Calculate the slope
and correlation for the revised data to
check your estimate.
22-03-2009 20:58
Lesson
3 de 13
c. Construct a residual plot of the revised

data. Does a linear model fit the data
well?
d. Refer to the scatterplot of Exam 2 versus
Exam 1 in Display 3.125. Does this plot
illustrate regression to the mean? Explain
your reasoning.
E70. Suppose you have the Exam 1 and Exam 2
scores of all students enrolled in U.S. History.
a. The slope of the regression line for
predicting the scores on Exam 2 from the
scores on Exam 1 is 0.51. The standard
deviation for Exam 1 scores is 11.6,
and the standard deviation for Exam 2
scores is 7.0. Use only this information to
find the correlation coefficient for these
scores.
b. Suppose you know, in addition, that the
means are 82.3 for Exam 1 and 87.8 for
Exam 2. Find the equation of the least
squares line for predicting Exam 2 scores
from Exam 1 scores.
E71. You are given a list of six values, 1.5, 0.5,
0, 0, 0.5, and 1.5, for x and the same list of six
values for y. Note that the list has mean 0 and
standard deviation 1.
a. Match each x-value with a y-value so
that the resulting six pairs ( x, y ) have
correlation 1.
Display 3.126
b. Match the x- and y-values again so that

the points have the largest possible
correlation less than 1.
c. Match the values again, this time to get a
correlation as close to 0 as possible.
d. Match the values a fourth time to get a
correlation of 1.
E72. Display 3.126 lists the values of six variables,
with a scatterplot matrix showing all 30
possible scatterplots for these variables. For
example, the first scatterplot in the first row
has variable B on the x-axis and variable
A on the y-axis. The first scatterplot in the
second row has variable A on the x-axis and
variable B on the y-axis.
a. For five pairs of variables the correlation
is exactly 0, and for one other pair it
is 0.02, or almost 0. Identify these six
pairs of variables. What do they have in
common?
b. At the other extreme, one pair of
variables has correlation 0.87; the next
highest correlation is 0.58, and the third
highest is 0.45. Identify these three pairs,
and put them in order from strongest to
weakest correlation.
c. Of the remaining six pairs, four have
correlations of about 0.25 ( give or take a
little ) and two have correlations of about
Data table for 6 variables and a scatterplot matrix of all 30 possible

scatterplots for the variables.
Chapter Summary
205
22-03-2009 20:58
Lesson
4 de 13
0.1 (give or take a little). Which four pairs

have correlations around 0.25?
d. Choose several scatterplots that you think
best illustrate the phrase the correlation
measures direction and strength but not
shape, and use them to show what you
mean.
E73. Decide whether each statement is true or
false, and then explain your decision.
a. The correlation is to bivariate data what
the standard deviation is to univariate
data.
b. The correlation measures direction and
strength but not shape.
c. If the correlation is near 0, knowing the
value of one variable gives you a narrow
interval of likely values for the other
variable.
d. No matter what data set you look at, the
correlation coefficient, r, and least squares
slope, b1, will always have the same sign.
E74. Look at the scatterplot of average SAT I math
scores versus the percentage of students
taking the exam in Display 3.7 on page 112.
a. Estimate the correlation.
b. What possibly important features of
the plot are lost if you give only the
correlation and the equation of the least
squares line?
c. Sketch what you think the residual plot
would look like if you fitted one line to all
the points.
E75. The correlation between in-state tuition and
out-of-state tuition, measured in dollars, for
a sample of public universities is 0.80.
a. Rewrite the sentence above so that
someone who does not know statistics
can understand it.
b. Does the correlation change if you
convert tuition costs to thousands of
dollars and recompute the correlation?
Does it change if you take logarithms
of the tuition costs and recompute the

correlation?
c. Does the slope of the least squares line
change if you convert tuition costs to
thousands of dollars and recompute
the slope? Does it change if you take
logarithms of the tuition costs and
recompute the slope?
E76. Display 3.127 shows a scatterplot divided
into quadrants by vertical and horizontal
lines that pass through the point of averages,
( x, y ).
a. For each of the four quadrants, give the
sign of zx (the standardized value of x),
zy (the standardized value of y), and their
product zx zy.
b. Which point(s) make the smallest
contribution to the correlation? Explain
why the contributions are small.
Display 3.127
A scatterplot divided into quadrants

by lines passing through the means.
E77. Rank these summaries for three sets

of bivariate data by the strength of the
relationship, from weakest to strongest.
22-03-2009 20:58
Lesson
5 de 13
E78. Theres an extremely strong relationship

between the price of books online and the
price at your local bookstore.
a. Does this mean the prices are almost the
same?
b. Explain why it is wrong to say that the
prices online cause the prices at your
local bookstore. Why is the relationship
so strong if neither set of prices causes the
other?
E79. Describe a set of cases and two variables for
which you would expect to see regression
toward the mean.
E80. Life spans. In Chapter 2, you looked at
the characteristics of mammals ( given in
Display 2.24 on page 43 ) one at a time. Now
you can look at the relationship between
two variables. For example, is longevity
associated with gestation period? The
variables are average longevity in years,
maximum longevity in years, gestation
period in days, and speed in miles per hour.
a. Construct a scatterplot of gestation
period versus maximum longevity.
Describe what you see, including an
estimate of the correlation.
Display 3.128
b. Repeat part a, with average longevity

in place of maximum longevity. Does
the average longevity or the maximum
longevity give a better prediction of the
gestation period?
c. Does speed appear to be associated with
average longevity?
E81. Spending for police. The data in Display
3.128 give the number of police officers, the
total expenditures for police officers, the
population, and the violent crime rate for a
sample of states in 2000.
a. Explore and summarize the relationship
between the number of police officers and
total expenditures for police.
b. Explore and summarize the relationship
between the population of the states
and the number of police officers they
employ.
c. Is the number of police officers
strongly related to the rate of violent
crime in these states? Explain. Find
a transformation that straightens
these data. Check the linearity of your
transformed data with a residual plot.
Number of police officers and related variables. [Source: U.S. Census Bureau,
Statistical Abstract of the United States, 20042005.]
Chapter Summary
207
22-03-2009 20:58
Lesson
6 de 13
E82. House prices. Display 3.129 gives the selling

prices for all houses sold in a Florida
community in one month.
a. Construct a model to predict the selling
price from the area, transforming any
variables, if necessary. Would you use
the same model for both new and used
houses?
b. Are there any influential observations
that have a serious effect on the model? If
so, what would happen to the slope of the
prediction equation and the correlation if
you removed this (or these) point(s) from
the analysis?
c. Predict the selling price of an old house

measuring 1000 sq ft. Do the same for an
old house measuring 2000 sq ft. Which
prediction do you feel more confident
about? Explain.
d. Explain the effect of the number of
bathrooms on the selling price of the
houses. Is it appropriate to fit a regression
model to price as a function of the
number of bathrooms and interpret
the results in the usual way? Why or
why not?
Display 3.129
22-03-2009 20:58
Lesson
7 de 13
Display 3.129
Chapter Summary
209
22-03-2009 20:58
Lesson
8 de 13
Display 3.129
Selling prices of houses in Gainesville, Florida. [Source: Gainesville Board of

Realtors, 1995.]
E83. Spending for schools. Display 3.130 provides

data on spending and other variables related
to public school education for 2001. The
variables are defined as
ExpPP
expenditure per pupil (in
dollars)
ExpPC
expenditure per capita (per
person in the state, in dollars)
average teacher salary (in
TeaSal
thousands of dollars)
%Dropout percentage who drop out of
school
Enroll
number of students enrolled
(in thousands)
Teachers number of teachers (in
thousands)
a. Examine the association between perpupil expenditure and average teacher

salary, with the goal of predicting perpupil expenditure. Is this a cause-andeffect relationship?
b. Analyze the effect of average teacher
salary on per-capita expenditure
( spending on public schools divided
by the number of people in the state ).
Compare the association to the
association in part a. Are the relative
sizes of the correlations about what you
would expect?
c. Are any variables good predictors of the
percentage of dropouts? Explain your
reasoning.
Display 3.130
22-03-2009 20:58
Lesson
9 de 13
Display 3.130 Public school education by state in 2001. [Source: U.S. Census Bureau, Statistical
Abstract of the United States, 20042005.]
Chapter Summary
211
22-03-2009 20:58
Lesson
10 de 13
AP1. This scatterplot shows the age in years

of the oldest and youngest child in 116
households with two or more children age
18 or younger living with their parents.
(Some points have been moved slightly to
show that multiple households are at each
coordinate.) Which of the following is not a
reasonable interpretation of this scatterplot?
There arent any points in the upper-left

region because older children tend
to move out to go to college or to get
married.
The older the oldest child in a household,
the older the youngest child tends to be.
There are no households represented
here in which the only children are twins.
The variability in the age of the youngest
child in these households tends to
increase with the age of the oldest child.
Few households have a range of more
than 12 years in the ages of all of the
children in the household.
AP2. In a study of 190 nations, the least squares
line for the relationship between birthrate
(per thousand per year) and female literacy
rate (in percent) is birthrate = 0.38
literacy + 53.5, with r = 0.8. Uganda has a
birthrate of 47 and a female literacy rate of
60. What is the residual for Uganda?
17.1
64.1
29.3
69.8
16.3
AP3. In a linear regression of the heights of a

group of trees versus their circumferences,
the pattern of residuals is U-shaped. Which
of the following must be true?
I. A nonlinear regression would be a
better model.
II. For trees near the middle of the range
of tree circumferences studied, the
predicted tree height tends to be too tall.
III. r will be close to 0.
II only
III only
I and II
I and III
I, II, and III.
AP4. A recent study models the relationship
between the number of teachers at a high
school and the number of sick days these
teachers take in a year. This scatterplot
shows data for all high schools in a county
during one year.
Which of the following is not a reasonable

way to proceed with the analysis?
Remove the four outliers permanently
from the data set.
Run the regression again without the
four outliers to see how much the slope
and correlation change.
Verify the values for the four outliers to
make sure they are correct.
Try to find a transformation that makes
the cloud of points more elliptical.
Make a residual plot in order to judge
the linearity of the points.
22-03-2009 20:58
Lesson
11 de 13
AP5. Upon checking out of a large hospital,

2000 patients rated their satisfaction
with their stay on a scale of 010, with
10 indicating complete satisfaction. The
relationship between the satisfaction rating
and the patients length of stay (in days) was
analyzed using linear regression. Here is part
of the computer printout for this regression.
Which is a correct interpretation of this

regression?
Patients who stay longer in the hospital
tend to be more satisfied than patients
with shorter stays, although this
relationship is weak.
The correlation is 0.228. However,
the sign on the correlation cannot be
determined.
The value of R2 indicates that the
relationship between satisfaction rating
and length of stay is weak but linear.
The y-intercept of the regression line
indicates that no patients rated their
satisfaction less than 4.
The slope of the regression line indicates
that as a patient stays longer in the
hospital, his or her satisfaction tends
to increase day by day.
AP6. The least squares equation to estimate
the population of the fictional country of
Barbaria is log10( population ) = 0.01t + 7,
where t is the number of years since 1950.
Which of the following is closest to the
predicted population of Barbaria for the
year 2000?
7.5
27
7,500,000
31,600,000
1,000,000,000,000,000,000,000,000,000
AP7. The Barbarian Aptitude Test (BAT) gives

each Barbarian two scores, one for pillaging
and one for burning. The scores range from
a low of 0 to a high of 50. The least squares
equation for a large group of Barbarians who
took the BAT is burning = 0.3 pillaging + 19,
with r = 0.6. Which is the best interpretation
of the slope of this line?
A Barbarian who studies harder and
improves her pillaging score by 1 point
on the next BAT will tend to increase her
burning score by about 0.3 point as well.
Barbarians tend to score about 0.3 point
higher on burning than on pillaging.
Barbarians score about 30% as many
points on burning as on pillaging.
The burning score is highly correlated
with the pillaging score.
A Barbarian who earned 1 more point
on pillaging than another Barbarian
tended to earn only 0.3 point more on
burning.
AP8. A least squares regression analysis using
a rating of each Barbarians personal
cleanliness as the explanatory variable and
the number of raids he or she has carried
out as the response variable found a positive
relationship with R2 = 0.81. Which is not a
correct interpretation of this information?
The correlation between personal
cleanliness and the number of raids is 0.9.
There is a strong relationship between
personal cleanliness and number of raids
among Barbarians.
A Barbarian who is more personally
clean than another also tends to have
made more raids.
There is an 81% chance that the
relationship between personal
cleanliness and number of raids is linear.
The variation in the residuals for the
number of raids among Barbarians
is about 19% of the variation in the
original responses.
AP Sample Test
213
22-03-2009 20:58
Lesson
12 de 13
Investigative Tasks
AP9. Siris equation. Athletes and exercise
scientists sometimes use the proportion
of fat to overall body mass as one measure
of fittness, but measuring the percentage
of body fat directly poses a challenge.
Fortunately, some good statistical detective
work by W. E. Siri in the 1950s provided
an alternative to direct measurement that
is still in use today. Siris method lets you
estimate the percentage of body fat from
body density, which you can measure
directly by hydrostatic (underwater)
weighing. In this exercise, youll see how
transformations and residual plots play a
crucial role in finding Siris model.
a. A first model. Use the data in Display
3.131.
Display 3.131 The percentage of body fat and body

density of 15 women. [Source: M. L. Pollock,
University of Florida, 1956.]
i. Plot percentage of body fat versus

body density, and from your plot
explain whether you think a line gives
a poor fit, a moderately good fit, or an
extremely good fit.
ii. Write the equation of the least squares
line.
iii. Does the value of r2 tend to con rm

your opinion about how well the line
fits?
iv. Construct a residual plot and describe
the pattern. Does the plot tend to
confirm or raise questions about your
opinion? Explain.
b. A new model.
i. Explain how knowing that fat is less
dense than the rest of the body might
have led Siri to plot the percentage of
body fat against the reciprocal
of density,
Construct this
plot and fit a least squares line, and
compare its equation to the one Siri
found:
% body fat = 450 + 495
Next, plot residuals versus
What features of this plot confirm that
the transformation has improved the
linear fit?
ii. Find the correlation between the
percentage of body fat and body
density and the correlation between
the percentage of body fat and the
reciprocal of body density. Comment
on using correlation as the only
criterion for assessing the usefulness
of a model.
iii. Suggest another model for the
percentage of body fat and body
density data that might work nearly
as well as Siris.
22-03-2009 20:58
Lesson
13 de 13
AP10. In AP9, you explored the relationship

between the percentage of body mass that
is fat and body density. Display 3.132 is
an extension of the data in AP9, including
skinfold measurements and data for men.
a. Does Siris model for relating percentage
of body fat to body density hold for men
as well as it did for women? That model
was
% body fat = 450 + 495
b. The variable skinfold is the sum of a
number of skinfold thicknesses taken
at various places on the body. ( The
units are millimeters.) The skinfold
measurements are used to predict
body density. Find a good model
for predicting density from skinfold
measurements based on these data
for women. Do your models require
re-expression?
Display 3.132 Percentage of body fat, body density,

and skinfold for 15 women and
14 men. [Source: M. L. Pollock, University
of Florida, 1956.]
AP Sample Test
215
22-03-2009 20:58
Lesson
1 de 3
Sample Surveys
and Experiments
What prompts a
hamster to prepare
for hibernation? A
student designed an
experiment to see
whether the number
of hours of light in
a day affects the
concentration of a key
brain enzyme.
22-03-2009 20:58
Lesson
2 de 3
Most of what youve done in Chapters 2 and 3, as well as in the first part of
Chapter 1, is part of data explorationways to uncover, display, and describe
patterns in data. Methods of exploration can help you look for patterns in just
about any set of data, but they cant take you beyond the data in hand. With
exploration, what you see is all you get. Often, thats not enough.
Pollster: I asked a hundred likely voters who they planned to vote for, and
fifty-two of them said theyd vote for you.
Politician: Does that mean Ill win the election?
Pollster: Sorry, I cant tell you. My stat course hasnt gotten to inference yet.
Politician: Whats inference?
Pollster: Drawing conclusions based on your data. I can tell you about the
hundred people I actually talked to, but I dont yet know how to
use that information to tell you about all the likely voters.
Methods of inference can take you beyond the data you actually have, but
only if your numbers come from the right kind of process. If you want to use
100 likely voters to tell you about all likely voters, how you choose those 100
voters is crucial. The quality of your inference depends on the quality of your
data; in other words, bad data lead to bad conclusions. This chapter tells you
how to gather data through surveys and experiments in ways that make sound
conclusions possible.
Here's a simple example.1 When you taste a spoonful of chicken soup and
decide it doesnt taste salty enough, thats exploratory analysis: Youve found a
pattern in your one spoonful of soup. If you generalize and conclude that the
whole pot of soup needs salt, thats an inference. To know whether your inference
is valid, you have to know how your one spoonfulthe datawas taken from
the pot. If a lot of salt is sitting on the bottom, soup from the surface wont be
representative, and youll end up with an incorrect inference. If you stir the soup
thoroughly before you taste, your spoonful of data will more likely represent the
whole pot. Sound methods for producing data are the statisticians way of making
sure the soup gets stirred so that a single spoonfulthe samplecan tell you
about the whole pot. Instead of using a spoon, the statistician relies on a chance
device to do the stirring and on probability theory to make the inference.
Soup tasting illustrates one kind of question you can answer using statistical
methods: Can I generalize from a small sample (the spoonful) to a larger
population (the whole pot of soup)? To use a sample for inference about a
population, you must randomize, that is, use chance to determine who or what
gets into your sample.
1The inspiration for this metaphor came from Gudmund Iversen,who teaches statistics at Swarthmore College.
22-03-2009 20:58
Lesson
3 de 3
The other kind of question is about comparison and cause. For example,
if people eat chicken soup when they get a cold, will this cause the cold to go
away more quickly? When designing an experiment to determine if a pattern in
the data is due to cause and effect, you also must randomize. That is, you must
use chance to determine which subjects get which treatments. To answer the
question about chicken soup, you would use chance to decide which of your
subjects eat chicken soup and which dont, and then compare the duration of
their colds.
The first part of this chapter is about designing surveys. A well-designed
survey enables you to make inferences about a population by looking at a sample
from that population. The second part of the chapter introduces experiments. An
experiment enables you to determine cause by comparing the effects of two or
more treatments.
In this chapter, you will learn
reasons for using samples when conducting a survey
how to design a survey by randomly selecting participants
how surveys can go wrong (bias)
how to design a sound experiment by randomly assigning treatments to

subjects
how experiments can determine cause
how experiments can go wrong (confounding)
22-03-2009 20:58
Lesson
1 de 12
4.1
Why samples are
necessary
Sampling saves money

and time.
Testing every unit can be

destructive.
Samples can give more
information than can
a cursory study of all
individuals.
Why Take Samples, and How Not To

Taking a sample survey can help you determine the percentage of people in a
population who have a particular characteristic. For example, the Gallup poll
periodically asks adults in the United States questions such as whether they
approve of the job the president is doing. Polls or surveys such as this rely on
samples to get their percentages; that is, they dont ask every adult in the United
States but instead ask only a sample of about 1500. Similarly, quality assurance
methods in a manufacturing plant do not call for checking the quality of every
item coming off the production line; rather, they recommend that a limited
number of items (a sample) be checked carefully for quality.
Cost in money or time is a primary reason to use samples. Imagine: Its
Sunday night at 8:00, and Nielsen Media Research is gathering data about what
proportion of TV sets are tuned to a particular program and what kinds of
people are watching that program. To find out how many TV sets are tuned to
the program, electronic meters have been hooked up to televisions in a sample
of households. To find out who is watching it, a sample of people are filling out
diaries. Why doesnt Nielsen Media Research include everyone in the United
States in these surveys? To hook up a meter to every TV set would cost more than
anyone would be willing to pay for the information. Also, to try to get a diary
from every TV viewer about what they were watching at 8:00 p.m. on Sunday
would take so much time that the information would no longer be very useful.
So for two reasonsmoney and timeNielsen ratings are based on samples.
Sampling lets a cook know how the soup will taste without eating it all just to
make sure. A light bulb manufacturer cant test the life of every bulb produced,
or there would be none left to sell. Whenever testing destroys the things you test,
your only choice is to work with a sample.
If time and money are limitedand they always aretheres a tradeoff
between the number of people in your sample and the amount and quality of
information you can expect to get from each person. Using a sample allows you
to spend more time and money gathering high-quality information from each
individual. This often produces greater accuracy in the results than you could get
from a quick, but error-prone, study of every individual.
Census Versus Sample

In statistics, the set of people or things that you want to know about is called
the population. The individual elements of the population sometimes are called
units. In everyday language, population often refers to the number of units in
the set, as when you say In 1990, the population of Massachusetts was about
6 million. In statistics, population refers to the set itself (for example, the people
of Massachusetts). The number of units is called the population size. Ordinarily,
you dont get to record data on all the units in the population, so you use a sample.
The sample is the set of units you do get to study. The special case where you
collect data on the entire population is a census.
Nielsen Media Research takes a survey so they can get an estimate of the
proportion of all U.S. households that are tuned to a particular television program.
219
22-03-2009 20:59
Lesson
2 de 12
S is for statistic and

sample: A statistic is a
numerical summary
of the sample. P is
for parameter and
population: A parameter
is a numerical summary
of the population.
The true proportion that Nielsen would get from a survey of every household is
called a population parameter. Nielsen uses the proportion in the sample as an
estimate of this parameter. Such an estimate from a sample is called a statistic.
Not all statistics are created equal. Some arent very good estimators of the
population parameter. For example, the maximum in the sample isnt a very good
estimator of the maximum in the populationit is almost always too small. You
will learn more about the properties of estimators in Chapter 7.

D1. In which of these situations do you think a census is used to collect data, and
in which do you think sampling is used? Explain your reasoning.
a. An automobile manufacturer inspects its new models.
b. A cookie producer checks the number of chocolate chips per cookie.
c. The U.S. president is determined by an election.
d. Weekly movie attendance figures are released each Sunday.
e. A Los Angeles study does in-depth interviews with teachers in order
to find connections between nutrition and health.
Bias: A Potential Problem with Survey Data

Samples offer many advantages, but some samples are more trustworthy than
others. In this section, you will learn about two ways to get untrustworthy results:
bias in the way you select your sample
bias in the way you get a response from the units in your sample
In Activity 4.1a, youll examine the kind of problem that can lie in wait for
the unwary sampler. Suppose you have just won a contract to estimate the average
length of stay in the childrens ward of a hospital. How will you gather the data?
Time in the Hospital
What youll need: one deck of cards for you and your partner
In this activity, youll estimate the average length of stay in a five-bed hospital
ward. Youll sample from the population lengths of stay, represented by the
numbers on the 40 cards (not counting face cards) in an ordinary deck of
cards. Youll estimate the average length of stay from a sample of five patients.
1. Shuffle your 40 cards several times. Deal out a row of five cards to represent
your first patients. The numbers on the cards tell you how many days they
will be on your ward. For example, suppose your first five cards are
These cards mean the patient in bed 1 will be in the hospital for 1 day, the
patient in bed 2 will be there for 2 days, the patient in bed 3 will be there
for 8 days, and so forth. Your partner should record this information on a
(continued)
220 Chapter 4 Sample Surveys and Experiments
22-03-2009 20:59
Lesson
3 de 12
chart like the one in Display 4.1. Place the cards that represent patients in a
stack separate from unused cards.
Display 4.1 Lengths of stay for the first five patients.
2. Deal out the other cards one at a time, assigning the next patient to a bed
as it becomes available. To continue with the example, suppose the next
patient is 9 . First available bed is bed 1, and the patient will be in it
for 9 days. The next available bed is bed 2, and the next patient, say, 9 , will
be there for 9 days. Display 4.2 shows a chart of the lengths of stays at this
stage. The next patient will go into bed 5.
Display 4.2 Lengths of stay for the first seven patients.
3. Continue dealing out patient cards, each representing a hospital stay. Record
these stays until all five beds have been filled for at least 20 days. You should
end up with something like Display 4.3. (Save your chart; youll need it later.)
Display 4.3 Lengths of stay for the patients during the first 20 days.
(continued)
221
22-03-2009 20:59
Lesson
4 de 12
(continued)
4. Select a day at random from days 1 through 20 (or however many days all
of your beds are full).
5. Compute the average length of stay for the five patients in the beds on
that day.
6. Pool your results with those of the rest of your class until you have
about 30 estimates of the average length of a stay. Make a dot plot of the
30 averages from your class. Where is this distribution centered? Is there
much variability in your estimates?
7. Compute the average stay for the whole population (the original deck of all
40 cards).
8. Compare your results in step 6 and step 7. On average, are your estimates
generally too low, too high, or about right? Why is this the case?
9. What are the units in this situation? Did every unit have an equal chance
of being in the sample? If so, explain. If not, which units had the greater
chance?
10. How could you improve the sampling method?
In everyday language, we say an opinion is biased if it unreasonably favors
one point of view over others. A biased opinion is not balanced, not objective.
In statistics, bias has a similar meaning in that a biased sampling method is
unbalanced.
A sampling method is biased if it produces samples such that the estimate
from the sample is larger or smaller, on average, than the population parameter
being estimated.
Sampling bias lies in
the method, not in the
sample.
Theres an important distinction here between the sample itself and the
method used for choosing the sample.
Investigator: What makes a good sample?
Statistician: A good sample is representative. That is, it looks like a small
version of the population. Proportions you compute from the
sample are close to the corresponding proportions you would
get if you used the whole population. The same is true for other
numerical summaries, such as averages and standard deviations
or medians and IQRs.
Investigator: How can I tell if my sample is representative?
Statistician: Theres the rub. In practice, you cant. You can tell only by
comparing your sample with the population, and if you know
that much about the population, why bother to take a sample?
22-03-2009 20:59
Lesson
5 de 12
Investigator: Great! First you tell me my sample should be representative, and

then you tell me theres no way to know whether it is. Is that the
best statisticians can do?
Statistician: Nope. Although you cant tell about any particular sample, it is
possible to tell whether a sampling method is good or not. Thats
where bias comes in.
Investigator: I thought biased was just a fancy word for nonrepresentative.
Not true?
Statistician: Now were getting to the point. Bias refers to the method, not the
samples you get from it. A sampling method is biased if it tends to
give nonrepresentative samples.
Investigator: Now I get it. I may not be able to tell whether my sample is
representative, but if I use an unbiased method, then I can be
confident that my sample is likely to be representative. Right?
Statistician: Now youre thinking like a statistician. Theres more detail to come,
but you have the big picture in focus.
Bias
D2. Explain the difference between nonrepresentative and biased as these
terms pertain to sampling.
D3. Which statements describe an event that is possible? Which describe an
event that is impossible?
A. A representative sample results from a biased sample-selection method.
B. A nonrepresentative sample results from a biased sample-selection
method.
C. A representative sample results from an unbiased sample-selection
method.
D. A nonrepresentative sample results from an unbiased sample-selection
method.
Sample Selection Bias
Size bias is one kind of

sample selection bias.
Sample selection bias, or sampling bias, is present in a sampling method if

samples tend to result in estimates of population parameters that systematically
are too high or too low. Various forms of this selection bias can undermine the
usefulness of samples and surveys.
You explored one kind of sampling bias in Activity 4.1a, in which patients
who spent more days in the hospital were more likely to be selected for the
sample. In fact, the chance of selection is proportional to the length of stay. A
5-day stay is five times as likely to be chosen as a 1-day stay. This type of sample
selection bias is called size bias. Suppose a wildlife biologist samples lakes in a
state by dropping grains of rice at random onto a map of the state and then selects
for study the lakes that have rice on them.This is another example of size bias.
223
22-03-2009 20:59
Lesson
6 de 12
Voluntary response
bias is another kind of
sampling bias.
Convenience sampling is
almost surely biased.
Judgment sampling,
even when taken by
experts, is usually biased.
The quality of your

sample depends on
having a good sampling
frame.
When a television or radio program asks people to call in and take sides on
some issue, those who care about the issue will be overrepresented, and those
who dont care as much might not be represented at all. The resulting bias from
such a volunteer sample is called voluntary response bias and is a second type of
sample selection bias.
Heres a simple sampling method: Take whatevers handy. For example, what
percentage of the students in your graduating class plan to go to work immediately
after graduation? Rather than find a representative sample of your graduating class,
it would be a lot quicker to ask your friends and use them as your samplequicker
and more convenient, but almost surely biased because your friends are likely to
have somewhat similar plans. A convenience sample is one in which the units
chosen are those that are easy to include. The likelihood of bias makes convenience
samples about as worthless as voluntary response samples.
Because voluntary response sampling and convenience sampling tend to
be biased methods, you might be inclined to rely on the judgment of an expert
to choose a sample that he or she considers representative. Such samples, not
surprisingly, are called judgment samples. Unfortunately, though, experts might
overlook important features of a population. In addition, trying to balance
several features at once can be almost impossibly complicated. In the early days
of election polling, local experts were hired to sample voters in their locale by
filling certain quotas (so many men, so many women, so many voters over the
age of 40, so many employed workers, and so on). The poll takers used their own
judgment as to whom they selected for the poll. It took a very close election (the
1948 presidential election, in which polls were wrong in their prediction) for the
polling organizations to realize that quota sampling was a biased method.
An unbiased sampling method requires that all units in the population have
a known chance of being chosen, so you must prepare a list of population
units, called a sampling frame or, more simply, frame, from which you select
the sample. If you think about enough real examples, youll come to see that
making this list is not something you can take for granted. For the Westvaco
employees in Chapter 1 or for the 50 U.S. states, creating the list is not hard, but
other populations can pose problems. How would you list all the people using the
Internet worldwide or all the ants in Central Park or all the potato chips produced
in the United States over a year? For all practical purposes, you canthere will
often be a difference between the populationthe set of units you want to know
22-03-2009 20:59
Lesson
7 de 12
aboutand the sampling framethe list of units you use to create your sample.
A sample might represent the units in the frame quite well, but how well your
sample represents the population depends on how well youve chosen your frame.
Quite often, a convenient frame fails to cover the population of interest (using a
telephone directory to sample residents of a neighborhood, for example), and a
bias is introduced by this incomplete coverage. If you start from a bad frame, even
the best sampling methods cant save you: Bad frame, bad sample.

D4. Identify the type of sampling method used in each of these surveys. Would
you expect the estimate of the parameter to be too high or too low?
a. You use your statistics class to estimate the percentage of students in your
school who study at least 2 hours a night.
b. You send a survey to all people who have graduated from your school in
the past 10 years. You use the mean annual income of those who reply
to estimate the mean annual income of all graduates of your school in
the past 10 years.
c. A study was designed to estimate how long people live after being
diagnosed with dementia. The researchers took a random sample of the
people with dementia who were alive on a given day. The date the person
had been diagnosed was recorded, and after the person died the date of
death was recorded.
D5. You want to know the percentage of voters who favor state funding for
bilingual education. Your population of interest is the set of people likely to
vote in the next election. You use as your frame the phone book listing of
residential telephone numbers. How well do you think the frame represents
the population? Are there important groups of individuals who belong to
the population but not to the frame? To the frame but not to the population?
If you think bias is likely, identify what kind of bias and how it might arise.
Response Bias
Bias doesnt always
come from the sampling
method.
Nonresponse bias can

occur when people do
not respond to surveys.
In all the examples so far, bias has come from the method of taking the sample.
Unfortunately, bias from other sources can contaminate data even from wellchosen sampling units.
Perhaps the worst case of faulty data is no data at all. It isnt uncommon
for 40% of the people contacted to refuse to answer a survey. These people
might be different from those who agree to participate. An example of this
nonresponse bias came from a controversial study that found that left-handers
died, on average, about 9 years earlier than right-handers. The investigators sent
questionnaires to the families of everyone listed on the death certificates in two
counties near Los Angeles asking about the handedness of the person who had
died. One critic noted that only half the questionnaires were returned. Did that
change the results? Perhaps. [Source: Left-Handers Die Younger, Study Finds,
Los Angeles Times,April 4, 1991.]
Questionnaire bias
Nonresponse bias, like bias that comes from the sampling method, arises
from who replies. Questionnaire bias arises from how you ask the questions.
225
22-03-2009 20:59
Lesson
8 de 12
The opinions people give can depend on the tone of voice of the interviewer, the
appearance of the interviewer, the order in which the questions are asked, and
many other factors. But the most important source of questionnaire bias is the
wording of the questions. This is so important that those who report the results of
surveys should always provide the exact wording of the questions.
For example, Readers Digest commissioned a poll to determine how the
wording of questions affected peoples opinions. The same 1031 people were asked
to respond to these two statements:
1. I would be disappointed if Congress cut its funding for public television.
2. Cuts in funding for public television are justified as part of an overall effort to
reduce federal spending.
Note that agreeing with the first statement is pretty much the same as disagreeing
with the second. However, 54% agreed with the first statement, 40% disagreed, and
6% didnt know, while 52% agreed with the second statement, 37% disagreed, and
10% didnt know. [Source: Fred Barnes, Can You Trust Those Polls? Readers Digest, July 1995,
pp. 4954.]
Incorrect response
Another problem polls and surveys have is trying to ensure that people tell
the truth. Often, the people being interviewed want to be agreeable and tend to
respond in the way they think the interviewer wants them to respond. Newspaper
columnist Dave Barry reported that he was called by Arbitron, an organization
that compiles television ratings. Dave reports:
So I figured the least I could do, for television, was be an Arbitron household.
This involves two major responsibilities:
1. Keeping track of what you watch on TV.
2. Lying about it.
At least thats what I did. I imagine most people do. Because lets face it:
Just because you watch a certain show on television doesnt mean you want
to admit it. [Source: Dave Barry, Dave Barry Talks Back, copyright 1991 by Dave Barry.
Used by permission of Crown Publishers, a division of Random House, Inc.]
22-03-2009 20:59
Lesson
9 de 12
Bias from incorrect responses might be the result of intentional lying, but it
is more likely to come from inaccurate measuring devices, including inaccurate
memories of people being interviewed in self-reported data. Patients in medical
studies are prone to overstate how well they have followed the physicians orders,
just as many people are prone to understate the amount of time they actually
spend watching TV. Measuring the heights of students with a meterstick that
has one end worn off leads to a measurement bias, as does weighing people on a
bathroom scale that is adjusted to read on the light side.
Response Bias
D6. Like Dave Barry, people generally want to appear knowledgeable and
agreeable, and they want to present a favorable face to the world. How might
that affect the results of a survey conducted by a school on the satisfaction of
its graduates with their education?
D7. Another part of the Readers Digest poll described on page 226 asked
Americans if they agree with the statement that it is not the governments
job to financially support television programming. The poll also asked them
if theyd be disappointed if Congress cut its funding for public television.
Which question do you think brought out more support for public television?
D8. How is nonresponse bias different from voluntary response bias?
Summary 4.1: Why Take Samples, and How Not To

The population is the set of units you want to know about. The sample is the set of
units you choose to examine. A census is an examination of all units in the entire
population. Important reasons for using a sample in many situations rather than
taking a census include these:
Testing sometimes destroys the items.
Sampling can save money.
Sampling can save time.
Sampling can make it possible to collect more or better information on
each unit.
227
22-03-2009 20:59
Lesson
10 de 12
A sampling method is biased if tends to give results that, on average, are too low
or too high.This can happen if the method of taking the sample or the method of
getting a response is flawed.
Sources of bias from the method of taking the sample include
using a method that gives larger units a bigger chance of being in the sample
(size bias)
letting people volunteer to be in the sample
using a sample just because it is convenient
selecting the sampling units based on expert judgment
constructing an inadequate sampling frame
Types of bias derived from the method of getting the response from the sample
include
nonresponse bias
questionnaire bias
incorrect response or measurement bias
Practice
P1. You want to estimate the average number of
TV sets per household in your community.
a. What is the population? What are the
units?
b. Explain the advantages of sampling over
conducting a census.
c. What problems do you see in carrying
out this sample survey?
Bias
P2. Four people practicing shooting a bow and
arrow made these patterns on their targets.
Display 4.4 Results of four archers.
a. Which person had shots that were biased

and had low variability?
b. Which person had shots that were biased
and had high variability?
c. Which person had shots that were
unbiased and had low variability?
d. Which person had shots that were

unbiased and had high variability?
e. Do you think it would be easiest to help
Al, Cal, or Dal improve?
P3. Describe the type of sample selection
bias that would result from each of these
sampling methods.
a. A county offcial wants to estimate the
average size of farms in a county in Iowa.
He repeatedly selects a latitude and
longitude in the county at random and
places the farms at those coordinates in
his sample. If something other than a
farm is at the coordinates, he generates
another set of coordinates.
b. In a study about whether valedictorians
succeed big in life, a professor traveled
across Illinois, attending high school
graduations and selecting 81 students to
participate. . . . He picked students from
the most diverse communities possible,
from little rural schools to rich suburban
schools near Chicago to city schools.
[Source: Michael Ryan, Do Valedictorians Succeed Big in
Life? PARADE, May 17, 1998, pp. 1415.]
22-03-2009 20:59
Lesson
11 de 12
c. To estimate the percentage of students

who passed the most recent AP Statistics
Exam, a teacher on an Internet discussion
list for teachers of AP Statistics asks
teachers on the list to report to him how
many of their students took the test and
how many passed.
d. To estimate the average length of the
pieces of string in a bag, a student reaches
in, mixes up the strings, selects one,
mixes them up again, selects another,
and so on.
e. In 1984, Ann Landers conducted a poll
on the marital happiness of women by
asking women to write to her.
P4. Suppose the Museum of Fine Arts in Boston
wants to estimate what proportion of people
who come to Boston from out of town
planned their trip to Boston mainly to visit
the museum. The sample will consist of
all out-of-town visitors to the museum on
several randomly selected days. On buying
a ticket to the museum, people will be
asked whether they came from out of town
and, if so, what the main reason for their
trip was. Do you expect the museums
estimate to be too high, too low, or just

about right? Why? What kind of sampling
method is this?
Response Bias
P5. Consider this pair of questions related to gun
control:
I. Should people who want to buy guns
have to pass a background check to make
sure they have not been convicted of a
violent crime?
II. Should the government interfere with an
individuals constitutional right to buy a
gun for self-defense?
Which question is more likely to show a
higher percentage of people who favor some
control on gun ownership?
P6. In one study, educators were asked to rank
Princetons undergraduate business program.
Every educator rated it in the top ten in
the country. Princeton does not have an
undergraduate business program. What kind
of bias is shown in this case? [Source: Anne Roark,
Guidebooks to Colleges Get As, Fs, Los Angeles
Times, November 21, 1982, Part I, p. 25.]
Exercises
E1. Suppose you want to estimate the percentage
of U.S. households have children under
the age of 5 living at home. Each weekday
from 9 a.m. to 5 p.m., your poll takers call
households in your sample. Every time they
reach a person in one of the homes, they
ask, Do you have children under the age of
5 living in your household? Eventually you
give up on the households that cannot be
reached.
a. Will your estimate of the percentage of
U.S. households with children under the
age of 5 probably be too low, too high, or
about right?
b. How does this example help explain why
poll takers are likely to call at dinnertime?
c. Is this a case of sampling bias or of
response bias?
E2. A wholesale food distributor has

commissioned a sample survey to estimate
the satisfaction level of his customers, who
are the owners of small restaurants. The
sampling firm takes a random sample from
the current list of customers, develops a
satisfaction questionnaire, and sends field
workers out to interview the owners of
the sampled restaurants. Field workers
realize that the time does not allow them
to interview all the owners selected for the
sample, so if the owner is busy when the field
worker arrives, the worker moves on to the
next business.
a. What is the population? The sample?
b. What kind of bias does this survey have?
c. What kind of restaurants would you
expect to be underrepresented in the
sample?
229
22-03-2009 20:59
Lesson
12 de 12
E3. Suppose you want to estimate the average

response to the question Do you like
math? on a scale from 1 (Not at all) to 7
(Definitely) for all students in your school.
You use your statistics class as a sample.
What kind of sample is this? What sort of
bias, if any, would be likely? Be as specific as
you can. In particular, explain whether you
expect the sample average to be higher or
lower than the population average.
E4. At a meeting of local Republicans, the
organizers want to estimate how well their
partys candidate will do in their district
in the next race for Congress.They use the
people present at their meeting as their
sample. What kind of sample is this? What
bias do you expect?
E5. You want to estimate the average number of
states that people living in the United States
have visited. If you asked only people at least
40 years old, would you expect the estimate
to be too high or too low? What bias might
you expect if you take your sample only from
those living in Rhode Island?
E6. For a study on smoking habits, you want to
estimate the proportion of adult males in
the United States who are nonsmokers, who
are cigarette smokers, and who are pipe or
cigar smokers. Tell why it makes more sense
to use a sample than to try to survey every
individual. What types of bias might show up
when you attempt to collect this information?
E7. Television today is more offensive than ever,
say the overwhelming majority92%of
readers who took part in USA Weekends
third survey measuring attitudes toward
the small screen. More than 21,600 people
responded to this write-in survey.
[Source: USA Weekend, May 1618, 1997, p. 20.]
a. What kind of sample is this?

b. Do you trust the results of the survey?
Why or why not?
c. What percentage of the entire U.S.
television-watching public do you think
would say that todays shows are more
offensive than ever: more than 92%, quite a
bit less than 92%, or just about 92%? Why?
E8. Are people willing to change their driving

habits in the face of higher gasoline prices?
At a time of steeply rising gas prices, a
Consumers Union poll of Internet users
who chose to participate showed these
results. [Source: www.consumersunion.org.]
a. What kind of sample is this?

b. Do you trust the results of this survey?
Why or why not?
c. What percentage of U.S. drivers do you
think drive less as a result of higher
gasoline prices: more than 53%, less than
53%, or just about 53%? Why do you
think that?
E9. To estimate the average number of children
per family in the city where you live, you
use your statistics class as a convenience
sample. You ask each student in the sample
how many children are in his or her family.
Do you expect the sample average to be
higher or lower than the population average?
Explain why.
E10. Suppose you wish to estimate the average
size of English classes on your campus.
Compare the merits of these two sampling
methods.
I. You get a list of all students enrolled in
English classes, take a random sample of
those students, and find out how many
students are enrolled in each sampled
students English class.
II. You get a list of all English classes, take a
random sample of those classes, and find
out how many students are enrolled in
each sampled class.
22-03-2009 20:59
Lesson
1 de 12
E11. A Gallup/CNN/USA Today poll asked this

question:
As you may know, the Bosnian Serbs
rejected the United Nations peace plan
and Serbian forces are continuing to
attack Muslim towns. Some people are
suggesting that the United States conduct
air strikes against Serbian military
forces, while others say we should not
get militarily involved. Do you favor or
oppose U.S. air strikes?
On the same day, the ABC News poll worded
its question this way:
Specifically, would you support or
oppose the United States, along with its
allies in Europe, carrying out air strikes
against Bosnian Serb artillery positions
and supply lines?
Explain which poll you think got a larger
response favoring air strikes.
4.2
Relying on chance helps
protect against bias.
E12. Using the gun control example in P5 on

page 229 as a guide, write two versions
of a question about a controversial issue,
with one version designed to get a higher
percentage in favor and the other designed
to get a higher percentage opposed. Test your
questions on a sample of 20 people.
E13. Display 1.1 (page 5) lists data on the 50
employees of the engineering department
at Westvaco. How would you take a sample
of 20 employees that is representative of all
50 with regard to all the features listed: pay
type, year of birth, year of hire, and whether
laid off?
E14. Bring to class an example of a survey from
the media. Identify the population and the
sample, and discuss why a sample was used
rather than a census. Do you see any possible
bias in the survey method?
Random Sampling: Playing It Safe by Taking Chances

The best-planned surveys leave a lot to chance. The key idea is to randomize, that
is, let chance choose the sampling units. This might seem like a paradox at first,
but it makes sense once you understand a basic fact about chancelike behavior:
Selecting your sample by chance is the only method guaranteed to be unbiased.
Youve seen how some common sampling methods are biased, tending to
give nonrepresentative samples. People who care enough to respond voluntarily
might not be typical of others. The members of your class, although theyre
convenient to survey, dont represent some groups well at all. Even relying on
expert judgment doesnt work as well as wed like. Over the long run, chance beats
all these methods.
Simple Random Samples

When each unit in the population has a fixed probability of ending up in the
sample, we call the sample a probability sample. One type of probability sample
is a simple random sample, or SRS.
In a simple random
sample of size
n,each possible
sample of size n has
the same chance of
being selected.
In simple random sampling, all possible samples of a given fixed size are
equally likely. All units have the same chance of belonging to the sample, all
possible pairs of units have the same chance of belonging to the sample,
all possible triples of units have the same chance, and so on.
4.2 Random Sampling: Playing It Safe by Taking Chances 231
22-03-2009 21:00
Lesson
2 de 12
The simplest way to let chance choose your simple random sample amounts
to putting all the individuals from your population into a gigantic hat, mixing
them thoroughly, and drawing individuals out one at a time until you have
enough for your sample. ( This is like mixing the soup before tasting.) Although
this method is exactly right in theory, in practice mixing thoroughly is almost
impossible. So its actually better to implement the theory using random numbers.
Steps in Choosing a Simple Random Sample
1. Start with a list of all the units in the population.
2. Number the units in the list.
3. Use a random number table or generator to choose units from the
numbered list, one at a time, until you have as many as you need.
Do you think you can select representative sampling units as well as a random
number table can? Activity 4.2a provides an opportunity for you to test yourself.
Random Rectangles
What youll need: Display 4.5, a method of producing random digits
Your goal is to choose a sample of 5 rectangles from which to estimate the
average area of the 100 rectangles in Display 4.5.
1. Without studying the display of rectangles too carefully, quickly choose five
that you think represent the population of rectangles on the page. This is
your judgment sample.
2. Find the areas of each rectangle in your sample of five and compute the
sample mean, that is, the average area of the rectangles in your sample.
3. List your sample mean with those of the other students in the class.
Construct a plot of the means for your class.
4. Describe the shape, center, and spread of the plot of sample means from
the judgment samples.
(continued)
22-03-2009 21:00
Lesson
3 de 12
Display 4.5 Random rectangles.
233
22-03-2009 21:00
Lesson
4 de 12
(continued)
5. Now generate five distinct random integers between 0 and 99. (The
rectangle numbered 100 can be called 0.) Find the rectangles whose
numbers correspond to your random numbers. This is your random
sample of five rectangles. [You can use a calculator to generate random
numbers. See Calculator Note 4A.]
6. Repeat steps 24, this time using your random sample.

7. Discuss how the two distributions of sample means are similar and how
they different.
8. Which method of producing sample means do you think is better if the
goal is to use the sample mean to estimate the population mean? (Your
instructor has the value of the population mean.)

D9. Use a table of random digits to take a simple random sample of six students
from your class. For what characteristics is the sample representative of this
class? Not representative?
D10. Describe how you could obtain a simple random sample of all students
enrolled in the English classes at your school.
D11. Readability. Youve decided you simply must know the proportion of
capital letters in this textbook because it can indicate the complexity of the
sentences on a page.Think of the book itself as your population and each
printed character as a unit. How could you get a simple random sample
of 10,000 characters? What are the advantages and disadvantages of this
sampling method?
Stratified Random Samples

Suppose you are planning a sample survey for an international outdoor clothing
manufacturer to see if their image has suffered among their retailers due to negative
press coverage regarding their use of sweatshop labor. The company headquarters
furnishes you with a list of retail outlets throughout the world that sell their
products. Because press coverage is largely national, it will be more informative to
organize this list by country and take a random sample from each country. Youll be
gathering data using each countrys phone or postal system in the language of that
country, so categorizing the retail outlets by country will also be more convenient.
22-03-2009 21:00
Lesson
5 de 12
Classifying the retail outlets by country is called stratification. If you can

divide your population into subgroups that do not overlap and that cover the
entire sampling frame, the subgroups are called strata. If the sample you take
within each subgroup is a simple random sample, your result is a stratified
random sample. This is a second type of probability sample.
Steps in Choosing a Stratified Random Sample
1. Divide the units of the sampling frame into non-overlapping subgroups.
2. Take a simple random sample from each subgroup.
Why stratify?
Stratification reduces
variability.
Why might you want to stratify a population? Here are the three main reasons:
Convenience in selecting the sampling units is enhanced. It is easier to sample
in smaller, more compact groups (countries) than in one large group spread
out over a huge area (the world).
Coverage of each stratum is assured.The company might want to have data
from each country in which it sells products; a simple random sample from
the frame does not guarantee that this would happen.
Precision of the results may be improved. That is, stratification tends to give
estimates that are closer to the value for the entire population than does an
SRS. This is the fundamental statistical reason for stratification.
An example might help clarify the last point.
Example: To Stratify or Not to Stratify

Suppose a geologist has a box of rocks from a mountain stream in Mexico.
Among other things, she wants to estimate the mean diameter of these discusshaped rocks so it can be compared to the mean diameter of similar rocks from
other locations. She has a sieve with a 2-inch mesh, which allows through only
rocks under 2 inches in diameter. To sieve or not to sieve before estimating, that
is the question. In other words, is it better to stratify the rocks into two groups
(small and large) and then sample from each stratum, or is it just as good to
sample from the pile of rocks without sieving?
To simplify the problem, consider a population of 100 rocks. After sieving,
50 turn out to have diameters of less than 2 inches. Display 4.6 shows dot plots
of the population of diameters and of that population divided into the two strata
(small and large rocks).
Twenty simple random samples, each of size 4, were taken from the
population, and the sample mean was calculated for each. A plot of the 20 sample
means is shown in the top half of Display 4.7. Next, 20 stratified random samples
were taken, with each stratified sample consisting of a simple random sample of
size 2 from the small rocks mixed with a simple random sample of size 2 from
the large rocks. A plot of these 20 sample means is shown in the bottom half of
Display 4.7. Note that for both sampling methods we have 20 samples, each of
size 4. Which method is better?
235
22-03-2009 21:00
Lesson
6 de 12
Display 4.6 Data on rock diameters, population and strata.
Display 4.7 Means of simple and stratified random samples of

rock diameters.
Solution
Each method is unbiased because the sample means are centered at the
population mean, about 2.51. However, the means from the stratified random
samples are less variable, tending to lie closer to the population mean. Because its
better to have estimates that are closer to the parameter than estimates that tend
to be farther away, stratification wins.
Make the strata as
different as possible.
Allocate units in the

sample proportionally to
the number of units in
the strata.
From this example, it certainly looks as though the stratification (sieving)

pays off in producing estimates of the mean with smaller variation. This will
be true generally if the stratum means are quite different. If the geologist had
decided to stratify based on color and if color was not related to diameter, then
the stratification would have produced little or no gain in the precision of the
estimates. The guiding principle is to choose strata that have very different means,
whenever that is possible.
One good way to choose the relative sample sizes in stratified random
sampling is to make them proportional to the stratum sizes (the number of units
in the stratum). Thus, if all the strata are of the same size, the samples should all
22-03-2009 21:00
Lesson
7 de 12
be of the same size. If a population is known to have 65% women and 35% men,
then a sample of 100 people stratified on gender should contain 65 women and
35 men. If the sample sizes are proportional to the stratum sizes, then the overall
sample mean (the mean calculated from the samples of all the strata mixed
together) will be a good estimate of the population mean. Proportional allocation
is particularly effective in reducing the variation in the sample means if the
stratum standard deviations are about equal. (If the stratum standard deviations
differ greatly, a more effective allocation of samples to strata can be found.)

D12. Suppose the geologist had a set of sieves that could divide the rocks into
many small piles with essentially equal-diameter rocks in each pile. Would
this process result in a more precise estimate of the mean diameter than that
of the two-strata scenario discussed earlier? Why or why not?
D13. An administrator wants to estimate the average amount of time high school
students spend traveling to school. The plan is to stratify the students
according to grade level and then take a simple random sample from each
grade. What is potentially good and what is potentially bad about this plan?
D14. Your assignment is to estimate the mean number of hours per week spent
studying by students in your school. Discuss how you would set up a
stratified random sampling plan to accomplish the task.
Other Methods of Sampling

Simple random samples and stratified random samples offer many advantages,
but you often run into a major practical problem in attempting to carry them out.
Sampling individual units from a population one at a time is often too costly, too
time consuming, or, if a good frame is not available, simply not possible. Several
commonly used sampling designs get around this problem by forming larger
sampling units out of groups of population units. ( Think of sampling households in
a neighborhood rather than individual residents.) If the groups are chosen wisely,
these designs are often nearly as good as simple or stratified random sampling.
Cluster Samples
To see how well U.S. 4th graders do on an arithmetic test, you might take a
simple random sample of children enrolled in the 4th grade and give each
child a standardized test. In theory, this is a reasonable plan, but it is not very
practical. For one thing, how would you go about making a complete list of all
the 4th graders in the United States? For another, imagine the work required to
track down each child in your sample and get him or her to take the test. Instead
of taking an SRS of 4th graders, it would be more realistic to take an SRS of
elementary schools and then give the test to all the 4th graders in those schools.
Getting a list of all the elementary schools in the United States is a lot easier than
getting a list of all the individuals enrolled in the 4th grade. Moreover, once youve
chosen your sample of elementary schools, its a relatively easy organizational
problem to give the test to all 4th graders in those schools. This is an example of
4.2 Random Sampling: Playing It Safe by Taking Chances 237
22-03-2009 21:00
Lesson
8 de 12
cluster sampling, with each elementary school a cluster of 4th-grade students. A

cluster sample is an SRS of non-overlapping clusters of units.
Steps in Choosing a Cluster Sample

1. Create a numbered list of all the clusters in your population.
2. Take a simple random sample of clusters.
3. Obtain data on each individual in each cluster in your SRS.
The situation of the 4th graders illustrated the two main reasons for using
cluster samples: You need only a list of clusters rather than a list of individuals,
and for some studies it is much more effcient to gather data on individuals
grouped by clusters than on all individuals one at a time.
Two-Stage Cluster Samples

In a two-stage cluster
sample, first take a
cluster sample and then
take an SRS from each
cluster.
The National Assessment of Educational Progress (NAEP) uses a variation on

cluster sampling. For example, to get a sample of 4th graders in Illinois, NAEP takes
a (stratified) random sample of elementary schools. Then, typically 30 4th-grade
students are selected randomly from each school to take the mathematics test.
[Source: nces.ed.gov.] This is an example of two-stage sampling because it involves two
randomizations.
Steps in Choosing a Two-Stage Cluster Sample

1. Create a numbered list of all the clusters in your population, and then take
a simple random sample of clusters.
2. Create a numbered list of all the individuals in each cluster already
selected, and then take an SRS from each cluster.
Reasons to use a
two-stage cluster sample
Two-stage cluster samples are useful when it is much easier to list clusters
than individuals but still reasonably easy and suffcient to sample individuals once
the clusters are chosen. Two-stage cluster sampling might sound like stratified
random sampling, but they are different.
In stratiffed random sampling, you want to choose strata so that the units
within each stratum dont vary much from each other. Then you sample from
every stratum.
In two-stage cluster sampling, you want to choose clusters so that the
variation within each cluster reflects the variation in the population, if
possible. Then you sample from within only some of your clusters.
Systematic Samples with Random Start
Systematic samples can

start with counting off.
To get a quick sample of the students in your class, use the common system for
choosing teams by counting off. Count off by, say, eights, and then select a digit
between 1 and 8 at random. Every student who calls out that digit is in the sample.
This is an example of a systematic sample with random start.
22-03-2009 21:00
Lesson
9 de 12
Steps in Choosing a Systematic Sample with Random Start

1. By a method such as counting off, divide your population into groups of
the size you want for your sample.
2. Use a chance method to choose one of the groups for your sample.
Suppose your population units are in a list, such as a list of names in a
directory, and you want to select a sample of a certain size. The steps in choosing
a systematic sample with random start are equivalent to choosing a random
starting point between units 1 and k and then taking every kth unit thereafter.
The value of k is determined by the sample size. For example, suppose you want
a systematic sample of 20 units selected from a list of 1000 units. Then k is 50,
because selecting a random start between 1 and 50 and then taking every 50th
unit thereafter will result in a systematic sample of 20 units.
When the units in the population are well mixed before the counting off, this
method will produce a sample much like an SRS. Systematic samples, like cluster
samples, are often easier to take than simple random samples of the same size.

D15. Why isnt taking a systematic sample with random start equivalent to taking
a simple random sample?
D16. The Los Angeles Times commissioned an analysis of St. Johns wort, an
over-the-counter herbal medicine, to determine whether the potency of the
pills matched the potency claimed on the bottle. The sampling procedure
was described as this: For the analysis, 10 pills were sampled from each
of three containers of one lot of each product. [Source: Los Angeles Times,
August 31, 1998, p. A10.]
a. What type of sampling was this?

b. Can you suggest an improvement in the sampling procedure?
D17. A newspaper reports that about 60% of the cars in your community come
from manufacturers based outside the United States. How would you design
a sampling plan for gathering the data to substantiate or refute this claim?
D18. Both cluster sampling and stratified random sampling involve viewing the
sampling frame as a collection of subgroups. Explain the difference between
these two types of sampling.
D19. Take a systematic sample of five students from your class.
Summary 4.2: Random Sampling

The main reason for using a chance method to choose the individuals for your
sample is that randomization protects against sample selection bias. And, as you
will see, random selection of the sampling units will make inference about the
population possible.
There are several ways to take a sample using random selection. Some of the
most common are described on the next page.
239
22-03-2009 21:00
Lesson
10 de 12
Simple random sample. Number the individuals and use random numbers to
select those to be included in the sample.
Stratified random sample. First divide your population into groups, called
strata, and then take a simple random sample from each group.
Cluster sample. First select clusters at random, and then use all the individuals
in the clusters as your sample.
Two-stage cluster sample. First select clusters at random, and then select
individuals at random from each cluster.
Systematic sample with random start. Count off your population by a number
k determined by the size of your sample, and select one of the counting numbers
between 1 and k at random; the units with that number will be in your sample.
Display 4.8 shows schematically how five common sampling designs might look
for sampling the blobs in the rectangles.
Display 4.8 Various sampling methods.
22-03-2009 21:00
Lesson
11 de 12
Practice
estimate the percentage of all adults in your
city with receding gums and were given a
P7. Decide whether these sampling methods
budget
suffcient for a random sample of
produce a simple random sample of students
1200 adults.
from a class of 30 students. If not, explain
a. Suppose that, in your city, 43% of adults
why not.
see their dentist twice a year, 32% once
a. Select the first six students on the class
a year, and 25% less frequently than
roll sheet.
once a year. If you stratify on number of
b. Pick a digit at random and select those
visits to a dentist, how many adults from
students whose phone number ends in
each stratum should you include in your
that digit.
sample?
c. If the classroom has six rows of chairs
b. In addition to number of visits to a
with five seats in each row, choose a row at
dentist, which of these variables would
random and select all students in that row.
be the best to consider stratifying on?
d. If the class consists of 15 males and
Explain your answer.
15 females, assign the males the numbers
gender (male/female)
from 1 to 15 and the females the numbers
age (over 40/under 40)
from 16 to 30. Then use a random digit
whether a person eats an apple a day
table to select six numbers from 1 to
30. Select the students assigned those
P9. Suppose your population is 65% women
numbers for your sample.
and 35% men. In a stratified random sample
of 100 women and 100 men, you find that
e. If the class consists of 15 males and
84 women pump their own gas and 69
15 females, assign the males the numbers
men pump their own gas. What is the best
from 1 to 15 and the females the numbers
estimate of the proportion of the entire
from 16 to 30. Then use a random digit
population who pump their own gas?
table to select three numbers from 1 to 15
and three numbers from 16 to 30. Select
the students assigned those numbers for
P10. Suppose 200 people are waiting in line for
your sample.
tickets to a rock concert. You are working for
f. Randomly choose a letter of the alphabet
the school newspaper and want to interview
and select for the sample those students
a sample of the people in line. Show how to
whose last name begins with that letter.
select a systematic sample of
If no last name begins with that letter,
a. 5% of the people in line.
randomly choose another letter of the
alphabet.
b. 20% of the people in line.
P11.
The
American Statistical Association
directory lists its roughly 17,000 members
P8. Receding gums are most common in people
in alphabetical order. You want a sample of
over age 40, but they can occur in younger
about 1000 members. Describe how to use
adults who have periodontal (gum) disease.
the alphabetical listing to take a systematic
There does not appear to be a great difference
sample with random start.
between the rates of males and females
with receding gums. You have been hired to
241
22-03-2009 21:00
Lesson
12 de 12
P12. A bank keeps its list of home mortgages in

chronological order, from earliest to latest
in terms of the date they were assumed. An
auditor wants to get a quick check on the
average balance of these mortgages by taking
a sample.
a. Describe how the bank could take a
systematic sample with random start.
b. Describe how to take a cluster sample.
c. Describe how to take a two-stage cluster
sample.
P13. More readability. Reread D11 on page 234,
which asked about taking an SRS to estimate
the proportion of capital letters in this book.
a. Instead of an SRS, consider taking a

cluster sample. What would you use for
your clusters? What are the advantages
of the cluster sample? Are there any
disadvantages?
b. Suppose you want to estimate the average
number of capital letters per printed
line, so now your sampling unit is a line.
Describe how to take a two-stage cluster
sample from this textbook.
c. Now suppose your sampling unit is an
individual printed character. Describe
how to take a three-stage cluster sample of
characters from this book.
Exercises
E15. A wholesale food distributor has hired you
to conduct a sample survey to estimate the
satisfaction level of the businesses he serves,
which are mainly small grocery stores and
restaurants. A current list of businesses
served by the distributor is available for the
selection of sampling units.
a. If the distributor wants good information
from both the small grocery store owners
and the restaurant owners, what kind of
sampling plan will you use?
b. If the distributor wants information from
the customers who frequent the grocery
stores and restaurants he serves, how
would you design the sampling plan?
E16. Cookies. Which brand of chocolate chip
cookies gives you the most chips per cookie?
For the purpose of this question, take as your
population all the chocolate chip cookies
now in the nearest supermarket. Each cookie
is a unit in this population.
a. Explain why it would be hard to take an
SRS.
b. Describe how to take a cluster sample of
chocolate chip cookies.
c. Describe how you would take a twostage cluster sample. What circumstances
would make the two-stage cluster sample
better than the cluster sample?
E17. Haircut prices. You want to take a sample of

students in your school in order to estimate
the average amount they spent on their last
haircut. Which sampling method do you
think would work best: a simple random
sample; a stratified random sample with
two strata, males and females; or a stratified
random sample with class levels as strata?
Give your reasoning.
E18. The Oxford Dictionary of Quotations, 3rd
edition, has about 600 pages of quotations.
Describe how you would take a systematic
sample of 30 pages to use for estimating the
number of typographical errors per page.
E19. An early use of sampling methods was in
crop forecasting, especially in India, where
an accurate forecast of the jute yield in the
1930s made some of the techniques (and
their inventors) famous. Your job is to
estimate the total corn yield, right before
harvest, for a county with five farms and
a total of 1000 acres planted in corn. How
would you do the sampling?
E20. You are called upon to advise a local movie
theater on designing a sampling plan for a
survey of patrons on their attitudes about
recent movies. About 64% of the patrons are
adults, 30% are teens, and 6% are children.
The theater has the time and money to
22-03-2009 21:00
Lesson
1 de 19
interview about 50 patrons. What design

would you use?
E21. Quality-control plans in industry often
involve sampling items for inspection. A
manufacturer of electronic relays (switches)
for the TV industry wants to set up a
quality-control sampling plan for these
relays as they come off its production line.
What sampling plan would you suggest if
4.3
Inference about cause
and effect
there is only one production line? If there are

five production lines?
E22. Suppose that a sample of 25 adults contains
only women. Two explanations are possible:
(1) The sampling procedure wasnt random,
or (2) a nonrepresentative sample occurred
just by chance. In the absence of additional
information, which explanation would you
be more inclined to believe?
Experiments and Inference About Cause

The previous two sections showed you how to collect data that will allow you
to generalize or infer from the sample you see to a larger but perhaps unseen
population. A second kind of inference, maybe even more important than the
first, takes you from a pattern you observe to a conclusion about what caused the
pattern. Does regular exercise cause your heart rate to go down? Does bilingual
education increase the percentage of non-native-English-speaking students who
graduate from high school? Will taking a special preparatory course raise your
SAT scores? Does smoking cigarettes cause lung cancer?
Cause and Effect

Children who drink more milk have bigger feet than children who drink less
milk. There are three possible explanations for this association:
I. Drinking more milk causes childrens feet to be bigger.
II. Having bigger feet causes children to drink more milk.
III. A lurking variable is responsible for both.
In this case, it is explanation III. The lurking variableone that lies in the
background and may or may not be apparent at the outset but, once identified,
could explain the pattern between the variablesis the childs overall size. Bigger
243
22-03-2009 21:00
Lesson
2 de 19
children have bigger feet, and they drink more milk because they eat and drink
more of everything than do smaller children.
But suppose you think that explanation I is the reason. How can you prove
it? Can you take a bunch of children, give them milk, and then sit and wait to see
if their feet grow? That wont prove anything, because childrens feet will grow
whether they drink milk or not.
Can you take a bunch of children, randomly divide them into a group that
will drink milk and a group that wont drink milk, and then sit and wait to see if
the milk-drinking group grows bigger feet? Yes! Such an experiment is just about
the only way to establish cause and effect.
A Real Experiment: Kellys Hamsters

The goal of every experiment is to establish cause by comparing two or more
conditions, called treatments, using an outcome variable, called the response .
Heres an example of a real experiment, planned and carried out by Kelly
Acampora as part of her senior honors project in biology at Mount Holyoke
College. What happens when an animal gets ready to hibernate? This question is
too general to answer with a single experiment, but if you know a little biology,
you can narrow the question enough for it to serve as the focus of an experiment.
Kelly relied on three previously known facts:
1. Golden hamsters hibernate.
2. Hamsters rely on the amount of daylight to trigger hibernation.
3. An animals capacity to transmit nerve impulses depends in part on an
enzyme called Na K ATP-ase.
Here are the components of Kellys experimental design:
Kellys Question: If you reduce the amount of light a hamster gets from
16 hours to 8 hours per day, what happens to the concentration of
Na K ATP-ase in its brain?
Subjects: Kellys subjects were eight golden hamsters.
Treatments: There were two treatments: being raised in long (16-hour) days
or short (8-hour) days.
Random assignment of treatments: To make her study a true experiment, Kelly
randomly assigned a day length of 16 hours or 8 hours to each hamster in such a
way that half the hamsters were assigned to be raised under each treatment.
Replication: Each treatment was given to four hamsters.
Response variable: Because Kelly was interested in whether a difference in
the amount of light causes a difference in the enzyme concentration, she
chose the enzyme concentration for each hamster as her response variable.
Results: The resulting measurements of enzyme concentrations (in
milligrams per 100 milliliters) for the eight hamsters were
Short days:
Long days:
12.500
6.625
11.625
10.375
18.275
9.900
13.225
8.800
22-03-2009 21:00
Lesson
3 de 19
You can imagine Kelly defending her design:

Kelly: I claim that the observed difference in enzyme concentrations
between the two groups of hamsters is due to the difference in the
number of hours of daylight.
Skeptic: Wait a minute. As you can see, the concentration varies from
one hamster to another. Some just naturally have higher
concentrations. If you happened to assign all the high-enzyme
hamsters to the group that got short days, youd get results like the
ones you got.
Kelly: I agree, and I was concerned about that possibility. In fact, thats
precisely why I assigned day lengths to hamsters by using random
numbers. The random assignment makes it extremely unlikely
that all the high-enzyme hamsters got assigned to the same
group. If you have the time, I can show you how to compute the
probability.
Skeptic: (Hastily) Thats okay for now. Ill take your word for it. But maybe
you can catch me in Chapter 5.
The characteristics of the plan Kelly used are so important that statisticians
try to reserve the word experiment for studies like hers that answer a question by
comparing the results of treatments assigned to subjects at random.
Cause and Effect

D20. Plot Kellys results in a dot plot, using different symbols for hamsters
raised in short days and those raised in long days. In Chapter 9, youll see a
formal method for analyzing data like these to decide whether the observed
difference between the long- and short-day hamsters is too big to be due
only to chance. This method will be similar to the sort of logic introduced
in the Martin v. Westvaco case in Chapter 1. For now, just offer your best
judgment: Do you think the evidence supports a conclusion that the number
of daylight hours causes a difference in enzyme concentration?
D21. Kelly has shown that hamsters raised in less daylight have higher enzyme
concentrations than hamsters raised with more daylight. In order for Kelly
to show that less daylight causes an increase in the enzyme concentration,
she must convince us that there is no other explanation. Has she done that?
Confounding in Observational Studies

For sample surveys, selecting the sample at random protects against bias,
which otherwise can easily mislead you into jumping to false conclusions. For
comparative studies, a threat called confounding also can lead to false conclusions.
Randomizing, when it is possible, protects against this threat in much the same
way it protects against bias in sampling.
In everyday language, confounded means mixed up, confused, at a dead end,
and the meaning in statistics is similar.
245
22-03-2009 21:00
Lesson
4 de 19
Confounding
makes it impossible
to say which of
two explanatory
variables causes the
response.
This thymus surgery

example offers a
cautionary tale about
confounding.
Two possible influences on an observed outcome are confounded if they are

mixed together in a way that makes it impossible to separate their effects on the
responses.
Studies that claim to show that review courses increase SAT scores often
ignore the important concept of confounding. In one study, students at a large
high school were offered an SAT preparation course, and SAT scores of students
who completed the course were higher than scores of students who chose not
to take the course. The positive effect of the review course was confounded with
the fact that the course was taken only by volunteers, who would tend to be
more motivated to do well on the SAT. Consequently, you cant tell if the higher
scores of those who took the course were due to the course itself or to the higher
motivation of the volunteers. As you read the next example, ask yourself which
influences are being mixed together. Wheres the confounding?
The thymus, a gland in your neck, behaves in a peculiar way. Unlike other
organs of the body, it doesnt get larger as you growit actually gets smaller.
Ignorance of this fact led early 20th-century surgeons to adopt a worthless and
dangerous surgical procedure. Imagine yourself in their situation. You know
that many infants are dying of what seem to be respiratory obstructions, and in
your search for a treatment you begin to do autopsies on infants who die with
respiratory symptoms. Youve done many autopsies in the past on adults who died
of various causes, so you decide to rely on those autopsy results for comparison.
What stands out most when you autopsy the infants is that they all have thymus
glands that look too big in comparison to their body size. Aha! That must be
it! The respiratory problems are caused by an enlarged thymus. It became quite
common in the early 1900s for surgeons to treat respiratory problems in children
by removing the thymus. In particular, in 1912, Dr. Charles Mayo (one of the two
brothers for whom the Mayo Clinic is named) published an article recommending
removal of the thymus. He made this recommendation even though a third of the
children who were operated on died. Looking back at the study, its easy to spot
the confounded variables.
The doctors couldnt know whether children with a large thymus tend to have
more respiratory problems, because they have no evidence about children with a
smaller thymus. Age and size of thymus were confounded.
The thymus study is an example of an observational study, not an experiment.
Conditions arent
randomly assigned
in an observational
study.
In an observational study, no treatments get assigned to the units by the

experimenterthe conditions of interest are already built into the units being
studied.
22-03-2009 21:00
Lesson
5 de 19
Observational studies are less desirable than experiments, but often they are
the only game in town. Suppose you want to study accident rates at a dangerous
intersection to see if the rate on rainy days is higher than the rate on dry days. It
turns out that most of the rainy days you have available for study are weekends
and almost all the dry days are during the week. You cannot assign rain to a
weekday! The best you can do is design your observational study to cover a long
enough period so that you are likely to have some weekdays and weekends in both
groups. Otherwise, you have a serious confounding problem.
Dont make the mistake of thinking that observational studies are mere
haphazard collections and observations of existing data. Good observational
studies, like good experiments, can be designed to help answer specific questions.
After researchers take into account every alternative explanation that they can think
of, interesting associations can be observed. For example, epidemiologists studying
the causes of cancer noticed that lack of exercise seemed to be associated with
getting cancer. The best test of a cause-and-effect relationship between amount of
exercise and getting cancer would be an experiment in which thousands of people
would be assigned randomly to exercise or to not exercise. Researchers would have
liked to conduct this experiment, but the practical difficulties were prohibitive in
cost and in getting people to comply over the many years the experiment would
have to run. To examine this issue further, one research study selected a group of
patients with cancer and then matched them according to as many background
variables as they could reasonably measureexcept how sedentary their jobs had
beenwith patients who did not have cancer. Those with cancer turned out to have
more sedentary jobs. But cause and effect can not be established based on such an
observational study alone. The researchers are worried that a confounding variable
is presentthat people who have sedentary jobs are different in some other way
from those who dont. As one researcher said, The problem is the things were not
smart enough to know about, the things we havent even thought of. [Source: Gina
Kolata, But Will It Stop Cancer?, New York Times, November 1, 2005.]
A randomized
experiment comparing
medical treatments is
called a clinical trial.
For drawing conclusions about cause and effect, a good randomized experiment
is nearly always better than a good observational study.
These days, any new medical treatment must prove its value in a clinical
triala randomized comparative experiment. Patients who agree to be part of
the study know that a chance method will be used to decide whether they get
the standard treatment or the experimental treatment. If Dr. Mayo had used a
randomized experiment to evaluate surgical removal of the thymus, he would
have seen that the treatment was not effective and many lives might have been
spared. However, at the time, randomized experiments were not often used in
the medical profession.

D22. Suppose the surgeons had examined infants without respiratory problems
and found that their thymus generally was small.
a. Have they now proved that a large thymus causes respiratory problems in
children? If so, why? If not, what is another possible explanation?
b. Design an experiment to determine whether removal of the thymus helps
children with respiratory problems.
247
22-03-2009 21:00
Lesson
6 de 19
D23. What variables might be confounded with amount of exercise in the

observational studies on the association of lack of exercise with cancer?
D24. What is the main difference between an experiment and an observational
study?
Factors and Levels

Suppose Kelly decides to add two different diets to her hamster experimentcall
them light and heavy. Now her experiment has two factors, diet and length of day,
as shown in the table. Each factor has two levels: long and short for length of day,
and heavy and light for diet. Choosing one level for length of day and one level for
diet gives four possible treatments.
In an observational study, nothing is treated, but the factor and level

terminology still works fine. For example, in the thymus study one factor is the
size of the thymus (large or small) and a second factor is age (child or adult).
The Importance of Randomizing

Randomizing can
protect against
confounding.
Confounding is the main threat to making reliable inferences about cause. You
can think of confounding as the two groups you want to compare differing in
some important way other than just on the conditions you want to compare.
How can you guard against confounding? The best solution, whenever possible,
is to randomize: Use a chance device to decide which people, animals, or objects
to assign to each set of conditions you want to compare. If you dont randomize,
its risky to generalize.
Wait a minute! Are statisticians asking you to believe that if, say, you want to
determine if smoking causes cancer, you should assign people to smoke or not on
the basis of a coin flip? Of course not. But it is true that if you cant use a chance
device to assign the conditions you want to compare, for example, smoker and
nonsmoker, it becomes very difficult to draw sound conclusions about cause
and effect.
Often, you can assign treatments at random to available subjects. For example,
to evaluate a special course to prepare students for the SAT, you can start with
a group of students who want to take the course. Randomly assign half to take
the course and the other half to a group that will not get any special preparation.
Then, after both groups of students have taken the SAT, compare their scores. This
time, if the students who took the course score 40 points better on average, you
can be more confident that the difference is due to the course rather than to some
confounding trait of the students willing to take a review course.
22-03-2009 21:00
Lesson
7 de 19
Why Randomization Makes Inference Possible

If you assign treatments to units at random, then there are only two
possible causes for a difference in the responses to the treatments: chance
or the treatments.
If the probability is small that chance alone will give you such a difference
in the responses, then you can infer that the cause of the difference was the
treatment.
Your goal in Activity 4.3a is to gain practice with randomization as it is used
in the design of experiments and to see the effects of randomization.
Randomization and Its Effect

What youll need: a box, equal-size slips of paper for a random drawing, a coin
The students in your class are to be the subjects in an experiment with two
treatments, A and B. The task is to find the best of three ways to assign the
treatments to the subjects.
1. Choose two leaders from the class, one for Treatment A and one for
Treatment B. The leaders should flip a coin to decide who goes first, then
alternately choose class members for their teams ( much like choosing sides
for a soft ball game ).
2. For each treatment group, record in a table the number of subjects, the
proportion of females, the proportion who have brothers or sisters, and the
proportion who like to read novels in their spare time. Do the two groups
look quite similar or quite different?
3. Next, divide the class by writing the names on the pieces of paper and
putting them in a box. Randomly draw out half of them (one by one) to be
assigned Treatment A. The names remaining are assigned Treatment B.
4. Repeat step 2 for these groups.
5. Finally, divide the class by having each person flip a coin. Those getting
heads are assigned Treatment A, and those getting tails are assigned
Treatment B.
6. Repeat step 2 for these new groups.
7. What are the strengths and weaknesses of each method of assigning
treatments to subjects? Which method is least random?
A Control or Comparison Group Is Vital

Anecdotal evidence
isnt proof.
An article on homeopathic medicine in Time [Is Homeopathy Good Medicine? September 25,
1995, pp. 4748] began with an anecdote about a woman who had been having pain in
her abdomen. She was told to take calcium carbonate, and after two weeks
249
22-03-2009 21:00
Lesson
8 de 19
A control group or a
comparison group
provides the basis of
comparison.
her pain had disappeared. It was reported that her family now turns first to
homeopathic medicine. This kind of personal evidenceIt worked for meis
called anecdotal evidence.
Anecdotal evidence can be useful in deciding what treatments might be
helpful and so should be tested in an experiment. However, anecdotal evidence
cannot prove, for example, that calcium carbonate causes abdominal pain to
disappear. Why not? After all, the pain did go away.
The problem is that pain often tends to go away anyway, especially when
a person thinks he or she is receiving good care. When people believe they are
receiving special treatment, they tend to do better. In medicine, this is called the
placebo effect.
A placebo is a fake treatment, something that looks like a treatment to the
patient but actually contains no medicine. Carefully conducted studies show
that a large percentage of people who get placebos, but dont know it, report
that their symptoms have improved. The percentage depends on the patients
problem but typically is over 30%. For example, when people are told they are
being treated for their pain, even if they are receiving a placebo, changes in their
brain cause natural painkilling endorphins to be released. Consequently, if people
given a treatment get better, it might be because of the treatment, because time
has passed, or simply because they are being treated by someone or something
they trust.
How can scientists determine if a new medication is effective or whether the
improvement is due either to the placebo effect or to the fact that many problems
get better over time? They use a group of subjects who provide a standard for
comparison. The group used for comparison usually is called a control group if
the subjects receive a placebo and a comparison group if the subjects receive a
standard treatment.
Patients in the treatment group are given the drug to be evaluated. The
patients given a placebo get a nontreatment carefully designed to be as much
like the actual treatment as possible. The control and treatment groups should be
handled exactly alike except for the treatment itself. If a new treatment is to be
compared with a standard treatment, the comparison group receives the standard
treatment rather than a placebo.
In order for the control or comparison group and the treatment group to be
treated exactly alike, both the subjects and their doctors should be blind.That is,
22-03-2009 21:00
Lesson
9 de 19
the patients should not know which treatment they are receiving, and the doctors
who evaluate how much the patients symptoms are relieved should be blind as to
what the patients received. If only the patients dont know, the experiment is said
to be blind. If both the patients and the doctors dont know, the experiment is said
to be double-blind.
Sample Medical Study Designs
Comparing a treatment and a placebo. To test the stimulus effect of caffeine,
30 male college students were randomly assigned to one of three groups of ten
students each. Each group was given one of three doses of caffeine (0, 100, and
200 mg), and 2 hours later the students were given a finger-tapping exercise. The
measured response was the number of taps per minute. The 0-mg dose can be
considered a placebo. Ideally, the three groups would have been given pills that
look exactly alike so that they could not tell which treatment they were receiving.
This design allows comparisons to be made among two active treatments and a
placebo to sort out the effect of caffeine. [Source: Draper and Smith, Applied Regression Analysis
(John Wiley and Sons, 1981).]
Comparing a new treatment and a standard treatment. In an early study

on AIDS-related complex (ARC), the drug zidovudine (commonly known as
AZT) combined with acyclovir (ACV) was compared to AZT alone (a standard
treatment). A total of 134 patients with ARC agreed to participate in the study,
which was designed and run as a double-blind, randomized clinical trial. Each
patient was randomly assigned either AZT by itself or a combination of AZT and
ACV. One of the outcome measures was how many of the ARC patients developed
AIDS during the 1 year of the study. A placebo was not used heresuch a
treatment would be unethical because of the health professions obligation to
administer a treatment thought to be effective. You will analyze the outcome of
this experiment in Chapter 8. [Source: David A. Cooper et al., The Efficacy and Safety of Zidovudine
Alone or as Cotherapy with Acyclovir for the Treatment of Patients with AIDS and AIDS-Related Complex:
A Double Blind, Randomized Trial, AIDS 7 (1993): 197207.]
Comparing a new treatment, a standard treatment, and a placebo. Another

experiment compared an undesirable side effect of some antihistamines
drowsinessfor two treatments (meclastine and promethazine) and a placebo.
At the time of the study, meclastine was a new drug and promethazine was a
standard drug known to cause drowsiness. Each of nine subjects was given each
of the three treatments on different days, in random order. The subjects were
blind as to which treatment they were receiving. The outcome measure was the
flicker frequency of patients ( the number of flicks of the eyelids per minute ). Low
flicker frequency is related to drowsiness, because the eyes are staying shut too
long. Note that each subject received all treatments; the only randomization was
in the order in which he or she received them. You will analyze the results of this
experiment in Chapter 9. [Source: D. J. Hand et al., A Handbook of Small Data Sets (London: Chapman
and Hall, 1994), p. 8.]
To summarize, a good experiment must have both a random assignment

of treatments to units and a control or comparison group that is compared
to the group getting the treatment of interest. Such an experiment is called a
randomized comparative experiment.
251
22-03-2009 21:00
Lesson
10 de 19

D25. Why is blinding or double-blinding important in an experiment?
D26. A report about a new study to test the effectiveness of the herb St. Johns
wort to treat depression says, The subjects will receive St. Johns wort, an
antidepressant drug, or a placebo for at least eight weeks and as long as six
months. Describe how you would design this study to compare the effects
of the three treatments. [Source: Los Angeles Times, August 31, 1998, p. A10.]
D27. How would you design an experiment to verify that a placebo effect exists
for people who think they are being treated for pain?
Experimental Units and Replication

In educational research, children appear in a grouptheir classroomso in order
to compare two methods of teaching reading, say, a researcher must assign entire
classrooms randomly to the two methods. In such cases, the classroom, not an
individual student, is the experimental unit.
Suppose a researcher had six classrooms, each with 25 students, available for
her study and assigned three classrooms to each method. The researcher might
like to say that each method was replicated on 75 students because the reading
ability of 75 students was measured, but, alas, she can claim only that each
method was replicated on three classrooms. The students within each classroom
were selected as a group and treated in a group setting and so do not contribute
independent responses.
Experimental units are the people, animals, families, classrooms, and so
on to which treatments are randomly assigned.
Replication is the random assignment of the same treatment to different
units.
Performing many replications in an experiment is the equivalent of having
a large sample size in a surveythe more good observations you have, the more
faith you have in your conclusions. ( Be careful, however, when someone uses the
word replication. Sometimes it is used to mean that your entire experiment was
repeated by someone else who came to the same conclusion you did.)
This box summarizes what you have learned about the characteristics of a
well-designed experiment.
Characteristics of a Well-Designed Experiment

Compare. A treatment group is compared to a control group, or two or
more treatment groups are compared to each other.
Randomize. Treatments are randomly assigned to the available
experimental units.
(continued)
22-03-2009 21:00
Lesson
11 de 19
Replicate. Each treatment is randomized to enough experimental units to

provide adequate assessment of how much the responses from the same
treatment vary.
A Tale of Two Sponges

A true story will show why it is critically important to get the units right when
you plan a study. Although this study happens to be observational, the lesson
about units applies to experiments as well.
It could have been the best of designs, but it was the worst of designs. It
looked like the biggest of samples, but in the end it turned out to be the smallest
of samples.
Once upon a time, many years ago, there lived a statistically innocent
graduate student of biology who wanted to compare the lengths of cells in green
sponges and white sponges. After consulting his advisorwho should have
known betterthe graduate student went to work. For hours at a time, he sat
hunched over his microscope, painstakingly measuring the lengths of tiny cells.
Weeks went by, and eventually, after he had measured cell lengths for about
700 cells per sponge, the graduate student decided he had enough data. It was
time, at last, to do the analysis. This dialogue between the Innocent Graduate
Student (IGS) and the project statistician, the Bearer of Bad News (BBN), reveals
the tragic ending.
IGS: Here are my pages and pages of numbers. Can I conclude that cells
from white and green sponges have different lengths?
BBN: That depends. Well have to look at three thingsthe size of the
difference in the white and green averages, the size of your sample,
and the size of the natural variability from one unit to the next.
IGS: No problem. My sample size is humongous: 700 cells of each kind.
And the variability from one cell to the next isnt all that big.
BBN: Seven hundred cells certainly is a lot. How many sponges did they
come from?
IGS: Uh, just two, one of each color. Does that matter?
BBN: Unfortunately, its critical. But Ill start with the good news. With
700 cells from each sponge, you have rock-solid estimates of
253
22-03-2009 21:00
Lesson
12 de 19
average cell length for the two sponges you actually looked at.
However, . . .
IGS: Right. And because ones green and ones white and the average
lengths are different, I can say that . . .
BBN: Not so fast. You have data that let you generalize from the cells
you measured to all the cells in your particular two sponges. But
thats not the conclusion you were aiming for.
IGS: Yeah. I want to say something about all white sponges compared
to all green ones.
BBN: So for you, the kind of variability that matters most is the
variability from one sponge to another. Your unit should be a
sponge, not a cell.
IGS: You mean to tell me Ive got only two units? After all that work!
BBN: Im afraid you have samples of size 1.
IGS: Woe is me! Id rather be a character in a bad parody of a Dickens
novel. But it will be a far, far better study that I do next time.
The next example describes an experiment with a design that is typical of
experiments done in industrial settings.
Example: Popping Corn

One Saturday afternoon, you decide to compare the number of kernels left
unpopped by a generic and a brand-name popcorn and by your hot air popper
and your oil popper. You have the time to pop 20 batches of popcorn before
hitting the gym.
a. What are the factors and levels? Describe the treatments.
b. How will you design this study? What are the experimental units? How would
randomization be a part of it? Name a variable that might be confounded
with your treatments if you didnt randomize.
c. Have you designed a sample survey, an experiment, or an observational study?
Solution
a. The factors are the type of popcorn and the type of popper. There are two
levels for type of popcorn: generic and brand-name. There are two levels for
type of popper: hot air and oil. The four treatments are the four combinations
of the two types of popcorn with the two types of popper. The treatments are
represented by the four cells in the table.
22-03-2009 21:00
Lesson
13 de 19
The things that receive

the randomization are
your experimental units.
The take-away message

is to randomize what
you can.
b. You will have time to pop five batches for each treatment. Because you cant
select bags of popcorn at random (all the bags in your grocery store probably
were produced at the same time), you will have to assumewithout much
justificationthat the bags of popcorn are pretty much all alike within each
type. You are using the poppers that you have; they arent selected at random
either.
What you can randomize is the order that you assign to the 20 batches to
be popped. You have 20 time periods available to pop batches of popcorn.
Your experimental unit is one of these time periods. Place 20 slips of paper
in a hat, five with hot air/generic written on them, five with hot air/brandname, five with oil/generic, and five with oil/brand-name. Draw them out
one at a time, popping the next batch according to the treatment. This type of
randomization minimizes confounding the treatments with variables such as
how hot the poppers get or the changing humidity in your kitchen.
c. The real world is messy and many designed studies cannot be categorized
perfectly as a survey, an observational study, or an experiment. This is one of
those studies.
The study described in part b is closest to an experiment. The experimental
units are the 20 time periods. Four treatments were to be compared.
Each unit had one of the four treatments assigned to it at random. Each
treatment was replicated on five time periods.
The random assignment over time reduces confounding with variables
related to time, such as the temperature of the room and of the poppers.
The statistics alone, however, do not allow generalization of the results
beyond the popcorn and poppers used. Such generalizations are made by
assuming that the popcorn and the poppers used are representative of the
brands or types, but these assumptions cannot be confirmed by the data
generated in the experiment. That is why real-world experiments generally
are repeated under varying conditions.
If you didnt randomize the four treatments to the time periods in your
answer to part b, there would be no randomization at all and you would
have an observational study.
If you had been able to use random samples of kernels from each of the two
different brands of popcorn, the design would be closer to a sample survey
than to an experiment. You would be surveying the two populations of
popcorn kernels to determine the percentage that would remain unpopped
in each of your two poppers.
Giving a name to the type of study isnt important. What is important is that
you randomize what you can. In the next section, you will learn more about how
to deal with variables that cannot be randomly assigned to the experimental units.

D28. In the teaching-methods study (page 252), why does it seem reasonable that
the experimental unit should be the classroom? Provide an example where
the experimental unit might be the school.
255
22-03-2009 21:00
Lesson
14 de 19
D29. To study whether women released from the hospital 1 day after childbirth
have more problems than women released 2 days after childbirth, many
hospitals in a large city were recruited to participate in an experiment.
Each participating hospital was randomly assigned to release all women
giving routine births either after 1 day or after 2 days of hospital care. An
assessment of problems encountered after the women returned home was
then done on a random sample of women released from each hospital. What
experimental unit should be used as the basis of this study?
D30. Why do you have more faith in your conclusions if your experiment has
many replications?
Summary 4.3: Experiments and Inference

About Cause
The goal of an experiment is to compare the responses to two or more treatments.
The key elements in any experiment are
randomization of treatments to units
replication of each treatment on a sufficient number of units
a control or comparison group
Randomly assigning treatments to units allows you to make cause-and-effect
statements and protects against confounding. If you cant randomize the
assignment of treatments to units, you have an observational study, and
confounding will remain a threat. Confounding makes it impossible to determine
whether the treatment or something else caused the response.
The amount of information you get from an experiment depends on the
number of replications. Recognizing experimental units is crucial.
So that you have a basis for comparison, experiments require either a
control group receiving no treatment or a comparison group receiving a second
treatment.
Practice
Cause and Effect
P14. Research has shown a weak association
between living near a major power line and
the incidence of leukemia in children. Such
a study might measure the incidence of
leukemia in children who live near a major
power line and compare it to the incidence
in children who dont live near a major
power line. Typically, the children in the
areas near major power lines are matched by
characteristics such as age, sex, and family
income to children in the areas not near
major power lines.
a. Identify the subjects, conditions, and
response variable in such a study.
b. Is this type of study a true experiment?

c. According to a newspaper article, While
there is a clear association between
high-voltage power lines and childhood
leukemia, there is no evidence that the
power lines actually cause leukemia.
[Source: Power lines tie to cancer unknown, The Paris
[Texas] News,July 30, 2006.] Why might the
newspaper come to this conclusion?

P15. Solar thermal systems use heat generated by
concentrating and absorbing the suns energy
to drive a generator and produce electrical
power. A manufacturer of solar power
generators is interested in comparing the
22-03-2009 21:00
Lesson
15 de 19
loss of energy associated with the reflectivity

of the glass on the solar panels. (A measure
of energy loss is available.) Three types of
glass are to be studied. The company has
12 test sites in the Mojave Desert, in the
southwestern United States. Each site has
one generator. The sites might differ slightly
in the amount of sunlight they receive.
a. How would you design an experiment to
compare the energy losses associated with
reflectivity?
b. What plays the role of subjects in
your experiment in part a? What is the
treatment?
c. How would you change your design if
each site had six generators?
P16. Each pair of variables here is strongly
associated. Does I cause II, or does II cause I,
or is a lurking variable responsible for both?
a. I. wearing a hearing aid or not
II. dying within the next 10 years or not
b. I. the amount of milk a person drinks
II. the strength of his or her bones
c. I. the amount of money a person earns
II. the number of years of schooling
P17. Show the confounding in the SAT study
described on page 246 by drawing and
labeling a table like the one in the thymus
example on page 246.
Factors and Levels
P18. Does the type of lighting or music in a
dentists waiting room have any effect on the
anxiety level of the patient? An experiment
to study this question could have nine
treatments, represented by this table.
a. What are the factors? What are the levels?

b. Describe a possible response variable.
P1920. Decades ago, when there was less
agreement than there is now about the bad effects
of smoking on health, a large study compared
death rates of three groups of mennonsmokers,
cigarette smokers, and pipe or cigar smokers. The
results are shown here.
Display 4.9 Death rates of males with various

smoking habits. [Source: Paul R. Rosenbaum,
Observational Studies (New York: Springer-Verlag,
1995).]
P19. The numbers seem to say that smoking a

pipe or cigars almost doubles the death
rate, from about 20 to 35 per 1000, but that
smoking cigarettes is pretty safe.
a. Do you believe that is true? If not, can
you suggest a possible explanation for the
pattern?
b. Is this study an observational study or an
experiment?
c. What is the factor? What are the levels?
What is the response variable?
P20. The investigators also recorded the ages
of the men in the study, so it was possible
to compare the average ages of the three
groups.
Display 4.10 Average age of males with various

smoking habits.
Does this information help you account for

the pattern of death rates? If so, tell how. If
not, tell why not. Now what are the factors in
this study?
257
22-03-2009 21:00
Lesson
16 de 19
P21. Driving age. You want to know whether

raising the minimum age for getting a
drivers license will save lives. You compare
the highway death rates for the 50 U.S. states,
grouped according to the legal driving age.
a. Is this study observational or
experimental?
b. What is the factor? What are the levels?
What is the response variable?
c. link of a variable that might be
confounded with legal driving age.
The Importance of Randomizing
P22. Bears in space. Congratulations! You have
just been appointed director of research for
Confectionery Ballistics, Inc., a company
that specializes in launching gummy bears
from a ramp using a launcher made of
tongue depressors and rubber bands.
The CEO has asked you to study a variety

of factors that are thought to affect launch
distance. Your first assignment is to study
the effects of the color of the bears. Do green
bears travel farther?
Display 4.11 shows actual data for bears
launched by members of a statistics class at
Mount Holyoke College. Each launch team
did ten launches, the first set of five launches
using red bears and the second set of five
launches using green bears. Launch distances
are in inches.
a. Which color bear tended to go farther?
b. Your CEO at Confectionery Ballistics,
Inc., wants to reward you for discovering
the secret of longer launches. Explain why
his enthusiasm is premature.
c. Your CEO is adamant: Color is the
key to better launches. In vain you
argue that color and launch order are
confounded. Finally, your CEO issues an
executive order: Prove it. Show me data.
Determined to meet the challenge, you
remember the basic rule of statistics: Plot
your data first. Discuss what the plots in
Display 4.12 show.
d. How would you change the design of the
experiment to eliminate confounding?
Display 4.11 Sample launch distance data, in inches, for red and green gummy bears.
22-03-2009 21:00
Lesson
17 de 19
Display 4.12 Plots of launch distances for red and green gummy bears.

P23. In Dr. Mayos thymus studies, described on
page 246, what did he use as a control group?
Could the placebo effect have been a factor
in any successful surgeries he had? Was it
possible for the study to be blind? Was it
possible for the doctors who evaluated how
well the patients were doing after surgery to
be blind?
P24. In a study to see if people have a magnetic
sensethe ability to use Earths magnetism
to tell directionstudents were blindfolded
and driven around in a van over winding
roads. Then they were asked to point in the
direction of home. They were able to do so
better than could be explained by chance
alone. [Source:Tests Point Away from 6th Sense,
Los Angeles Times, July 19, 1982.]
a. Are you satisfied with the design of

this study? Are there any other possible
explanations for the results other than a
magnetic sense?
To improve the design, the experimenter
placed magnets on the back of some
students heads, which were supposed
to confuse their magnetic sense, and

nonmagnetic metal bars on the back of other
students heads.
b. What is the control group in the new
design?
c. One very important thing wasnt
mentioned in the description of the
design of this experiment. What is that?
d. Are you satisfied with the new design?
Later it was determined that the magnetic
bars tended to stick to the metal wall of the
van.
e. Was this a blind experiment? Was this a
double-blind experiment?
P25. You want to compare two different textbooks
for the AP Statistics course. Ten classes
with a total of 150 students will take part in
the study. To judge the effectiveness of the
books, you plan to use the students scores
on the AP Statistics Exam, taken at the end
of the year. The classes are randomly divided
into two groups of five classes each. The
first group of five classes, with 80 students
259
22-03-2009 21:00
Lesson
18 de 19
total, uses the first of the two books. The

remaining five classes, with 70 students total,
use the other book. Identify the treatments
and the experimental units in this study, and
then identify the effective sample size for
comparing treatments.
P26. You have a summer job working in a
greenhouse. The manager says that she has
discovered a wonderful new product that
will help carnations produce larger blooms,
and you decide to design an experiment to
check it out. What are your experimental
units? How will you use randomization
and replication? Do you need a control or
comparison group?
P27. On another lazy Saturday afternoon, you
decide to compare the strength of two
brands of paper towels, Brand A and
Brand B, each under the conditions wet
and dry. You have one roll of each brand on
hand and have time to test 20 towels from
each roll before hitting the library. One by
one, you stretch each towel in an embroidery
hoop, either leaving the towel dry or wetting
it with a tablespoon of water, and then place
the hoop over a bowl. You place a penny in
the center of the hoop, wait 3 seconds, and

place another penny on top, continuing
until the towel breaks. You record the
number of pennies on the paper towel
before the towel broke.
a. What are the factors and levels? What is

the response variable?
b. How should you use randomization in
this study? What are the experimental
units?
c. Is your study most like a survey, an
observational study, or an experiment?
Exercises
E23. A psychologist wants to compare children
from 1st, 3rd, and 5th grades to determine
the relationship between grade level and
how quickly a child can solve word puzzles.
Two schools have agreed to participate in the
study. Would this be an observational study
or an experiment?
E24. An engineer wants to study traffic flow at
four busy intersections in a city. She chooses
the times to collect data so that they cover
morning, afternoon, and evening hours on
both weekdays and weekends. Is this an
experiment or an observational study?
E25. Buttercups. Some buttercups grow in bright,
sunny fields; others grow in woods, where it
is both darker and damper. A plant ecologist
wanted to know whether buttercups in
sunny locations have adapted in a way that

makes them less successful in the shade
than in the sun. For his study, he dug up ten
plants in sunny locations. Five plants were
chosen at random from these ten plants
and then were replanted in the sun; the
remaining five plants were replanted in the
shade. At the end of the growing season, he
compared the sizes of the plants.
a. Does this study meet the three characteristics
(randomization, replication, control or
comparison group) of a true experiment?
b. Why did the plant ecologist bother to dig
up the plants that were just going to be
replanted in the sunny location? Use the
word confounded in your answer.
22-03-2009 21:00
Lesson
19 de 19
E26. A metallurgist is studying the properties of

copper disks produced by sintering (heating
powder until it becomes a solid mass). He
has two types of powder available and wants
to consider three sintering temperatures.
One oven is available for the sintering,
and many disks can be made in each run.
However, each run can involve only one type
of powder.
a. What are the factors and levels? How
many treatments are there?
b. How should the metallurgist incorporate
randomization into the design of his
study?
c. How could you make this study blind?
E27. The college health service at a small
residential college wants to see whether
putting antibacterial soap in the dormitory
bathrooms will reduce the number of
visits to the infirmary. In all, 1800 students
from 20 dormitories participate. Half the
dormitories, chosen at random, are supplied
with the special soap; the remaining ones
are supplied with regular soap. At the end
of one semester, the two groups of students
are compared based on the average number
of visits to the infirmary per person per
semester.
a. What are the units?
b. How many units are there?
c. Is this an observational study or an
experiment?
E28. If youve studied chemistry or biology, you
may know that different kinds of sugar have
different molecular structures. Simple sugars,
like glucose and fructose, have six carbon
atoms per molecule; sucrose, a compound
sugar, has twice as many atoms. A biologist
wants to know whether complex sugars can
sustain life longer than simple sugars. She
prepares eight petri dishes, each containing
ten potato leafhoppers. Two dishes are
assigned to a control group (no food) and
two each to a diet of glucose, fructose, and
sucrose. The response variable is the time it
takes for half the leafhoppers. in a dish to die.
a. What are the units?

b. How many units are there?
c. Is this an observational study or an
experiment?
E29. Climate and health. Does living in a colder
climate make you healthier? To study
the effects of climate on health, imagine
comparing the death rates in two states with
very different climates, such as Florida and
Alaska. (The death rate for a given year tells
how many people out of every 100,000 died
during that year.) In 2000, Floridas death
rate was more than twice as high as Alaskas.
Why is death rate used as a response variable
rather than number of deaths? Why might
even death rate not be a good choice for a
response variable? (Think about the kinds of
people who move to the two states.) Is this
an experiment or an observational study?
[Source: National Center for Health Statistics, www.cdc.gov.]
E30. You are to direct a study to compare the

graduation rates of the 35 high schools in
your county.
a. How will you define graduation rate?
Why is it important to look at rates rather
than the number of graduates?
b. What issues will determine whether
you do a census, a sample survey, an
observational study, or an experiment?
c. Will you be able to determine the cause
of low graduation rates from the data
collected in your study?
261
22-03-2009 21:00
Lesson
1 de 16
E31. Suppose you have 50 subjects and want

to assign 25 subjects Treatment A and 25
Treatment B. You flip a coin. If it is heads,
the first subject goes into Treatment A; if it is
tails, the first subject goes into Treatment B.
You continue flipping the coin and assigning
subjects until you have 25 subjects in one of
the treatment groups. Then the rest of the
subjects go into the other treatment group.
Does this method randomly divide the 50
subjects into the two treatment groups?
Explain.
E32. Suppose you have recruited 50 subjects and
want to assign 25 subjects to Treatment A
and 25 to Treatment B. For the first subject
4.4
to arrive on the day of the study, you flip a

coin. If it is heads, the first subject goes into
Treatment A; if it is tails, the first subject
goes into Treatment B. The second subject
to arrive then goes into the other treatment
group. For the third subject you flip the coin
and again make the random assignment.
The fourth subject goes into the opposite
treatment from that of the third, and so on.
(You flip the coin for the odd-numbered
subjects only) Does this method randomly
divide the 50 subjects into the two treatment
groups? What good property does this
method have?
Designing Experiments to Reduce Variability

In the previous section, you learned the basics of a good experiment: Randomly
assign one of two or more treatments to each experimental unit, and then handle
them as alike as possible except for the treatment itself. Within this framework,
however, you have further choices about experimental design. If you can protect
yourself against confounding by randomizing, designing a good experiment
becomes mainly a matter of managing variability. In the first part of this section,
you will learn about two types of variability in experiments: one type that you
want because it reveals differences between treatments, and another type that you
dont want because it obscures differences between treatments. The rest of the
section will show you how a good experimental design can reduce the bad kind
of variability.
Differences Between Treatments Versus Variability

Within Treatments
The results for Kellys experiment with hamsters raised in short days and
hamsters raised in long days (page 244) are given again in Display 4.13. There
is a difference in the response variable between the two treatments:The average
enzyme concentration for long days is 8.925, and the average concentration for
short days is 13.906, a difference of 4.981. There also is variability within each
treatmentlike all living things, hamsters will vary, even when treated exactly
alike. In fact, for short days, the difference between the largest and smallest
enzyme concentrations (18.275 11.625, or 6.65) is even larger than the
difference between the two treatment means. Still, because the concentration
22-03-2009 21:01
Lesson
2 de 16
for each of the short days is larger than the concentration for each of the long
days, you probably believe the treatment made a difference.
Display 4.13 Dot plots of enzyme concentration in Kellys

hamsters, in milligrams per 100 milliliters.
Suppose Kelly had gotten results with more spread in the values but the same
means. As Display 4.14 shows, with the values more spread out, its no longer
obvious that the treatment matters.
Display 4.14
Dot plots of altered hamster data.
In order to conclude that the treatments make a difference, the difference

between the treatments has to be large enough to overshadow the variation within
each treatment.
A good experimental plan can reduce within-treatment variability and will
allow you to measure the size of the variability that remains both between and
within treatments.
A Design for Every Purpose

Now that youve seen some of the general ideas behind designing sound
experiments, its time to put them into practice. In Activity 4.4a, you will serve
as an experimental unit ( subject ) for three different versions of an experiment.
The purpose of the experiment is to see whether there is a detectable difference
in heart rate measured under two treatments:
Treatment 1: standing , with eyes open
Treatment 2: sitting , with eyes closed, relaxing
The goal of the activity as a whole is for you to think about the advantages and
disadvantages of three different versions of the experiment. If you dont do the
activity, read through it so you will understand the three designs.
263
22-03-2009 21:01
Lesson
3 de 16
Sit or Stand
You will take your pulse several times during this
activity, as directed here.
a. Get ready, sitting or standing.
b. Have your teacher time you for 30 seconds.
When your teacher says Go, start counting
beats until your teacher says Stop.
c. Double your count to get your heart rate in
beats per minute.
Part A: Completely Randomized Design
In Part A, you and your classmates will be assigned

randomly to one treatment or the other.
1. Random assignment. Your teacher will pass around a box with slips of
paper in it. Half say stand and half say sit. When the box comes to you,
mix up the slips and draw one. Depending on the instruction you get,
either stand with your eyes open, or sit with your eyes closed and relax.
2. Measurement. Take your pulse.
3. Record data. Record your heart rate and that of the other students in your
treatment group.
4. Summaries. Display the data using side-by-side stemplots for the two
treatments. Then compute the mean and standard deviation for each
group. Do you think the treatment makes a difference?
Part B: Randomized Paired Comparison Design (Matched Pairs)
In Part B, you and your classmates will first be sorted into pairs based on an
initial measurement. Then, within each pair, one person will be randomly
chosen to stand and the other will sit.
1. Initial measurement. Take your pulse sitting with your eyes open.
2. Forming matched pairs. Line up in order, from fastest heart rate to slowest;
pair off, with the two fastest in a pair, the next two fastest in a pair, and
so on.
3. Random assignment within pairs. Either you or your partner should
prepare two slips of paper, one that says sit and one that says stand. One
of you should then mix the two slips and let the other person choose one.
Thus, within each pair, one of you randomly ends up sitting and the other
ends up standing.
4. Measurement. Take your pulse.
5. Record data. Calculate the difference standing minus sitting for each pair
of students.
(continued)
22-03-2009 21:01
Lesson
4 de 16
6. Summaries. Display the set of differences in a stemplot. Then compute the

mean and standard deviation of the differences. What should the mean be
if the treatment makes no difference? Do you think the observed difference
is real or simply due to variation among individuals?
Part C: Randomized Paired Comparison Design (Repeated Measures)
This time each person is his or her own matched pair. Each of you will take
your pulse under both treatments: standing and sitting. Youll flip a coin to
decide the order.
1. Random assignment. Flip a coin. If it lands heads, you will sit first and then
stand. If it lands tails, you will stand first.
2. First measurement. Take your pulse in the position chosen by your coin
flip.
3. Second measurement. Take your pulse in the other position.
4. Record data. Record your heart rates in a table with the heart rates of other
students. Calculate the difference standing minus sitting for each student.
5. Summaries. Display the set of differences in a stemplot. Then compute
the mean and standard deviation of the differences. Do you think the
treatment makes a difference?

If you did not do Activity 4.4a, look at the sample results on pages 266268 before
discussing these questions.
D31. Which design do you think is best for studying the effect of position on
heart rate: the completely randomized design, the randomized paired
comparison design (matched pairs), or the randomized paired comparison
design (repeated measures)? Explain what makes your choice better than the
other two designs.
D32. Which of the three designs do you think is least suitable? Explain what
makes this design less effective than the other two. Make up and describe a
new scenario (choose a response and two treatments to compare) for which
the least suitable design would be more suitable than the other two.
D33. Describe the variation within treatments for the design in Part A. How did
the other two designs reduce variation within treatments?
The Completely Randomized Design
In a completely
randomized design,
treatments are randomly
assigned to units
without restriction.
In Activity 4.4a, the design in Part A is an example of what statisticians call a

completely randomized design. The treatments (sit or stand) are assigned to units
completely at random, that is, without any prior sorting or restrictions. For a
completely randomized design (CRD), use a chance device to randomly assign a
treatment to each experimental unit.
265
22-03-2009 21:01
Lesson
5 de 16
Steps in Creating a Completely Randomized Design (CRD)

1. Number the available experimental units from 1 to n.
2. If you have three treatments, for example, use a random number table or
your calculator to pick n/3 integers at random from 1 to n, discarding any
repetitions. The units with those numbers will be given the first treatment.
Again pick n/3 integers, discarding any repetitions. The units with those
numbers will be given the second treatment. The remaining units will get
the third treatment.
Kelly Acamporas hamster experiment used a completely randomized design.
Each hamster was randomly assigned a treatment. Kelly did this in such a way
that her design was balanced ; that is, each treatment was assigned to the same
number of units.
Sample Results for Sit or Stand: Completely Randomized Design
Display 4.15 provides data for the completely randomized design of Activity 4.4a
from one class of 22 students. Each student was randomly assigned to sit or to
stand. The parallel dot plots show that the pulses for the stand group might be a
little higher on average, but the large within-treatment variability obscures a clear
view of any real differences between the treatment means.
Display 4.15 Pulse rates for a completely randomized design.
The Randomized Paired Comparison Design

Matched pairs and repeated measures, the designs in Parts B and C in Activity
4.4a, are examples of randomized paired comparison designs.
22-03-2009 21:01
Lesson
6 de 16
Steps in Creating a Randomized Paired Comparison

(Matched Pairs) Design
In a randomized
paired comparison
design, pairs
of similar units
are randomly
assigned different
treatments.
Blocks are groups of

similar units.
1. Sort your available experimental units into pairs of similar units. The two
units in each pair should be enough alike that you expect them to have a
similar response to any treatment.
2. Randomly decide which unit in each pair is assigned which treatment. For
example, you could flip a coin, with heads meaning the first unit gets the
first treatment and tails meaning the first unit gets the second treatment.
The other unit in the pair gets the other treatment.
In the matched pairs design of Part B, the units were sorted into pairs of
students having similar sitting heart rates, with treatments randomly assigned
within each pair. In the repeated measures design of Part C, both treatments
were assigned, in random order, to each person (the ultimate in matching). In
experiments of this type, the matched units or the individual units that receive
all treatments in random order are called blocks.
Sample Results for Sit or Stand: Matched Pairs and Repeated
Measures Designs
Display 4.16 provides data for the matched pairs design (Part B) of Activity 4.4a
for the same 22 students. Students were paired based on their sitting pulse rate.
Each pair then flipped a coin to decide who would sit and who would stand.
Because the data are paired, it is appropriate to look at the differences between
the standing and sitting pulse rates as measures of increase in pulse rate due to
standing. The dot plot of these differences shows that well over half the differences
are greater than 0.
Display 4.16 Difference in pulse rates for a matched pairs design.
Display 4.17 completes the look at data from Activity 4.4a. In this repeated
measures design (Part C), each student in the same class as before received each
267
22-03-2009 21:01
Lesson
7 de 16
treatment. Each student flipped a coin to determine whether to sit first or stand
first. The dot plot shows that only one of the differences is negative and that 18 of
the 22 differences are greater than 0.
Display 4.17 Difference in pulse rates for a repeated measures

design.
Sit or Stand
D34. Does the randomized paired comparison designs or the completely
randomized design provide stronger evidence that the pulse rate tends to be
higher when standing than when sitting?
D35. Which of the two randomized paired comparison designs provides stronger
evidence that the pulse rate tends to be higher when standing than when
sitting?
The Randomized Block Design
A randomized paired comparison design is a special type of randomized block
design. The difference is that in a generalized block design more than two
experimental units may be in each block.
22-03-2009 21:01
Lesson
8 de 16
Steps in Creating a Randomized Block Design

In a randomized
block design,
treatments are
randomly assigned
within blocks of
similar units.
1. Sort your available experimental units into groups (blocks) of similar units.
The units in each block should be enough alike that you expect them to
have a similar response to any treatment. This is called blocking.
2. Randomly assign a treatment to each unit in the first block. (Its usually
best if the same number of units is assigned to each treatment.)Then go
to the second block and randomly assign a treatment to each unit in this
block. Repeat for each block.
The use of the term block in statistics comes from agriculture. One of the
earliest published examples of a block design appeared in R. A. Fishers The Design
of Experiments (1935). The goal of the experiment was to compare five types of
wheat to see which type gave the highest yield. The five types of wheat were the
treatments; the yield, in bushels per acre, was the response; and eight blocks of
farmland were available for planting.
There were many possible sources of variability in the blocks of land, mainly
differences in the composition of the soil: what nutrients were present, how well
the soil held moisture, and so on. Because many of these possibly confounding
influences were related to the soil, Fisher made the reasonable assumption that
the soil within each block would be more or less uniform and that the variability
to be concerned about was the variability from one block to another. His goal in
designing the experiment was to keep this between-block variability from being
confounded with differences between wheat types.
Fishers plan was to divide each of the eight blocks of land into five plots
running from end to end of the block, and lying side by side, making forty plots in
all. These 40 plots were his experimental units. Fisher then used a chance device to
assign one type of wheat to each of the five plots in a block. Display 4.18 shows how
his plan might have looked. A, B, C, D, and E represent the five types of wheat.
Display 4.18
An experimental design using blocks.
Suppose it turned out that one of the blocks had really poor (or really
favorable) conditions for wheat. Then Fishers blocking accomplished two things:
All five types of wheat would be affected equally, and the variation within a block
could be attributed to the types of wheat.
The effectiveness of blocking depends on how similar the units are in each
block and how different the blocks are from each other. Here similar units are
269
22-03-2009 21:01
Lesson
9 de 16
Similar units in a block

make for effective
blocking.
units that would tend to give similar values for the response if they were assigned
the same treatments. The more similar the units within a block, the more effective
blocking will be.
Randomized Designs
D36. What is the main difference between a completely randomized design and a
randomized block design?
D37. Why is blocking sometimes a desirable feature of a design? Give an example
in which you might want to block and an example in which you might not
want to block.
A Capstone Activity on Experimental Design:

Bears in Space
A good protocol for an

experiment attempts to
reduce within-treatment
variation.
In this capstone activity, you will return to your job as director of research for
Confectionery Ballistics, Inc., the company that specializes in launching gummy
bears into space (see P22 on page 258).
Discovering the secret of long launches might not be as important as finding
a cure for cancer, but, unlike examples from real science, the bears are something
everyone in your class can know firsthand. Youre going to return to the bears,
but this time color wont matter. For Activity 4.4b, there are two treatments: one
book under the ramp and four books under the ramp. The response variable is
launch distance. Each team will do a set of ten launches, with teams randomly
assigned to use either a steeper or a flatter launch angle. To judge the effects of the
treatments, it is important to keep all other potential influences on the response as
nearly constant as possible. One important way to do this is to write out a careful
protocol that tells you exactly how the experiment is to be done. To write a good
protocol, you first have to identify the important sources of within-treatment
variation. This requires that you know the situation fairly well, which is one of the
main reasons for using the bear-launch study as an example.
Bears in Space
What youll need: for each team, a
launch ramp and launcher, a coin, a
tape measure or yardstick, a supply
of gummy bears, four copies of this
textbook
1. Form launch teams and construct
your launcher. Your class should
have an even number of teams, at
least four. Make your launcher
from tongue depressors and
rubber bands. First wind a rubber band enough times around one end of a
stick to keep it firmly in place. Then place that stick and another together
( continued )
22-03-2009 21:01
Lesson
10 de 16
2.
3.
4.
5.
and wind a second rubber band tightly around the other end of the two
sticks to bind them firmly together. Insert a thin pencil or small wooden
dowel between the two sticks to act as a fulcrum.
Randomize. Use random numbers or a random ordering of cards
containing team names to assign each team to produce data for one or
the other of the two treatments:
Steeper ramp (four books)
Flatter ramp (one book )
The same number of teams should be assigned each treatment.
Organize your team. Decide who
will do which job or jobs: hold the
launcher on the ramp, load the
gummy bear, launch the bear, take
measurements, record the data.
(When you launch, use a coin
instead of your fingers, or you
could end up with sore fingers.)
Gather data. Each team should do
ten launches. After each launch,
record the distance. Measure the
distance from the front of the ramp,
and measure only in the direction
parallel to the ramp.
Summarize. Plot your teams ten launch distances on a dot plot. Summarize
the spread by computing the standard deviation.
Your plots in Activity 4.4b show variability from three sources: the
launch angle, that is, the variability due to the two different treatments (one
book, four books); the particular team, that is, the variability from one team
to another for teams with the same launch angle; and the individual launches,
that is, the variability between launches for the same team. The first is the
difference you want to seethe difference between the two treatments. The
second two are nuisance sources of variability for launches with the same angle;
these are to be minimized. In the discussion questions, you will examine these
sources of variation.
Bears in Space: Within-Team Variability

D38. Why do the launch distances vary so much for your team? List as many
explanations as you can think of, and then order them, starting with the one
you think has the most effect.
D39. One important source of variation between launches is the effect of practice.
Can you think of an alternative strategy that would reduce or come close to
eliminating the variation due to practice?
271
22-03-2009 21:01
Lesson
11 de 16
D40. Use your answers to D38 and D39 to write an experimental protocola
set of rules and steps to follow in order to keep launch conditions as nearly
constant as possible. Your protocol should include rules for deciding when,
if ever, you should not count a bad launch. After each team has written
a protocol, your class should discuss them and combine them into a single
protocol to be used for future launches.
Bears in Space: Between-Team Variability

D41. Which measure of centermean or mediangives a better summary of a
teams set of ten launches? Or are they about equally good? Give reasons for
your judgment.
D42. Record the team summaries (means or medians) in parallel dot plots, one
for each launch angle. Compute the standard deviation of the means (or
medians) for each launch angle. Is there more variation between teams with
the same launch angle or between launches for the same team (calculated in
Activity 4.4b, step 5)? Why do you think the team summaries vary as much
as they do?
List as many explanations as you can think of; then order them, starting
with the one you think has the most effect. If you were going to summarize
a teams results with a single-number measure of center anyway, why not do
only one launch per team and use that number?
D43. What strategies would you recommend for managing the variability between
the teams that used the same launch angle?
Bears in Space: Difference Between Treatments

D44. Use the data from the entire class to construct parallel dot plots showing
individual launches by team. Think about ways to use the data to measure
and compare the sizes of the three kinds of variabilitybetween-treatment,
between-team, and within-team. Can you say which variability is largest?
Smallest?
D45. Without doing a formal analysis, choose a tentative conclusion from among
these three statements, and give reasons for your choice.
I. Launch angle clearly makes a big difference.
II. Launch angle may very well make a difference, but theres too much
variation from other sources to be able to isolate and measure any
effect due to launch angle.
III. Theres not much variability anywhere in the data, so its safe to
conclude that the effect due to launch angle, if any, is quite small.
D46. Combine the data from all the teams that launched with one book into a
single group. Do the same for the data from the four-book teams. Make
parallel dot plots of one-book launches and four-book launches. Show all
individual launches, ignoring teams. Compare the centers and spreads of the
two sets of launches. Why might this analysis be misleading?
22-03-2009 21:01
Lesson
12 de 16
More on Blocking
In Activity 4.4b, you found a lot of variability, even for the same launch angle.
Some teams are more skillful than others in launching the bear.
Within a team, some launches go more smoothly than others.
These two kinds of variability within the treatment can obscure a difference
caused by the launch angle.
Ordinarily, when teams carry out the bear-launch experiment, their results
show that the variability between teams is so large that it is hard to tell how much
difference the angle of the launch ramp makes. Following a careful protocol can
reduce the variability somewhat, but there will still tend to be big differences from
one team to another. One solution to the problem is to increase the number of
experimental units, that is, use more teams.
Although increasing the number of units is a reliable way to get more
information about the difference due to treatments, it can be costly or
time-consuming. As youll see in Activity 4.4c, sometimes there are much
better ways to manage variability.
Block Those Bears!

What youll need: the same equipment as in Activity 4.4b
1. Form teams and assign jobs, as in Activity 4.4b.
2. Randomize. Each team will conduct two sets of five launches, one set with
four books under the ramp and one set with just one book. Each team
should flip a coin to decide the order of the treatments.
Heads: four books first, then one book
Tails: one book first, then four books
3. Gather data. Carry out the launches, following the protocol your class
developed in D40, and record the launch distances.
4. Compute averages. List your teams averages for the two treatments and
their difference in a table. Pool the results from all the teams. Based on the
differences between the averages, do you think it is reasonable to conclude
that launch angle makes a difference, or is it impossible to tell?
Block Those Bears!

D47. What kind of experimental design is used in Activity 4.4c? What are the
blocks?
D48. Units. Remember that you randomly assign a treatment to a unit. For
the design of Activity 4.4c, which is the unit: a single launch, a set of five
launches, or a set of ten launches by a team?
D49. Another block design. Suppose you were to redo Activity 4.4c. Can you think
of a way to improve on the design?
273
22-03-2009 21:01
Lesson
13 de 16
Summary 4.4: Designing Experiments

to Reduce Variability
Random assignment of treatments to experimental unitsthe fundamental
principle of good designcan be accomplished through two basic plans:
The completely randomized design is characterized by randomly assigning a
treatment to each experimental unit while keeping the number of units given
each treatment as equal as possible.
The randomized block design is characterized by placing similar experimental
units into different groups, called blocks, and randomly assigning the
treatments to the units within each block.
The reason for the randomized block design is that, in any experiment,
responses vary not only because of different treatments (between-treatment
variation) but also because different subjects respond differently to the same
treatment (within-treatment variation). In a well-designed experiment, the
experimenter should minimize the amount of variability within the treatment
groups because it can obscure the variability resulting from differences between
treatments. By taking the blocks into account in analyzing the data, the effect of
the within-treatment variability can be minimized.
You have learned two special cases of block design:
A randomized paired comparison (matched pairs) design involves randomly
assigning two treatments within pairs of similar units, such as to twins or to
left and right feet.
A randomized paired comparison (repeated measures) design involves
the assignment of all treatments, in random order, to each unit so that
comparisons can be made on the same units. An example is a study in which
each patient in a clinic is assigned each of three treatments for asthma, in
random order.
Practice
Differences Between Treatments Versus
Variability Within Treatments
P28. Review the antibacterial soap experiment
in E27.
a. List at least two sources of withintreatment variability.
b. Is the point of randomization to reduce
the within-treatment variability or to
equalize it between the treatment groups?

P29. To test a new drug for asthma, both the new
drug and the standard treatment, in random
order, will be administered to each subject in
the study.
a. What kind of design is this?
b. An observant statistician cries, No, no!
Use two similar subjects in each pair,
randomized to each treatment. What
kind of design is this?
c. Which design do you prefer, and why?
22-03-2009 21:01
Lesson
14 de 16
P30. In one design of the heart-rate experiment

in Activity 4.4a, each person provided one
measurement on each treatment. Which
design was that?
P31. For each experiment, describe the withintreatment variability that might obscure
any difference between treatments. Then
describe an experimental design that
includes blocking, and define a response
variable.
a. To determine whether studying with
the radio on helps or hurts the ability to
memorize, there will be two treatments:
listening to the radio and not listening
to the radio. The subjects available are all
seniors in the same school.
b. To determine whether adding MSG
to soup makes customers eat more
soup, a large restaurant will assign two
treatments: adding MSG to the soup and
not adding MSG. The subjects available
are all customers during one evening.
P32. In the experiment to compare drowsiness
caused by two antihistamines (Section 4.3,
page 251), a new drug, meclastine, was
compared to a standard drug, promethazine,
and to a placebo. Each subject was given
each treatment. Such a design can be used
only because the effect of the antihistamine
wears off in a relatively short period of
time, allowing more than one treatment
to be applied to each subject in the study.
The response was the average number of
eyelid flicks per minute, because low flicker
frequency is related to drowsiness. Results
are given in Display 4.19.
a. Do the parallel dot plots show much
evidence that one drug is performing
better than the other? Is it easy to answer
the question from these plots? Explain.
b. Display 4.20 shows the scatterplot for
the paired data on Treatments A and C.
The line on the scatterplot is the line
y = x. Does it now look as if Treatment A
is better than Treatment C? Is it easier to
answer this question from the scatterplot
that shows the paired data?
Display 4.19 Average number of eyelid flicks

per minute.
Display 4.20 Average number of eyelid flicks

per minute for subjects given both
Treatments A and C.
c. How else might you compare the

responses for Treatments A and C?
d. What is the design of this study?
Explain why this design is better than a
completely randomized design.
275
22-03-2009 21:01
Lesson
15 de 16
Exercises
E33. The SAT discussion from Section 4.3
a substance that helps transmit nerve
ended with a revised plan that was truly
impulses. To some extent, you can relieve
experimental (see pages 248249). To which
the symptoms by eating a restricted diet,
of the two types does that design belong,
one low in the amino acid phenylalanine. A
completely randomized or randomized
study was designed to measure the effect of
block? Summarize the design by giving the
such a diet on the levels of dopamine in the
experimental units, factors, levels, response
body. The subjects were ten PKU patients.
variable, and blocks (if any).
Each patient was measured twice, once after
a week on the low phenylalanine diet and
E34. School academic programs often are
again after a week on a regular diet. Identify
evaluated on what are called gain scores,
blocks (if any), treatments, and experimental
the gain in academic achievement (as shown
units, and describe the design used in this
by standardized test scores) over the course
study.
of an academic year. Such scores often are
reported at the school level, not the student
E36. In P15, you considered this problem: Solar
level. That is, each school will receive only
thermal systems use heat generated by
the mean score of all its children in a given
concentrating and absorbing the suns energy
grade. Suppose ten schools have agreed to
to drive a generator and produce electrical
participate in a study to evaluate the effect
power. A manufacturer of solar power
of calculator use in teaching mathematics
generators is interested in comparing the
(extensive use versus limited use).
loss of energy associated with the reflectivity
of the glass on the solar panels. (A measure
a. In the first design, five participating
of energy loss is available.) Three types of
schools are randomly assigned one of
glass are to be studied. The company has
the two treatments and the other five
12 test sites in the Mojave Desert, in the
schools are assigned the other treatment.
southwestern United States. Each site has
The gain scores in mathematics for
one generator. The sites might differ slightly
the two treatment groups are to be
in terms of the amount of sunlight they
compared at the end of the year. Is this a
receive.
completely randomized or a randomized
block design? Identify the treatments,
Now the manufacturer wants to expand
experimental units, and blocks (if any).
the test sites to include generators that are
available in the mountains of Tennessee,
b. In the second design, the participating
the desert of northern Africa, and central
schools first are paired according to
Europe. Each of these locations has 12 test
last years gain scores in mathematics.
Within each pair of schools, one school
is randomly assigned to extensive use
of calculators and the other to limited
use. The gain scores in mathematics are
to be compared at the end of the year. Is
this a completely randomized design or
a randomized block design? If the latter,
what type of randomized block design is
it? Identify the treatments, experimental
units, and blocks (if any).
E35. PKU (phenylketonuria) is a disease in which
the shortage of an enzyme prevents the
body from producing enough dopamine,
22-03-2009 21:01
Lesson
16 de 16
sites, with one generator at each site. Further,

there are two designs of panels in which to
place the glass. Describe how you would
design this study, identifying treatments,
experimental units, and blocks (if any).
E37. Finger tapping. Many people rely on the
caffeine in coffee to get them going in the
morning or to keep them going at night.
But wouldnt you rather have chocolate?
Chocolate contains theobromine, an alkaloid
quite similar to caffeine both in its structure
and in its effects on humans. In 1944, C. C.
Scott and K. K. Chen reported the results
of a study designed to compare caffeine,
theobromine, and a placebo. Their design
used four subjects and randomly assigned
the three treatments to each subject, one
treatment on each of three different days.
Subjects were trained to tap their fingers in a
way that allowed the rate to be measured
presumably, this training eliminated any
practice effect. The response was the rate
of tapping 2 hours after taking a capsule
containing either one of the drugs or the
placebo. Are treatments assigned to the
experimental units completely at random or
using a block design? Identify the response,
experimental units, treatments, and blocks
(if any). [Source: C. C. Scott and K. K. Chen, Comparison
of the Action of 1-ethyl Theobromine and Caffeine in
Animals and Man, Journal of Pharmacologic Experimental
Therapy 82 (1944): 8997.]
E38. Walking babies. The goal of this experiment

was to compare four 7-week training
programs for infants to see whether special
exercises could speed up the process of
learning to walk. One of the four training
programs was of particular interest: It
involved a daily 12-minute set of special
walking and placing exercises. The second
program (the exercise control group)
involved daily exercise for 12 minutes but
without the special exercises. The third and
fourth programs involved no regular exercise
(parents were given no instructions about
exercise) but differed in their follow-up:
Infants in the third program were checked
every week, whereas those in the fourth
program were checked only at the end of
the study. Twenty-three 1-week-old babies

took part, and each was randomly assigned
to one of the groups. The response was the
time, in months, when the baby first walked
without help.
Tell how the treatments are assigned to
the experimental unitscompletely at
random or using a block design. Identify the
response, experimental units, treatments,
and blocks (if any).
E39. Sawdust for lunch? Twenty-four tobacco

hornworms served as subjects for this
experiment, which was designed to see how
worms raised on low-quality food would
compare with worms raised on a normal
diet. The two dozen worms were randomly
divided into two groups, and the lucky
half was raised on regular worm food. The
unlucky half got a mixture of 20% regular
food and 80% cellulose. Cellulose has no
more food value for a hornworm than it
has for youneither you nor a hornworm
can digest the stuff. The experimenter kept
track of how much each hornworm ate and
computed a response value based on the total
amount eaten in relation to body weight.
Tell how treatments were assigned to
experimental units in this experiment
completely at random or using a block
design. Identify the response, experimental
units, treatments, and blocks (if any).
277
22-03-2009 21:01
Lesson
1 de 8
E40. Arthritis is painful, and those who suffer

from this disease often take pain relief and
anti-inflammatory medication for long
periods of time. One of the side effects of
such medications is that they often cause
stomach damage such as lesions and ulcers.
The goal of research, then, is to find an
anti-inflammatory drug that causes minimal
stomach damage. An experiment was
designed to test two treatments for arthritis
pain (old and new) against a placebo by
measuring their effect on lesions in the
stomachs of laboratory rats. Rats with
similar stomach conditions at the start of
the experimental period were randomly

assigned to one of the three treatments. Total
length of stomach lesions (in millimeters)
was measured in each rat after a 2-week
treatment period.
a. Was the treatment assignment in this
experiment completely at random or in
blocks? Does this seem like the best way
to make an assignment in this case?
b. What are the experimental units? What is
the response measurement?
c. Is the use of a placebo essential here?
Chapter Summary
There are two main types of chance-based methods of data collection: sampling
methods and experimental designs. Sampling methods, studied in the first part
of this chapter, use chance to choose the individuals to be studied. Typically,
you choose individuals in order to ask them questions, as in a Gallup poll;
thus, samples and surveys often go together. Experiments, introduced in
the second part of this chapter, are comparative studies that use chance to
assign the treatments you want to compare. An experiment should have three
characteristics: random assignment of treatments to units, two or more treatments
to compare, and replication of each treatment on at least two subjects.
The purposes of sampling and experimental design are quite different. Sample
surveys are used to estimate the parameters of fixed, well-defined populations.
Experiments are used to establish cause and effect by comparing treatments.
Display 4.21 summarizes the differences between a survey and an experiment.
Display 4.21 Differences between a sample survey and an

experiment.
Randomization is the fundamental principle in both types of study because

it allows the use of statistical inference (developed in later chapters) to generalize
results. In addition, random selection of the sample reduces bias in surveys,
22-03-2009 21:01
Lesson
2 de 8
and random assignment of treatments to subjects reduces confounding in

experiments.
A careful observational study can suggest associations between variables, but
the possibility of a lurking variable being responsible for the association means
inferences cannot be made.
Display 4.22 shows how sample surveys, experiments, and observational
studies fit together. Any statistical study has at its core the selection of units on
which measurements will be made and the possible assignment of conditions
(or treatments) to those units. The sample surveys discussed in this chapter
fall into the upper-right box, the experiments into the lower-left box, and the
observational studies into the lower-right box. Studies of the type described in
the upper-left box, which use both random sampling and random assignment of
treatments, have not been discussed in this chapter.
Display 4.22 Essentials of designed statistical studies. [Adapted from

F. L. Ramsey and D. W. Schafer, 2002, The Statistical Sleuth, 2nd ed.,
Duxbury, 9, and the notes of Linda Young, University of Florida.]
Review Exercises
E41. For each situation, tell whether it is better to E42. You want to estimate the percentage of people
take a sample or a census, and give reasons
in your area with heart disease who also
for your answer.
smoke cigarettes. The people in your area who
have heart disease make up your population.
You take as your frame all records of patients
hospitalized in area hospitals within the last
5 years with a diagnosis of heart disease. How
well do you think this frame represents the
population? If you think bias is likely, identify
what kind of bias it would be and explain how
it might arise.
Chapter Summary 279
22-03-2009 21:01
Lesson
3 de 8
E43. Consider using your statistics class as

a convenience sample in each of these
situations. For each, tell whether you think
the sample will be reasonably representative
and, if not, in what way(s) you expect your
class to differ from the given population of
interest.
E44. The U.S. Bureau of Labor Statistics collects

data on occupational variables from a
nationwide sample of households. This
paragraph is from its news bulletin Union
Members in 1996 [January 31, 1997, p. 3]:
The data also are subject to nonsampling
error. For example, information on jobrelated characteristics of the worker, such as
industry, occupation, union membership,
and earnings, are sometimes reported by a
household member other than the worker.
Consequently, such data may reflect
reporting error by the respondent.
Moreover, in some cases, respondents might
erroneously report take-home pay rather
than gross earnings, or may round up or
down from actual earnings.
a. Would reporting take-home pay rather
than gross earnings increase or decrease
the estimate of the earnings of various
types of workers?
b. What other sources of nonsampling error
might this study contain?
E45. A friend who wants to be in movies is
interested in how much actors earn and
has decided to gather data using a simple
random sample. The World Almanac and
Book of Facts has a list of actors that your
friend plans to use as a sampling frame.
Would you advise against using that list as a
frame? What sort of bias do you expect?
E46. What is the average number of

representatives per state in the U.S. House
of Representatives? If you really want to
know, you should use a census rather than
a sample, but because you like statistics
so much youve decided to use a random
sample. Tell which of these two sampling
methods is biased and describe the bias: Will
estimates tend to be too high or too low?
Method I. Start with a list of all the current
members of the House. Take a simple
random sample of 80 members, and for each
representative chosen record the number of
representatives from that persons state. Then
take the average.
Method II. Start with a list of all 50 states.
Take a simple random sample of 5 states, and
for each state chosen record the number of
representatives from that state. Then take the
average.
E47. For an article on used cars, a writer wants to
estimate the cost of repairs for a certain 2000
model of car. He plans to take a random
sample of owners who bought that model
from a compilation of lists supplied by all
U.S. dealers of that make of car. He tells
his research assistant to send letters to all
the people in the sample, asking them to
report their total repair bills for last year. His
research assistant tells him, You might want
to think again about your sampling frame.
Im afraid your plan will miss an important
group of owners. What group or groups?
E48. If you look up Shakespeare in just about
any book of quotations, youll find that
the listing goes on for several pages. If you
look up less famous writers, you find much
shorter listings. Whats the average number
of quotations per author in, say, the Oxford
Dictionary of Quotations, 3rd edition?
Suppose you plan to base your estimate on
a random sample taken by this method:
You take a simple random sample of pages
from the book. For each page chosen, you
find the first quotation in the upper-left
corner of that page and record the number
of quotations by that author. Then you find
22-03-2009 21:01
Lesson
4 de 8
the average of these values for all the pages

in your sample. Whats wrong with this
sampling method?
E49. Youve just been hired as a research
assistant with the state of Maine. For
your first assignment, youre asked to
get a representative sample of the fish in
Moosehead Lake, the states largest and
deepest lake. Your supervisor, who knows
no statistics and cant tell a minnow from
a muskellunge, tells you to drag a net with
1-in. mesh (hole size) behind a motorboat,
up and down the length of the lake. No,
you tell him. That method is biased! What
is the bias? What kinds of fish will tend to be
overrepresented? Underrepresented? (You
dont have to know the difference between a
minnow and a muskie to answer this. You do
have to know a little statistics.)
E50. You are asked to provide a forestry
researcher with a random sample of trees on
a 1-acre lot. Your supervisor gives you a map
of the lot showing the location of each tree
and tells you to choose points at random on
the map and, for each point, to take the tree
closest to the point for your sample. Sorry,
sir, you tell him. That sampling method
was once in common use, but then someone
discovered it was biased. It would be better
to number all the trees in the map and use
random numbers to take an SRS. Draw a
small map, with about ten trees, that you
could use to convince your supervisor that
his method is biased. Put in several younger
trees, which grow close together, and two or
three older trees, whose large leaf canopies
discourage other trees from growing nearby.
Then explain which trees are more likely to
be chosen by the method of random points
and why.
E51. Youre working as an assistant to a
psychologist. For subjects in an experiment
on learning, she needs a representative
sample of 20 adult residents of New York
City. She tells you to run an ad in the New
York Times, asking for volunteers, and
to randomly choose 20 from the list of
volunteers. Whats the bias?
E52. Give an example in which nonresponse bias

is likely to distort the results of a survey.
E53. Tell how to carry out a four-stage random
sample of voters in the United States, with
voters grouped by precinct, precincts
grouped by congressional district, and
congressional districts grouped by state.
E54. Tell how to carry out a multistage sample for
estimating the average number of characters
per line for the books on a set of shelves.
E55. A researcher designs a study to determine
whether young adults who follow a
Mediterranean diet (lots of fruits, vegetables,
grains, and olive oil; little red meat) end up
having fewer heart attacks in middle age
than those who follow an average American
diet. She randomly selects 1000 young adults
in Greece and 1000 young adults in the
United States. These 2000 young adults are
categorized according to whether they follow
a Mediterranean diet or a typical American
diet. Thirty years later, she finds that a
lower percentage of those who followed the
Mediterranean diet have had a heart attack.
a. For the purpose of determining whether
a Mediterranean diet results in fewer
heart attacks than a typical American
diet, is this an observational study or an
experiment? Explain.
b. Explain why the researcher cannot
conclude that following a Mediterranean
diet results in a lower chance of having
a heart attack than following an average
American diet.
c. What factors might be confounded with
diet?
E56. A mother does not keep cola in her house
because she is convinced that cola makes
her 7-year-old daughter hyperactive.
Nevertheless, the daughter gets a cola
whenever she has fast food or goes to a
birthday party, and the mother has noticed
that she then acts hyperactive.
a. Has the mother done an experiment or an
observational study?
Chapter Summary
281
22-03-2009 21:01
Lesson
5 de 8
b. Name two factors that might be

confounded with cola, and explain how
they are confounded with cola.
c. Design a study that would allow the
mother to determine whether giving her
daughter cola makes her hyperactive. Are
there any potential problems with such a
study?
E57. For each science project described,
i. classify the project as a survey, an
observational study, or an experiment
ii. if a survey, name the explanatory
variables; if an observational study or
experiment, name the factors and levels
iii. name the response variable
iv. if a survey, give the sample size; if an
observational study, give the number of
observed units; if an experiment, give the
number of experimental units
a. A student wanted to study the effect
of social interaction on weight gain
in mice. A large number of baby mice
from different mothers were weaned,
weighed, and then randomly divided into
four groups. One group lived with their
mother and siblings, one group lived
with their mother but not their siblings,
one group lived only with their siblings,
and each mouse in the last group lived
in isolation. All groups were otherwise
treated alike and were given as much food
as they wanted. After 90 days, the mice
were weighed again.
b. A student picked four different sites on an
isolated hillside at random. At each site,
he measured off a 10-ft-by-10-ft square.
At each site, a sample of soil was taken
and the amounts of ten different nutrients
were measured. The student counted the
number of species of plants at each site,
hoping to be able to predict the number
of species from the amounts of the
nutrients.
c. A college professor helped his daughter
with a 2nd-grade science project titled
Does Fruit Float? They tested 15 different
kinds of fruit and classified them as

floaters and nonfloaters. For the display,
they took a photograph of a watermelon
floating in a swimming pool and a grape
sitting at the bottom. The daughter wrote
her own explanation of what determines
whether a particular type of fruit floats.
E58. You want to test whether your companys
new shampoo protects against dandruff
better than the best-selling brand.
Unfortunately, the best-selling brand has
a very distinctive color, which you are
unable to duplicate in your new shampoo.
Design a randomized paired comparison
experiment to test your shampoo. Discuss
any confounding you expect to encounter.
E59. Investigators found seven pairs of identical
twins in which one twin lived in the city
and the other in the country, and both were
willing to participate in a study to determine
how quickly their lungs cleared after they
inhaled a spray containing radioactive Teflon
particles. Display 4.23 gives the percentage of
radioactivity remaining 1 hour after inhaling
the spray.
Display 4.23 Percentage of radioactivity remaining

after 1 hour. [Source: Per Camner and Klas
Phillipson, Urban Factor and Tracheobronchial
Clearance, Archives of Environmental Health 27
(1973): 82. Reprinted in Richard J. Larson and
Morris L. Marx, An Introduction to Mathematical
Statistics and Its Applications, 2nd ed. (Englewood
Cliffs, N.J.: Prentice Hall, Inc., 1986).]
a. What is the factor? What are its levels?

Identify a block.
22-03-2009 21:01
Lesson
6 de 8
b. Is this an observational study or an

experiment? Explain.
c. Is there more variation in response within
the urban twins, within the rural twins,
or in the differences? Do you believe the
study demonstrates that environment
makes a difference in this case?
d. Why were twins used? What kind of
variation is reduced by using identical
twins?
E60. Design a taste-comparison experiment.
Suppose you want to rate three brands of
chocolate chip cookies on a scale from 1
(terrible) to 10 (outstanding). Tell how to
run this experiment as a randomized block
design.
E61. Exercise bikes. You work for a gym and
have been asked to design an experiment to
decide which of two types of exercise bikes
will get the most use. Your exercise bike
room has space for eight bikes, which you
will place as in Display 4.24. People enter
the room through the door at the top of the
diagram. A counter on each bike records the
number of hours it has been used.
Design an experiment to compare the
number of hours the two different brands of
exercise bike are used that takes into account
the fact that some bikes are in locations
that make them more likely to be used than

others when the gym isnt full.
Display 4.24 Diagram of spaces for the eight

exercise bikes.
E62. Read one of these articles from Statistics:

A Guide to the Unknown (Brooks/Cole,
2006), and write a paragraph explaining how
randomization and control or comparison
groups were used in the study reported in
the article.
1. Jennifer Hill, Evaluating School Choice
Programs
2. William F. Eddy and Margaret L. Smykla,
The Last Frontier: Understanding the
Human Mind
3. William Kahn and Leonard Roseman,
Advertising as an Engineering Science
Chapter Summary
283
22-03-2009 21:01
Lesson
7 de 8
AP1. Researchers want to estimate the mean

AP4. To conduct a survey to estimate the mean
number of children per family, for all
number of minutes adults spend exercising,
families that have at least one child enrolled
researchers stratify by age before randomly
in one of ten similarly sized county high
selecting their sample. Which of the following
schools. Which sampling plan is biased?
is not a good reason for choosing this plan?
Randomly select 100 families from those
Without stratification by age, age will be
families in the county that have at least
confounded with the number of minutes
one child in high school. Compute the
reported.
mean number of children per family.
Researchers will be sure of getting adults
Randomly select one high school, and
of all ages in the sample.
compute the mean number of children
Researchers will be able to estimate the
in each family with a child or children in
mean number of minutes for adults in
that school.
various age groups.
Randomly select 20 students from each
Adults of different ages may exercise
of the ten high schools, and ask each
different amounts on average, so
one how many children are in his or her
stratification will give a more precise
family. Compute the mean number of
estimate of the mean number of minutes
children per family.
spent exercising.
From a list of the families in all ten
All of these are good reasons.
high schools, choose a random starting
AP5.
A
movie
studio runs an experiment in order
point and then select every tenth family.
to
decide
which of two previews to use for
Compute the mean number of children
its advertising campaign for an upcoming
per family.
movie. One preview features the movies
None of the above is a biased plan.
romantic scenes and is expected to appeal
AP2. A radio program asked listeners to call in
more to women. The other preview features
and vote on whether the notorious band, The
the movies action scenes and is expected
Rolling Parameters, should perform at the
to appeal more to men. Sixteen subjects
Statistics Day celebration. Of the 956 listeners
take part in this experiment, eight women
who responded, 701 answered that The
and eight men. After viewing one of the
Rolling Parameters should perform. Which
previews, each person will rate how much
type of sampling does this example use?
he or she wants to see the movie. Which of
the following best describes how blocking
stratified random
cluster
should be used in this experiment?
systematic
quota
Use blocking, with the men in one block
voluntary response
and the women in the other.
AP3. To select students to explain homework
Use blocking, with half the men and half
problems, a teacher has students count off
the women in each block.
by 5s. She then randomly selects an integer
Do not block, because the preview that is
from 1 through 5. Every student who counted
chosen will have to be shown to audiences
off that integer is asked to explain a problem.
consisting of both men and women.
Which type of sampling plan is this?
Do not block, because the response will
convenience
systematic
be confounded with gender.
simple random
stratified random
Do not block, because the number of
cluster
subjects is too small.
22-03-2009 21:01
Lesson
8 de 8
AP6. A recent study tried to determine

whether brushing or combing hair results
in healthier-looking hair. Forty male
volunteers were randomly divided into two
groups. One group only brushed their hair
and the other group only combed their hair.
Other than that, the volunteers followed
their usual hair care procedures. After two
months, an evaluator who did not know the
treatments the volunteers used, scored each
head of hair by how healthy it looked. There
was almost no difference in the scores of
the two treatment groups. Which statement
best summarizes this study?
completely randomized with blocking

completely randomized with no blocking
randomized paired comparison
(matched pairs)
(repeated measures)
two-stage randomized
Investigative Tasks
AP9. Needle threading. With your eye firmly fixed

on winning a Nobel prize, you decide to
make the definitive study of the effect of
background color (white, black, green, or
If a male wants healthy looking hair,
red) on the speed of threading a needle with
it probably doesnt matter whether he
white thread. Design three experimentsone
brushes or combs it.
that uses no blocks, one that creates blocks by
grouping subjects, and one that creates blocks
You cant tell whether brushing
by
reusing subjects. Tell which of the three
or combing is better, because the
plans
you consider most suitable, and why.
treatments are likely to be confounded
AP10. For the 2000 U.S. Census, controversy
with variables such as which kind of
shampoo a male uses.
erupted over the Census Bureau proposal
to use sampling to adjust for the anticipated
You cant come to a conclusion, because
undercount. Here is a simplified version
the study wasnt double-blind.
of the plan: The Census Bureau collects
You cant come to a conclusion, because
the information mailed in by most of
only volunteers were included.
the residents of a region. Some residents,
The sample size is too small for any
however, did not receive forms or did not
conclusion to be drawn.
return them for some reason; these are the
uncounted persons. The bureau now selects
AP7. Which of the following is not a necessary
a sample of blocks (neighborhoods) from
component of a well-designed experiment?
the region and sends field workers find all
There must be a control group that
residents in the sampled blocks. The residents
receives a placebo.
the field workers found are matched to the
Treatments are randomly assigned to
census data, and the number of residents
experimental units.
uncounted in the original census is noted. The
census count for those blocks is then adjusted
The response variable is the same for all
according to the proportion uncounted.
treatment groups.
(For example, if one-tenth of the residents
There are a suffcient number of units in
were uncounted, the original census figures
each treatment group.
are adjusted upward by 11%.) In addition,
All units are handled as alike as possible,
the same adjustment factor is used for
except for the treatment.
neighboring regions that have characteristics
AP8. In a clinical trial, a new drug and a placebo
similar to the region sampled. Comment on
are administered in random order to each
the strengths and weaknesses of this method.
subject, with six weeks between the two
treatments. Which best describes this design?
AP Sample Test
285
22-03-2009 21:01
Lesson
1 de 2
Probability Models
Were some types of

people on the Titanic
more likely to go down
with the ship than
others? Basic concepts
of probability can help
you decide.
22-03-2009 21:02
Lesson
2 de 2
You have been using the idea of probability since Chapter 1. In your study of
Martin v. Westvaco, a question was this: If you select three workers at random for
layoff from ten workers ages 25, 33, 35, 38, 48, 55, 55, 55, 56, and 64, what is the
probability that the mean of the three ages will be 58 or more? You estimated
this probability by repeating the process of randomly drawing three ages and
computing the mean age. Because all possible sets of three ages are equally likely,
you divided the number of times you got a mean age of 58 or more by the total
number of runs in your simulation:
You had a modelselecting three workers at randomthat let you predict

how things are supposed to behave. You also had datafrom the court case
that told you how things actually did behave. The key question, then and for
the rest of this book, is this: Are the data consistent with the model? Or should
you scrap the model and look for some other way to account for the data? As in
Chapter 1, the answer will use the language of probability.
Before going any further into statistics, you need to learn more about
building and using probability models. In Chapter 1, you used simulation to get
an approximate distribution of all possible average ages when three workers are
picked at random for layoff. Sometimes, instead of doing a simulation, you can use
the rules of probability to construct exact distributions. Thats what youll learn to
do in this chapter and the next.
In this chapter, you will learn to

list all possible outcomes of a chance process in a systematic way
design simulations and use them to estimate probabilities
use the Addition Rule to compute the probability that event A or event B
(or both) occurs
use the Multiplication Rule to compute the probability that event A and
event B both occur
compute conditional probabilities, the probability that event B occurs given
that event A occurs
As you work through the chapter, you will see a number of problems
involving coins, cards, and dice. These familiar devices effectively demonstrate
fundamental principles that apply to many probabilistic situations. As you gain
understanding, youll be able to apply what youve learned to new statistical
problems.
Chapter 5 Probability Models
22-03-2009 21:02
Lesson
1 de 13
5.1
Constructing Models of Random Behavior

Food companies routinely conduct taste tests to make sure their new products
taste at least as good as the competing brand. Jack and Jill, co-owners of Downhill
Research, just won a contract to determine if people can tell tap water from
bottled water. They will give each person in their sample both kinds of water, in
random order, and ask which is the tap water.
Jack and Jill know they have a long climb ahead of them, but theyre prepared
to take it a step at a time. Before they start, they want to know what might happen
if people cant identify tap water. For example, they want to know: If we ask n
people which is the tap water and none of the people really know, what is the
probability they all will guess correctly?
Jack and Jill know several basics about probability, described here.
Fundamental Facts About Probability
You can write _

event _
not A as A ,
so P( A) = 1 P (A).
Event not A is also
sometimes written
A or AC.
An event is a set of possible outcomes from a random situation.

Probability is a number between 0 and 1 (or between 0% and 100%) that
tells how likely it is for an event to happen. At one extreme, events that
cant happen have probability 0. At the other extreme, events that are
certain to happen have probability 1.
If the probability that event A happens is denoted P( A ), then the
probability that event A doesnt happen is P( not A ) = 1 P( A ). The event
not A is called the complement of event A.
If you have a list of all possible outcomes and all outcomes are equally
likely, then the probability of a specific outcome is
1
_____________________________
the number of equally likely outcomes
and the probability of an event is
the number of outcomes in the event
_____________________________
the number of equally likely outcomes
Jack and Jill begin by formulating a model that specifies that if a person cant
identify tap water, then he or she will choose the tap water, T, with probability 0.5
and the bottled water, B, with probability 0.5. If they have only one taster, that is,
n = 1, then the probability that the person will guess correctly is written P( T ) = 0.5.
Onward and upward: What if they have two tasters, that is, n = 2? Assuming
that the tasters cant identify tap water, what is the probability that both people
will guess correctly and choose T ? Here the research stumbles.
Jack: There are three possible outcomes: Neither person chooses T, one
chooses T, or both choose T. These three outcomes are equally
288 Chapter 5 Probability Models
22-03-2009 21:02
Lesson
2 de 13
likely, so each outcome has probability . In particular, the

probability that both choose T is .
Jill: Jack, did you break your crown already? I say there are four
equally likely outcomes: first taster chooses T and the second
also chooses T ( TT ); the first chooses T and the second chooses
B( TB ); the first chooses B and the second chooses T( BT ); or both
choose B( BB ). Because these four outcomes are equally likely,
each has probability . In particular, the probability that both
choose T is , not .
A probability distribution gives all possible values resulting from a random
process and the probability of each. Display 5.1 summarizes Jills probability
distribution.
Display 5.1 Jills probability distribution for n = 2.
D1. Copy Jills table, leaving out the probabilities. Fill in new probabilities so they
represent Jacks probability distribution.
D2. What is the sum of the probabilities in a probability distribution? Why must
this be so?
D3. Who do you think is right, Jack or Jill? Give your reason(s). How could Jack
and Jill decide which of them is right?
Where Do Probabilities Come From?

To determine which of them has assigned the correct probabilities, Jack and Jill
decide to be philosophical and think about how probabilities can be assigned.
Probabilities come from three different sources:
Observed data (long-run relative frequencies). For example, observation of
thousands of births has shown that about 51% of newborns are boys. You can
use these data to say that the probability of the next newborn being a boy is
about 0.51.
Symmetry (equally likely outcomes). If you flip a coin and catch it in the air,
nothing about the physics of coin flipping suggests that one side is more likely
than the other to land facing up. Based on symmetry, it is reasonable to think
that heads and tails are equally likely, so the probability of heads is 0.5.
Subjective estimates. Whats the probability that youll get an A in this statistics
class? Thats a reasonable, everyday kind of question. The use of probability
289
22-03-2009 21:02
Lesson
3 de 13
is meaningful in this situation, but you cant gather data or list equally likely
outcomes. However, you can make a subjective judgment.
Based on symmetry, Jack and Jill agree that if n = 1, then P( T ) = 0.5. They
both used symmetry to justify their models for n = 2, saying that their outcomes
were equally likely, but they know that one of them must be wrong. To decide
which one, they decide to collect some data. They use two flips of a coin to
simulate the taste-test experiment with two tasters who cant identify tap water.
Two tails represented neither person choosing the tap water, one heads and one
tails represented one person choosing the tap water and the other choosing the
bottled water, and two heads represented both people choosing the tap water.
Hours later, they have their results, which are given in Display 5.2.
Display 5.2 Results of 3000 simulations for two tasters when

P (T ) 0.5.
[You can use calculator lists to quickly calculate relative frequencies for a frequency
table. See Calculator Note 5A.]
Jack: Whew! Now I see you must be right. The relative frequencies from
the simulation match your probabilities fairly well. I admit I fell
down a bit here. I forgot that there are two ways to get one person
choosing correctly first chooses correctly and the second
doesnt, or the second chooses correctly and the first doesnt.
There is only one way for two people to choose correctly: first
person chooses correctly and the second person chooses correctly.
Jill: Now that this has been settled, lets try to construct the probability
distribution for three people, or n = 3.
Jack: Let me redeem myself. Following your reasoning that different
orders should be listed separately, there are eight possible outcomes.
22-03-2009 21:02
Lesson
4 de 13
Jill: Ill count the number of people in each of the eight equally likely
outcomes who picked the tap water. That gives us the probability
distribution in Display 5.3. The probability that all three people
choose correctly is .
Display 5.3 Jack and Jills probability distribution for n = 3.
Jack: This makes me think we need a larger sample size. Even if nobody
can identify tap water, there is a one-eighth chance everyone will
choose correctly! The reputation of Downhill Research is at stake!
Sample Spaces
Jack and Jill both used the same principle: Start by making a list of possible
outcomes. Over the years, mathematicians realized that such a list of possible
outcomes, called a sample space, must satisfy specific requirements.
A sample space for a chance process is a complete list of disjoint outcomes. All
of the outcomes in a sample space must have a total probability equal to 1.
Complete means that every possible outcome is on the list. Disjoint means
that two different outcomes cant occur on the same opportunity. Sometimes
the term mutually exclusive is used instead of disjoint. This book will alternate
between the two terms so you can get used to both of them.
Deciding whether two outcomes are disjoint sounds easy enough, but be
careful. You have to think about what an outcome means in your situation. If your
outcome is the result of a single coin flip, your sample space is heads (H ) and tails
(T ). These two outcomes are mutually exclusive because you cant get both heads
and tails on a single flip. But suppose you are thinking about what happens when
you flip a coin three times. Now your sample space includes outcomes like HHT
and TTT. Even though you get tails on the third flip in both HHT and TTT, these
are disjoint outcomes. Your sample space consists of triples of flips, and these
arent the same triple.
Jacks sample spaceneither person chooses T, one chooses T, two choose
Thas outcomes that are complete and disjoint, but they arent equally likely. He
has a legitimate sample space; he has simply assigned the wrong probabilities.
Sample Spaces
D4. In a slightly different form, Jack and Jills argument holds an important
place in the early history of probability. One famous French mathematician,
291
22-03-2009 21:02
Lesson
5 de 13
Jean dAlembert (171783), wrote in a 1750s text that the probability of

getting two heads in two flips of a fair coin is . Heres his list of outcomes,
which he said were equally likely:
The first flip is tails.
The first flip is heads, the second is tails.
The first flip is heads, the second is heads.
Is his list of outcomes complete? Are the outcomes disjoint? Are the
outcomes equally likely?
Data and Symmetry

How can you tell if the outcomes in your sample space are equally likely? Theres
no magic answer, only two useful principlesdata and symmetry. A die has
six sides, and the only thing that makes one side different from another is the
number of dots. Its hard to imagine that the number of dots would have much
effect on which side lands facing up, so its reasonable to assume that the six sides
are equally likely to land facing up. This seems like a good model for a die, but
to verify it, you need data: Roll the die many times and compare your models
predictions with the actual results to see if you have a good fit.
In Jack and Jills model, they are temporarily assuming that symmetry
appliesthat people are as likely to choose the tap water as the bottled water. In
fact, its probably not the right model. When Jack and Jill collect data from enough
real people, they might be able to reject it. Or perhaps the model is correct and
they wont be able to reject it. In either case, they will have done what they were
hired to dodecide if people can distinguish the tap water from the bottled water.
Activity 5.1a will help you see some of the difficulties involved in coming up
with a realistic probability model.
Spinning Pennies
What youll need: one penny per student
In flipping a penny, heads and tails have the same probability. Is the same
probability model true for spinning a penny?
1. Use one finger to hold your penny on edge
on a flat surface, with Lincolns head right
side up, facing you. Flick the penny with
the index finger of your other hand as you
let go so that it spins around many times
on its edge. When it falls over, record
whether it lands heads up or tails up.
2. Repeat, and combine results with the rest of your class until you have a
total of at least 500 spins.
3. Are the data consistent with a model in which heads and tails are equally
likely outcomes, or do you think the model can safely be rejected?
(continued)
22-03-2009 21:02
Lesson
6 de 13
4. Suppose you spin a penny three times and record whether it lands heads up
or tails up.
a. How many possible outcomes are there?
b. Are these outcomes equally likely? If not, which is most likely? Least
likely?
The Law of Large Numbers
The Law of Large

Numbers guarantees
that sample proportions
converge to true
probabilities when
sampling is random.
Why was Jack convinced that his probability model was wrong after simulating the
taste test with 3000 flips of two coins (Display 5.2 on page 290)? In Activity 5.1a,
why might you be convinced that the equal probability model does not work for
coin spinning? Results from observed data seem to preclude the proposed model.
That is as it should be, according to a mathematical principlethe Law of Large
Numbersestablished in the 1600s by Swiss mathematician Jacob Bernoulli
(16541705).
The Law of Large Numbers says that in random sampling, the larger the
sample, the closer the proportion of successes in the sample tends to be to the
proportion in the population. In other words, the difference between a sample
proportion and the population proportion must get smaller (except in rare
instances) as the sample size gets larger. You can see a demonstration of this law
in Display 5.4. Think of spinning a coin in such a way that it comes up heads
with probability 0.4. The graph in Display 5.4 shows a total of 150 spins, with
the proportion of heads accumulated after each spin plotted against the number
of the spin. Notice that the coin came up heads for the first few spins, so the
proportion of heads starts out at 1. This proportion quickly decreases, however, as
the number of spins increases, and it ends up at about 0.377 after 150 spins.
Display 5.4 Proportion of heads for the given number of spins of

a coin, with P (heads) 0.4.
[You can use a calculator to perform this experiment yourself. See Calculator Note 5B.]
Most people intuitively understand the Law of Large Numbers. If they want
to estimate a proportion, they know it is better to take a larger sample than
a smaller one. After Jack saw the results from 3000 flips of two coins, he
immediately rejected his model that there are three equally likely outcomes. If
there had been only 10 pairs of coin flips, he couldnt have been so sure that his
model was wrong.
293
22-03-2009 21:02
Lesson
7 de 13

D5. An opinion pollster says, All I need to do to ensure the accuracy of the
results of my polls is to make sure I have a large sample.
A casino operator says, All I need to do to ensure that the house will win
most of the time is to keep a large number of people flocking into my
casino.
A manufacturer says, All I need to do to keep my proportion of defective
light bulbs low is to manufacture a lot of light bulbs.
Comment on the correctness of each of these statements. In particular, do
the people speaking understand the Law of Large Numbers?
D6. Flip a coin 20 times, keeping track of the cumulative number of heads and
the cumulative proportion of heads after each flip. Plot the cumulative
number of heads versus the flip number on one graph. Plot the cumulative
proportion of heads versus the flip number on another graph. On each
graph, connect the points. Repeat this process five times so that you have
five sets of connected points on each plot. Comment on the differences in
the patterns formed by the five lines on each graph.
The Fundamental Principle of Counting

Jack and Jill are now ready to tackle samples of size 5. They are beginning to
realize that using Jills method will result in a long list of possible outcomes,
and they have become a bit discouraged. The key is for them to notice these
two things:
1. They can always think of asking their question in stages (one person
at a time).
2. They can list the possible outcomes at each stage.
These two observations will let them count outcomes by multiplying. To see how,
note that Jack and Jill also could have used a tree diagram like the one in Display
5.5 to count the number of outcomes for n = 2.
Display 5.5 A tree diagram of all possible outcomes for n = 2.
22-03-2009 21:02
Lesson
8 de 13
There are two main branches, one for each of the two ways the first person
can answer. Each main branch has two secondary branches, one for each way
the second person can answer after the first person has answered. In all, there
are 2 2, or 4, possible outcomes.
For n = 3, you would add two branches to the end of each of the four
branches in Display 5.5. This would give a total of 2 2 2, or 8, possible
outcomes, as Jack and Jill already discovered. This idea is called the
Fundamental Principle of Counting.
Fundamental Principle of Counting
For a two-stage process with n1 possible outcomes for stage 1 and n2 possible
outcomes for stage 2, the number of possible outcomes for the two stages taken
together is n1 n2.
More generally, if there are k stages, with ni possible outcomes for
stage i, then the number of possible outcomes for all k stages taken together
is n1 n2 n3 . . . nk.
With five people, or n = 5, and two possible outcomes for each person, Jack
and Jill have 2 2 2 2 2, or 32, possible outcomes. If their model is correct
that is, if people are just guessing which is the tap waterthese 32 outcomes
are all equally likely. So the probability that each of the five people will correctly
choose the tap water is
When a process has only two stages, it is often more convenient to list them
using a two-way table.
Example: Rolling Two Dice

Make a two-way table that shows all possible outcomes
when you roll two fair dice. What is the probability that
you get doubles (both dice show the same number)?
Solution
Because there are six faces on each die and the dice are fair, there are 6 6, or 36,
equally likely outcomes, as shown in Display 5.6. Six of these outcomes are doubles,
so the probability of rolling doubles is
Display 5.6 The 36 outcomes when rolling two dice.
295
22-03-2009 21:02
Lesson
9 de 13

D7. Suppose Jack and Jill have five people taste three well-known brands of cola.
The people are asked if they prefer the first cola, the second cola, or the third
cola or cant tell any difference. How many possible outcomes will be in Jack
and Jills list for this taste test? Are these outcomes equally likely?
D8. Suppose you flip a fair coin seven times.
a. How many possible outcomes are there?
b. What is the probability that you will get seven heads?
c. What is the probability that you will get heads six times and tails once?
Summary 5.1: Constructing Models of Random

Behavior
A probability model is a sample space together with an assignment of
probabilities. The sample space is a complete list of disjoint outcomes for which
these properties hold:
Each outcome is assigned a probability between 0 and 1.
The sum of all the probabilities is 1.
Often you can rely on symmetry to recognize that outcomes are equally likely.
If they are, then you can compute probabilities simply by counting outcomes:
The probability of an event is the number of outcomes that make up the event
divided by the total number of possible outcomes. In statistics, the main practical
applications of equally likely outcomes are in the study of random samples and
in randomized experiments. As you saw in Chapter 4, in a survey, all possible
simple random samples are equally likely. Similarly, in a completely randomized
experiment, all possible assignments of treatments to units are equally likely.
The only way to decide whether a probability model is a reasonable fit to a real
situation is to compare probabilities derived from the model with probabilities
estimated from observed data.
The Fundamental Principle of Counting says that if you have a process
consisting of k stages with ni outcomes for stage i, the number of outcomes for all
k stages taken together is n1 n2 n3. . . nk.
Practice
Where Do Probabilities Come From?
P1. Suppose Jack and Jill use a sample of four
people who cant tell the difference between
tap water and bottled water.
a. Construct the probability distribution
for the number of people in the sample
who would choose the tap water just
by chance.
b. What is the probability that all four

people will identify the tap water
correctly?
c. Is four people a large enough sample to
ease Jacks concern about the reputation
of Downhill Research?
22-03-2009 21:02
Lesson
10 de 13
P2. Display 5.7 gives the actual low temperature

(to the nearest 5F) in Oklahoma City on
days when the National Weather Service
forecast was for a low temperature of 30F.
b. Are all outcomes in your sample space

equally likely?
c. What is the probability that the two
youngest people are the ones laid off?
d. What is the probability that the mean age
of those laid off is 40 or more?
P5. Jean dAlembert (whom you met in D4)
was coauthor (with another Frenchman,
Denis Diderot, 171384) of a 35-volume
Encyclopdie. In it, he wrote that the
probability of getting heads at least once
in two flips of a fair coin is . This time, he
Display 5.7 Temperatures in Oklahoma City on days
said that these three outcomes were equally
with a forecasted low temperature of
likely:
30F. [Source: Harold E. Brooks et al., Verification
heads on the first flip
of Public Weather Forecasts Available via the
Media, Bulletin of the American Meteorological
heads on the second flip
Society 78 (October 1997): 216777.]
heads on neither flip
a. Suppose the forecast for tomorrow is for
a.
Is
this list of outcomes complete?
a low temperature of 30F. What is your
b.
Are
the outcomes disjoint?
estimate of the probability that the low
temperature really will be approximately
c. Are the three outcomes equally likely?
30F?
d. Is dAlembert correct about the
b. Using these data, make a table that gives
probability of getting heads at least once?
the estimated probability distribution
for the actual low temperature when the
forecast is 30F.
c. Does the method of forecasting appear to
give a prediction that tends, on average,
to be too warm, or too cold?
Sample Spaces
P3. Suppose flip a coin and then roll a die.
If you get heads and a 3, then your outcome
is H3.
a. List a sample space that has outcomes
that are disjoint and complete.
b. Are all outcomes in your sample space
equally likely?
c. What is the probability that you get heads
and a 3?
P4. You randomly choose two workers to be laid
off from a group of workers ages 28, 35, 41,
47, and 55.
a. List a sample space that has outcomes
that are disjoint and complete.
Jean dAlembert (171783)
Data and Symmetry

P6. Suppose you pick four students at random
from your school and check whether they
are left-handed or right-handed.
a. Can you list a sample space?
b. Can you determine the probability that
all four students are right-handed?
297
22-03-2009 21:02
Lesson
11 de 13
dataset link:

P7. The results of 50 spins of a penny are plotted
in Display 5.8. The horizontal axis gives the
number of the spin, and the vertical axis
gives the cumulative proportion of spins so
far that were heads.
a. Was the first spin heads or tails? The
second? The third? The fourth? The 50th?
b. Use these data to estimate the probability
that this penny will land heads up when it
is spun.

P8. Suppose you ask a person to taste a
particular brand of strawberry ice cream and
evaluate it as good, okay, or poor on flavor
and as acceptable or unacceptable on price.
a. Show all possible outcomes on a tree
diagram.
b. How many possible outcomes are there?
c. Are all the outcomes equally likely?
P9. A dental clinic has three dentists and seven
dental hygienists.
a. If you are assigned a dentist and a dental
hygienist at random when you go in, how
many different pairs could you end up
with?
b. What is the probability that you get your
favorite dentist and your favorite dental
hygienist?
c. Illustrate your answer in part a with a
two-way table.
d. Illustrate your answer in part a with a
tree diagram.
Display 5.8 Results of 50 spins of a penny.
Exercises
E1. Suppose you flip a coin five times and count E3. Refer to the sample space for rolling two
dice shown in Display 5.6 on page 295.
the number of heads.
Determine each of these probabilities.
a. List all possible outcomes.
a. not getting doubles
b. Make a table that gives the probability
distribution for the number of heads.
b. getting a sum of 5
c. What is the probability that you get at
c. getting a sum of 7 or 11
most four heads?
d. a 5 occurring on the first die
E2. Refer to the sample space for rolling two dice
e. getting at least one 5
shown in Display 5.6 on page 295. Make a
f. a 5 occurring on both dice
table that gives the probability distribution
g. the larger number is a 5 (if you roll
for the sum of the two dice. The first column
doubles, the number is both the smaller
should list the possible sums, and the second
number and the larger number)
column should list their probabilities.
h. the smaller number is a 5
i. the difference of the larger number and
the smaller number is 5
22-03-2009 21:02
Lesson
12 de 13
E4. A tetrahedral die has four sides, with the

numbers 1, 2, 3, and 4.
a. How many equally likely outcomes are

there when two fair tetrahedral dice are
rolled?
b. Using a two-way table, show all possible
equally likely outcomes.
c. What is the probability of getting doubles?
d. What is the probability that the sum is 2?
e. What is the probability that the larger
number is a 2? (If you roll doubles, the
number is both the smaller number and
the larger number.)
f. What is the probability that the smaller
number is a 2?
g. What is the probability that the larger
number minus the smaller number is
equal to 2?
h. What is the probability that the smaller
number is less than 2?
E5. Suppose you roll a tetrahedral die twice
(see E4). Determine whether each proposed
sample space is disjoint and complete. If the
sample space is disjoint and complete, assign
probabilities to the given outcomes. If it is
not, explain why not.
a. {no 4 on the two rolls, one 4, two 4s}
b. {the first roll is a 4, the second roll is a 4,
neither roll is a 4}
c. {the first 4 comes on the first roll, the first
4 comes on the second roll}
d. {the sum of the two rolls is less than 2, the
sum is more than 2}
e. {the first roll is a 4, neither roll is a 4}
E6. Data or symmetry?
a. A taste test will be conducted to see if
people prefer the taste of food A or food
B. In fact, food A is identical in taste

to food B. Is there any way to tell from
symmetry alone (as opposed to needing
data) that food A will be preferred over
food B half the time? Give an argument
in support of your answer.
b. Is there any way to tell from symmetry (as
opposed to data) that a randomly selected
mathematics major will be female with
probability about 0.5? Give an argument
in support of your answer.
E7. The results of 50 rolls of a pair of dice are
plotted in Display 5.9. The horizontal axis
gives the number of the roll, and the vertical
axis gives the cumulative proportion of rolls
so far that were doubles.
Display 5.9 Cumulative proportion of doubles for

50 rolls of a pair of dice.
a. Was the first roll a double? The second?

The third? The fourth?
b. How many rolls were doubles?
c. Use this sample to estimate the
probability of rolling doubles with a
pair of dice.
d. Why do the lengths of the line segments
tend to get shorter as the number of spins
increases?
E8. Business is slow at Downhill Research. Jack
and Jill decide to investigate the probability
that a penny they found will land heads up
when spun.
a. Jack spins the penny 500 times and gets
227 heads. What is his estimate of the
299
22-03-2009 21:02
Lesson
13 de 13
probability that the penny will land heads

up when spun?
b. Jill spins the penny 50 times. Would you
expect Jack or Jill to have an estimate
closer to the true probability of the penny
landing heads up? Why?
c. Jack and Jill now flip the penny 10,000
times. Their results are recorded
in Display 5.10. Fill in the missing
percentages. Do their results illustrate the
Law of Large Numbers? Explain.
Display 5.10 Percentage of heads for various

numbers of coin flips.
d. Does the number of heads get closer to

or farther away from half the number of
flips?
E9. Miguel is designing an experiment to test
which combination of colors makes the type
on computer screens easiest to read. His two
factors are the color of the background (blue,
green, yellow, white, or beige) and the color
of the type (brown, black, navy, or gray).
a. How many different possible treatments
are there?
b. Show the possible treatments on a tree
diagram.
c. Suppose Miguel adds a third factor,
brightness, to his experiment. He will
have two levels of brightness on the
screen, high and low. How many possible
distinct treatments are there?
d. Make a tree diagram showing this new
situation.
E10. Tran has six shirts (blue, green, red, yellow,
and two whiteone long-sleeved, one shortsleeved) and four pairs of pants (brown,
black, blue, and gray).
a. Use the Fundamental Principle of
Counting to find the number of possible
outfits Tran can wear.
b. Show all possible outfits in a two-way

table.
c. If Tran selects a shirt and a pair of pants
at random, determine the probability of
Trans wearing
i. a white shirt
ii. the gray pants
iii. a white shirt and the gray pants
iv. a white shirt or the gray pants
E11. In a Culver City, California election, the
two candidates for a seat on the Board of
Education tied with 1141 votes each. To
avoid a long and costly run-off election, the
city code called for a chance process to select
the winner. The city officials made up two
bags. One candidate was given a bag that
contained eight red marbles and one white
marble. The other candidate was given a bag
that contained eight blue marbles and one
white marble. Simultaneously, the candidates
drew a marble. If neither marble was white,
the candidates continued to draw marbles
simultaneously, one marble at a time. The
first candidate to draw a white marble was
the winner. The winner drew a white marble
on the fourth round. [Source: Los Angeles Times,
December 2, 2003, pages B1, B7.]
a. Is this a fair process?

b. How could this process once again result
in a tie? Whats the chance of a tie on the
first round?
c. How could you change the process to
reduce the probability of a tie?
E12. You have 27 songs on your MP3 player and
randomly select one to play. Its your favorite.
When it finishes, you again randomly select
a song to play, and its the same one!
a. What is the chance of this happening?
b. What is the chance that both random
selections will be the same song?
E13. Equally likely outcomes?
a. Use the Fundamental Principle of
Counting to find the number of outcomes
for the situation of flipping a fair coin six
times. Can you find the probability that
you will get heads all six times?
22-03-2009 21:02
Lesson
1 de 14
b. Use the Fundamental Principle of

Counting to find the number of outcomes
for the situation of rolling a fair die six
times. Can you find the probability that
you will get a 3 all six times?
c. Use the Fundamental Principle of
Counting to find the number of
outcomes for the situation of picking
1200 U.S. residents at random and asking

if they go to school. Can you find the
probability that all 1200 will say yes?
E14. How many three-digit numbers can you
make from the digits 1, 2, and 7? You can use
the same digit more than once. If you choose
the digits at random, what is the probability
that the number is less than 250?
Using Simulation to Estimate Probabilities

In Activity 1.1a, you used simulation to estimate a probability. You wrote the
ages of ten hourly workers on slips of paper, drew three slips to represent those
workers to be laid off, and computed their average age. After repeating this
process many times, you were able to estimate the probability of getting, just by
chance, an average age as large as or even larger than the actual average age. In
this section, you will learn a more efficient method of conducting a simulation,
using random digits.
A table of random digits is a string of digits that is constructed in such a way
that each digit has probability
of being 0, 1, 2, 3, 4, 5, 6, 7, 8, or 9. Further, each
digit is selected independently of the previous digits. A table of random digits
appears in Table D on page 828 of this book. When you need a string of random
digits, you can use this table or you can generate a string of digits using your
calculator. [See Calculator Note 4A.]
These steps, and the five examples that follow, illustrate how to use a table of
random digits to estimate answers to questions about probability.
The Steps in a Simulation That Uses Random Digits

1. Assumptions. State the assumptions you are making about how the real-life
situation works. Include any doubts you might have about the validity of
your assumptions.
2. Model. Describe how you will use random digits to conduct one run of a
simulation of the situation.
Make a table that shows how you will assign a digit (or a group of digits)
to represent each possible outcome. (You can disregard some digits.)
Explain how you will use the digits to model the real-life situation. Tell
what constitutes a single run and what summary statistic you will record.
3. Repetition. Run the simulation a large number of times, recording the
results in a frequency table. You can stop when the distribution doesnt
change to any significant degree when new results are included.
4. Conclusion. Write a conclusion in the context of the situation. Be sure to
say that you have an estimated probability.
301
22-03-2009 21:03
Lesson
2 de 14
Example: Hourly Workers at Westvaco

The ages of the ten hourly workers involved in Round 2 of the layoffs were 25, 33,
35, 38, 48, 55, 55, 55, 56, and 64. The ages of the three workers who were laid off
were 55, 55, and 64, with average age 58. Use simulation with random digits to
estimate the probability that three workers selected at random for layoff would
have an average age of 58 or more.
Solution
1. Assumptions.
You are assuming that each of the ten workers has the same chance of being
laid off and that the workers to be laid off are selected at random without
replacement.
2. Model.
Assign each worker a random digit as in Display 5.11.
Display 5.11 Assignment of random numbers to

Westvaco workers.
When using a table of

random digits, always
start at a random spot.
Start at a random place in a table of random digits and look at the next
three digits. Those digits will represent the workers selected to be laid off
in a single run of the simulation. Because the same person cant be laid
off twice, if a digit repeats, ignore it and go to the next digit. Find the
average of the ages of the three workers laid off.
3. Repetition.
Suppose the string of random digits begins like this:
32416 15000 56054
In the first run of the simulation, the digits 3, 2, and 4 represent the workers
ages 35, 33, and 38. Record their average age, 35.33.
In the second run of the simulation, the digits 1, 6, and 5 represent the
workers laid off. The second 1 was skipped because that worker was already
selected in this run. The ages are 25, 55, and 48. Record the average age, 42.67.
22-03-2009 21:03
Lesson
3 de 14
In the third run of the simulation, the digits 0, 5, and 6 represent the workers
laid off. The second and third 0s were skipped because that worker was already
selected in this run. The ages are 64, 48, and 55. Record the average age, 55.67.
Display 5.12 shows the results of 2000 runs of this simulation, including the
three described.
Display 5.12 Results of 2000 runs of the layoff of three workers,

using a table of random digits.
4. Conclusion.
From the histogram, the estimated probability of getting an average age of
58 years or more if you pick three workers at random is
or 0.045. This
probability is fairly small, so it is unlikely that the process Westvaco used for
layoffs in this round was equivalent to picking the three workers at random.
According to the Law of Large Numbers, the more runs you do, the closer
you can expect your estimated probability to be to the theoretical probability.
The simulation in the previous example contained 2000 runs. When you do a
simulation on a computer, there is no reason not to do many thousands of runs.
Example: Hand Washing in Public Restrooms

The American Society of Microbiology periodically estimates the percentage of
people who wash their hands after using a public restroom. (Yes, these scientists
spy on people in public restrooms, but its for a good cause.) They conducted the
first study in 1996, when 67% of people washed their hands. [Source: Sports Illustrated,
Less Chop, More Soap, September 22, 2005, www.SI.com.] Suppose you watched four
randomly selected people using a public restroom back in 1996. Use simulation to
estimate the probability that all four washed their hands.
Solution
1. Assumptions.
You are assuming that the four people were selected at random and
independently from the population of restroom users. That is, you didnt
sample by doing something like taking the next four people in one randomly
selected public restroom.
You also are assuming that 67% is the true percentage of all restroom users
who wash their hands. This assumption may or may not be true, because the
percentage was estimated from a (large) sample.
303
22-03-2009 21:03
Lesson
4 de 14
Summarizing, the situation to be modeled is taking a random sample of

size 4 from a large population with 67% hand-washers and 33%
non-hand-washers.
2. Model.
Because there are two decimal places in 0.67, group the random digits into
pairs. The pairs will be assigned to outcomes as in Display 5.13.
Display 5.13 Assignment of random numbers to hand-washers

and non-hand-washers.
Look at four pairs of random digits. Count and record how many of the
four pairs of digits represent hand-washers. This time, if a pair of digits
repeats, use it again because each pair of digits doesnt represent a specific
person as in the previous example.
3. Repetition.
01420 94975 89283 40133 48486
In the first run of the simulation, the pairs 01, 42, 09, and 49 represent
hand-washer, hand-washer, hand-washer, and hand-washer. Record a 4, for
four hand-washers in this run.
In the second run of the simulation, the pairs 75, 89, 28, and 34 represent
non-hand-washer, non-hand-washer, hand-washer, and hand-washer. Record
a 2, for the two hand-washers in this run.
In the third run of the simulation, the pairs 01, 33, 48, and 48 represent four
hand-washers. Record a 4, for the four hand-washers in this run. Note that
the 48 was used twice in the same run because it doesnt represent only a
single hand-washer.
The frequency table and histogram in Display 5.14 give the results of 10,000
runs of this simulation, including the three already described.
Display 5.14 Results of 10,000 runs of the hand-washer

simulation.
22-03-2009 21:03
Lesson
5 de 14
<
4. Conclusion.
From the frequency table, the estimated probability that all four randomly
selected people will wash their hands is
or 0.197.
Example: Men and Women and Hand Washing

Things have improved. The American Society of Microbiology now reports that
75% of men and 90% of women wash their hands after using a public restroom.
[Source: www.harrisinteractive.com.] Suppose you pick a man and a woman at random as
they use a public restroom. Estimate the probability that both wash their hands.
Solution
1. Assumptions.
You are assuming that the man and the woman were selected randomly and
independently from the population of restroom users.
Your second assumption is that 75% and 90% are the true percentages.
That assumption may or may not be true, because those percentages were
estimated from a (large) sample.
2. Model.
In this example, you will learn how a calculator can be used to generate
pairs of random integers to simulate this situation. Because there are two
decimal places in 0.75 and 0.90, youll generate two integers from 0 to 99.
The first integer will represent a man, and the second will represent a
woman. The integers will be assigned to outcomes as in Display 5.15.
Display 5.15 Assignment of random integers to male and female

hand-washers and non-hand-washers.
Record whether the first integer represents a man who washed or didnt
wash and whether the second integer represents a woman who washed or
didnt wash. If an integer repeats, use it again.
3. Repetition.
A calculator gives this sequence of pairs of random integers. [See Calculator
Note 4A to learn how to generate pairs of random integers.]
305
22-03-2009 21:03
Lesson
6 de 14
In the first run of the simulation, the integers 72 and 43 represent a man who
washed and a woman who washed.
In the second run of the simulation, the integers 6 and 81 represent a man
who washed and a woman who washed.
In the third run of the simulation, the integers 76 and 81 represent a man
who didnt wash and a woman who washed.
Display 5.16 gives the results of 5000 runs of this simulation, including
the three already described. [See Calculator Note 5C to learn how to use your
calculator to do many runs of a simulation and report the results.]
Display 5.16 Results of 5000 runs of the male/female

hand-washer simulation.
4. Conclusion.
From the frequency table, the estimated probability that both the randomly
selected man and the randomly selected woman wash their hands is
or
about 0.67.
Example: Waiting for a Teen Who Flunks a Treadmill Test

About one-third of teenagers show a poor level of cardiovascular fitness on an
8-minute treadmill test. [Source: Journal of the American Medical Association, December 21, 2005,
in Los Angeles Times, January 26, 2006, page F7.]
Suppose you want to interview a teen who flunks a treadmill test about his
or her health habits. On average, to how many teens would you have to give a
treadmill test before you find one who flunks?
Solution
1. Assumptions.
You are assuming that you are testing randomly and independently selected
teens. You also are assuming that one-third is the true proportion of teens
who would flunk a treadmill test.
2. Model.
Assign the random digits as shown in Display 5.17. Ignore the digits
4 through 9 and 0.
Display 5.17 Assignment of random numbers to teens who flunk

or pass a treadmill test.
22-03-2009 21:03
Lesson
7 de 14
Look at random digits until you find a 1, which represents a teen who
flunks the treadmill test. Record how many digits you have to look at in
order to get a 1. Dont count any digits 4 through 9 or 0.
3. Repetition.
22053 71491 32923 71926
In the first run of the simulation, the digit 2 represents a teen who passes
the test. The second digit also represents a teen who passes. Skip the 0 and
the 5. The 3 also represents a teen who passes. Skip the 7. The 1 represents
a teen who flunks. Record a 4, because you had to test four teens to find
one who flunked.
In the second run of the simulation, skip the 4 and the 9. The 1 represents a
teen who flunks the test. Record a 1, because you had to test only one teen to
find one who flunked.
In the third run of the simulation, the 3 and the 2 represent teens who
pass. Skip the 9. The 2 and the 3 represent teens who pass. Skip the 7. The 1
represents a teen who flunks. Record a 5, because you had to test five teens to
find one who flunked.
The histogram and frequency table in Display 5.18 give the results of 3000
runs of this simulation, including the three described.
To compute the average number of teens who have to be tested, use the
formula from page 67. The average is about 3.
4. Conclusion.
The estimate from this simulation is that, on average, you will have to test
about three teens to find one who flunks the treadmill test.
Display 5.18 Results of 3000 runs of the treadmill simulation.
307
22-03-2009 21:03
Lesson
8 de 14
Example: Collecting Blends of Coffee

A coffee house rotates daily among six different blends of coffee. Suppose Sydney
goes into the coffee house on three randomly selected days. Estimate the probability
that the coffee house will be offering a different blend on each of the three days.
Solution
1. Assumptions.
You are assuming that each time Sydney goes into the coffee house, the
probability is
that any given blend will be offered and that her three trips are
selected independently of one another.
2. Model.
Assign the digits 1 through 6 to the six blends of coffee, as in Display 5.19.
Display 5.19 Assignment of random numbers to blends of coffee.
This time you will practice using a calculator to generate three random
integers between 1 and 6. Record whether the three digits are different.
3. Repetition.
A calculator gives these sets of random integers. [See Calculator Note 4A.]
The first run of the simulation resulted in 1, 1, and 2. Because Sydney got
Blend 1 twice, record that the three blends werent all different.
In the second run of the simulation, the three numbers are different, so
record that Sydney got three different blends.
In the third run of the simulation, the three numbers are different, so again
record that Sydney got three different blends.
Display 5.20 gives the results of 2000 runs of this simulation.
Display 5.20 Results of 2000 runs of the coffee blend simulation.
22-03-2009 21:03
Lesson
9 de 14
4. Conclusion.
From the frequency table, the estimated probability that Sydney will get
three different blends is
or about 0.56.

D9. What characteristics would you expect a table of random digits to have?
D10. Jason needs a table of random digits with the digits selected from 1 to 6. He
has a fair six-sided die that has the digits 1 through 6 on its faces. To construct
his table, he rolls the die and writes down the number that appears on the
top of the die for the first random digit. To speed things up, Jason writes the
number that appears on the bottom of the die for the next random digit. For
example, if a 2 appears on the top of the die, a 5 would be on the bottom, so
Jasons sequence of digits would begin 2 5. He continues in this way.
a. Does each of the digits 1 through 6 have an equally likely chance of
appearing in any given position in Jasons table?
b. Are the digits selected independently of previous digits?
c. Has Jason constructed a table of random digits?
D11. For the example Hand Washing in Public Restrooms, on page 303, describe
another way the random digits could have been assigned using pairs of
random digits. Describe a way that uses triples of random digits.
D12. Describe how to use a die to simulate the situation in the coffee house
example.
Summary 5.2: Using Simulation to Estimate

Probabilities
In this section, you have learned to model various situations involving probability
and to estimate their solutions using simulation. Some of the problems you worked
on were amusing, but dont let this mislead you. Simulation is an important method
of estimating answers to problems that are too complicated to solve theoretically.
You have used random digits to model situations because it is quick and
easy to generate random digits on calculators and computers. However, many
situations may be easily modeled using other chance devices such as coins, dice,
or spinners. The key to designing a simulation is to be clear about what an
outcome on the chance device corresponds to in the real-life situation.
When you solve a problem using simulation, always include these four steps:
1. State your assumptions.
2. Describe how you will use random digits to conduct one run of the
simulation.
3. Run the simulation a large number of times, recording the results in a
frequency table.
4. Write a conclusion in the context of the situation.
309
22-03-2009 21:03
Lesson
10 de 14
Practice
P10. How would you use a table of random digits
to conduct one run of a simulation of each
situation?
a. There are eight workers, ages 27, 29, 31,
34, 34, 35, 42, and 47. Three are to be
chosen at random for layoff.
b. There are 11 workers, ages 27, 29, 31, 34,
34, 35, 42, 42, 42, 46, and 47. Four are to
be chosen at random for layoff.
For P11P13, complete ad.
a. Assumptions. State your assumptions.
b. Model. Make a table that shows how you
are assigning the random digits to the
outcomes. Explain how you will use the
digits to model the situation and what
summary statistic you will record.
c. Repetition. Conduct ten runs of the
simulation, using the specified row of
Table D on page 828. Add your results to
the frequency table given in the practice
problem.
d. Conclusion. Write a conclusion in the
context of the situation.
P11. Researchers at the Macfarlane Burnet
Institute for Medical Research and Public
Health in Melbourne, Australia, noticed that
the teaspoons had disappeared from their
tearoom. They purchased new teaspoons,
numbered them, and found that 80%
disappeared within 5 months. [Source: Megan S. C.
Display 5.21 Results of 4990 runs of the

disappearing teaspoons.
P12. A catastrophic accident is one that involves

severe skull or spinal damage. The National
Center for Catastrophic Sports Injury
Research reports that over the last 21 years,
there have been 101 catastrophic accidents
Lim, Margaret E. Hellard, and Campbell K. Aitken, The Case
among female high school and college
of the Disappearing Teaspoons, British Journal of Medicine
athletes. Fifty-five of these resulted from
331 ( December 2005 ): 14981500, bmj.bmjjournals.com.]
cheerleading. [Source: www.unc.edu.]
Suppose that 80% is the correct probability
Suppose you want to study catastrophic
that a teaspoon will disappear within
accidents in more detail, and you take a
5 months and that this group purchases
random sample, without replacement, of 8 of
ten new teaspoons. Estimate the probability
these 101 accidents. Estimate the probability
that all the new teaspoons will be gone in
that at least half of your eight sampled
5 months.
accidents resulted from cheerleading.
Start at the beginning of row 34 of Table D
on page 828, and add your ten results to the
on page 828, and add your ten runs to the
frequency table in Display 5.21, which gives
the results of 4990 runs.
22-03-2009 21:03
Lesson
11 de 14
P13. The winner of the World Series of baseball is

the first team to win four games. That means
the series can be over in four games or can
go as many as seven games. Suppose the two
teams playing are equally matched. Estimate
the probability that the World Series will go
seven games before there is a winner.
on page 828, and add your ten runs to the
Display 5.23 Results of 4990 runs of the number of

games in the World Series.
Display 5.22 Results of 990 runs of sampling eight

catastrophic accidents.
Exercises
For E15E20, complete ad.
b. Model. Make a table that shows how
you are assigning the random digits
to the outcomes. Explain how you will
use the digits to conduct one run of the
simulation and what summary statistic
you will record.
c. Repetition. Conduct ten runs of the
Table D on page 828. Add your results to
the frequency table given in the exercise.
E15. About 10% of high school girls report that
they rarely or never wear a seat belt while
riding in motor vehicles. Suppose you
randomly sample four high school girls.
Estimate the probability that no more than
one of the girls says that she rarely or never

wears a seat belt. [Source: www.nhtsa.dot.gov.]

girls, out of four, who rarely or never
wear a seat belt.
311
22-03-2009 21:03
Lesson
12 de 14
E16. The study cited in E15 says that 18% of high

school boys report that they rarely or never
wear a seat belt. Suppose you randomly
sample nine high school boys. Estimate the
probability that none of the boys say that
they rarely or never wear a seat belt.


correct answers, out of ten guesses.

boys, out of nine, who rarely or never
wear a seat belt.
E17. You forgot to study for a ten-question

multiple-choice test on the habits of the
three-toed sloth. You will have to guess
on each question. Each question has four
possible answers. Estimate the probability
that you will answer half or more of the
questions correctly.
E18. A Harris poll estimated that 25% of U.S.

residents believe in astrology. Suppose
you would like to interview a person
who believes in astrology. Estimate the
probability that you will have to ask four or
more U.S. residents to find one who believes
in astrology. [Source: www.sciencedaily.com.]
22-03-2009 21:03
Lesson
13 de 14
Display 5.28 Results of 1990 runs of the number

of babies born to a couple who have
babies until they have a girl.
E20. Boxes of cereal often have small prizes in

them. Suppose each box of one type of cereal
contains one of four different small cars.
Estimate the average number of boxes a
parent will have to buy until his or her child
gets all four cars.

people asked in order to find the first
who believes in astrology.
E19. The probability that a baby is a girl is about

0.49. Suppose a large number of couples each
plan to have babies until they have a girl.
Estimate the average number of babies per
couple.

cereal boxes bought by parents who
buy boxes until their child gets all
four cars.
313
22-03-2009 21:03
Lesson
14 de 14
E23. There are six different car keys in a drawer,

For E21E26, complete ad.
including yours. Suppose you grab one key
at a time until you get your car key. Estimate
b. Model. Make a table that shows how you
the probability that you get your car key on
are assigning the random digits to the
the second try. Start at the beginning of row
outcomes. Explain how you will use the
13 of Table D on page 828.
digits to model the situation and what
E24. A deck of 52 cards contains 13 hearts.
summary statistic you will record.
Suppose you draw cards one at a time,
c. Repetition. Conduct 20 runs of the
without replacement. Estimate the
probability
that it takes you four cards or
Table D on page 828. Make your own
more to draw the first heart. Start at the
frequency table.
beginning of row 28 of Table D on page 828.
E25.
Every so often, a question about probability
gets the whole country stirred up. These
E21. A group of five friends have backpacks that
problems involve a situation that is easy to
all look alike. They toss their backpacks on
understand, but they have a solution that
the ground and later pick up a backpack
seems unreasonable to most people.
at random. Estimate the probability that
The most famous of these problems is the
everyone gets his or her own backpack. Start
so-called Monty Hall problem or the
at the beginning of row 31 of Table D on
problem of the car and the goats. To put
page 828.
it mildly, tempers have flared over this
E22. One method of testing whether a person has
problem. Heres the situation: At the end
extrasensory perception (ESP) involves a set
of the television show Lets Make a Deal, a
of 25 cards, called Zener cards. Five cards
contestant would be offered a choice of three
have a circle printed on them, five a square,
doors. Behind one door was a good prize,
five a star, five a plus sign, and five have wavy
say a car. Behind the other two doors were
lines (see Display 5.30). The experimenter
lesser prizes, say goats. After the contestant
shuffles the cards and concentrates on one
chose a door, the host, Monty Hall, would
at a time. The subject is supposed to identify
open one of the two doors that the contestant
which card the experimenter is looking at.
hadnt chosen. The opened door always had
Once the experimenter looks at a card, it is
a goat. Once, after the door with a goat had
not reused (and the subject is not told what
been opened, the contestant asked if he now
it is). Estimate the probability that a subject
could switch from his ( unopened ) door to
will identify the first five cards correctly just
the other unopened door. Was this a good
by guessing circle, square, star, plus sign,
strategy for the contestant, was it a bad
and wavy lines, in that order. Start at the
strategy, or did it make no difference? Start
beginning of row 41 of Table D on page 828.
page 828.
Display 5.30 Zener cards.
22-03-2009 21:03
Lesson
1 de 10
E26. This question appeared in the Ask Marilyn

column.
I work at a waste-treatment plant, and we
do assessments of the time-to-failure and
time-to-repair of the equipment, then input
those figures into a computer model to make
plans. But when I need to explain the process
to people in other departments, I find it
diffcult. Say a component has two failure
modes. One occurs every 5 years, and the
5.3
In mathematics, A or B
means A or B or both.
other occurs every 10 years. People usually

say that the time-to-failure is 7.5 years, but
this is incorrect. Its between 3 and 4 years.
Do you know of a way to explain this that
people will accept? [Source: PARADE, March 13,
2005, p. 17.]
Use simulation to estimate the expected

time-to-failure for this component. Start
page 828.
The Addition Rule and Disjoint Events

One of the shortest words in the English languageoris often misunderstood.
This is because or can have two different meanings. For example, if you are told
at a party that you can have apple pie or chocolate cake, you might not be sure
whether you must pick only one or whether it would be acceptable to say Both!
In statistics, the meaning of or allows both as a possibility: You can have apple pie
or you can have chocolate cake or you can have both. Thus, in statistics, to find
P(A or B), you must find the probability that A happens or B happens or both A
and B happen.
Disjoint and Complete Categories

Statistical data often are presented in tables like the one in Display 5.31, which
gives basic figures for employment in the United States for all nonmilitary adults
who were employed or seeking employment.
Display 5.31 The U.S. labor force as of July 2006.

[Source: Bureau of Labor Statistics, www.bls.gov.]
The categories are disjoint, which makes computing probabilities easy. For
example, if you pick one of the people in the civilian labor force at random, you
can find the probability that he or she is employed by adding the employees on
farm payrolls to those on nonfarm payrolls and then dividing by the total number
in the civilian labor force:
P (employed) = P(on farm payroll or on nonfarm payroll)
315
22-03-2009 21:03
Lesson
2 de 10
However, in the next example, you will see that you cant always compute the
probability of A or B simply by adding.
Example: Leisure Activities of American Adults

Display 5.32 lists some common leisure activities and the percentage of the
213 million U.S. adults who participated in the activity at least once in the
prior 12 months.
a. Are these categories disjoint for categorizing an adults leisure activities?
Are they complete?
b. How many adults read books for a leisure activity?
c. Can you find the percentage of U.S. adults who surfed the net or went to
the beach?
Display 5.32 Percentage of U.S. adults who engaged in various

leisure activities in 2004. [Source: U.S. Census Bureau,
Statistical Abstract of the United States, 2006, Table 1227.]
Solution
a. These categories arent disjoint because, for example, the same person might
have dined out and read books. Also, for this table, you can tell that the
categories arent disjoint because the percentages sum to more than 100%.
The categories arent complete, because there are many other leisure activities.
b. Thirty-six percent of 213,000,000, or 76,680,000 adults, read books for leisure.
c. From this information, it is impossible to determine the percentage of
U.S. adults who surfed the net or went to the beach. If you add the two
percentages, you are counting the people twice who did both. Because you
dont know how many people that is, you are stuck.
The Addition Rule for Disjoint Events

While doing Activity 5.3a, notice when you can find P( A or B ) simply by adding.
Exploring or
1. For each question, record the number of students in your class who
answer yes.
Is blue your favorite color?
Is red your favorite color?
(continued)
22-03-2009 21:03
Lesson
3 de 10
2. From the information in step 1 alone, can you determine the number of
students in your class who say blue is their favorite color or who say red is
their favorite color? If so, explain how. If not, what additional information
do you need?
3. For each of these questions, record the number of students in your class
who answer yes.
Are you female?
4. From the information in step 3 alone, can you determine the number of
students in your class who are female or who have blue as their favorite
color? If so, explain how. If not, what additional information do you need?
Two events are
disjoint if they cant
happen on the
same opportunity.
In Activity 5.3a, you can get the answer to one question by adding because the
categories are disjoint. For the other question, adding doesnt work because the
categories are not disjoint.
Two useful rules emerge from Activity 5.3a. First, if two events are disjoint,
the probability of their occurring together is 0.
A Property of Disjoint ( Mutually Exclusive ) Events
P( A or B ) is
sometimes written
P( A B ).
If event A and event B are disjoint, then
P( A or B ) is
sometimes written
P( A B ).
Second, if two events are disjoint, then you can add probabilities to find
P(A or B).
Using set notation,

this rule is written
P( A B ) = P( A ) + P( B )
where A B is read
A union B .
The Addition Rule for Disjoint (Mutually Exclusive) Events

If event A and event B are disjoint, then
P(Aand B) = 0
P(A or B) = P(A) + P(B)

The Addition Rule for Disjoint Events can be generalized. For example, if each
pair of events A, B, and C is disjoint, then
P( A or B or C ) = P( A ) + P( B ) + P( C )
Example: The Addition Rule with Dice

Use the Addition Rule for Disjoint Events to find the probability that if you roll
two dice, you get a sum of 7 or a sum of 8.
317
22-03-2009 21:03
Lesson
4 de 10
e. is left-handed; is right-handed
Solution
Let A be the event sum of 7. Let B be the event sum of 8. On the same roll, these
two events are mutually exclusive, so P( A and B ) = 0. Then P( A or B ) =
P( A ) + P( B ). Using the sample space of equally likely outcomes from Display
5.6 on page 295,
P( sum of 7 or sum of 8 ) = P( sum of 7 ) + P( sum of 8 )
Example: The Addition Rule with Tap Water

Use the Addition Rule for Disjoint Events to compute the probability that exactly
one person out of two correctly chooses the tap water in Jack and Jills model.
Solution
P( exactly one T among two tasters )
= P [( 1st chooses T and 2nd chooses B ) or (1st chooses B and 2nd chooses T )]
= P ( TB or BT ) = P( TB ) + P( BT ) =
You can add the probabilities because the two outcomes TB and BT are disjoint.

D13. Suppose you select a person at random from
your school. Are these pairs of events mutually
exclusive?
a. has ridden a roller coaster; has ridden a
Ferris wheel
b. owns a classical music CD; owns a jazz CD
c. is a senior; is a junior
d. has brown hair; has brown eyes
f. has shoulder-length hair; is male
D14. If you select a person at random from your
classroom, what is the probability that the person is a junior or a senior?
D15. Suppose there is a 20% chance of getting a mosquito bite each time you go
outside on a summer evening. Can you use the Addition Rule for Disjoint
Events to compute the probability that you will get bitten if you go outside
three times? If you go outside six times?
22-03-2009 21:03
Lesson
5 de 10
The Addition Rule

What about computing P( A or B ) when you dont have disjoint categories? In that
situation, addressed in Activity 5.3b, you must add an extra step.
More on or
1. For each question, record the number of students in your class who answer
yes.
Are you female?
Is it true that you are female and blue is your favorite color?
2. From the information in step 1 alone, can you determine the number
of students in your class who are female or have blue as their favorite
color? ( Remember, when we use the word or in probability, we mean one
or the other or both. ) If so, explain how to do it. If not, what additional
information do you need?
3. Copy and complete this two-way table showing the number of members of
your class whose responses fall into each cell of the table.
4. Find the number of students who are female or have blue as their favorite
color
a. as a row total plus a column total minus a cell count
b. as the sum of three cell counts
5. Let B be the event that the favorite color of the randomly chosen student is
blue. Let F be the event that the randomly chosen student is female. Write
two formulas for computing P( B or F ), one corresponding to part a and
one to part b in step 4. Use each formula to find P( B or F ).
The ideas you explored in Activity 5.3b can be stated formally as the
Addition Rule.
Using set notation,
P( A B ) = P( A ) +
P( B ) P( A B )
where A B is read
A intersect B.
Addition Rule
For any two events A and B,
P( A or B ) = P( A ) + P( B ) P( A and B )
319
22-03-2009 21:03
Lesson
6 de 10
Example: The Addition Rule for Events That Arent Disjoint

Use the Addition Rule to find the probability that if you roll two dice, you get
doubles or a sum of 8.
Solution
Let A be the event getting doubles, and let B be the event
getting a sum of 8. These two events are not mutually
exclusive, because they have a common outcome (4, 4),
which occurs with probability
From the Addition Rule,
P( A or B ) = P( A ) + P( B ) P( A and B ).
P( doubles or sum of 8 )
= P( doubles ) + P( sum of 8 ) P( doubles and sum of 8 )
You can also apply the Addition Rule backward; that is, you can compute
P( A and B ) when you know the other probabilities.
Example: Computing P( A and B )

In a local school, 80% of the students carry a backpack ( B ) or a wallet ( W ).
Forty percent carry a backpack, and 50% carry a wallet. If a student is selected
at random, find the probability that the student carries both a backpack and
a wallet.
Solution
B and W arent disjoint, so the general form of the Addition Rule applies:
P( B or W ) = P( B ) + P( W ) P( B and W )
0.80 = 0.40 + 0.50 P( B and W )
Solving, P( B and W ) = 0.10. The probability that the student carries both a
backpack and a wallet is 0.10.
The Addition Rule

D16. Thse diagrams are called Venn diagrams:
a. Which diagram illustrates mutually exclusive events?

b. Use these diagrams to justify the two forms of the Addition Rule.
22-03-2009 21:03
Lesson
7 de 10
D17. What happens if you use the general form of the Addition Rule,
P( A or B ) = P( A ) + P( B ) P( A and B )
in a situation where A and B are mutually exclusive?
Summary 5.3: The Addition Rule and Disjoint Events

In statistics, P(A or B) indicates the probability that event A occurs, event B
occurs, or both occur in the same random outcome. The word or always allows
for the possibility that both events occur. In this section, you learned the Addition
Rule for any events A and B:
P( A or B ) = P( A ) + P( B ) P(A and B)
Two events are disjoint ( mutually exclusive ) if they cannot occur on the same
opportunity. In the case of disjoint events, P( A and B ) = 0, so the Addition Rule
simplifies to
P( A or B ) = P( A ) + P( B )
Two-way tables often are helpful in clarifying the structure of a probability
problem and in answering questions about the joint behavior of two events.
Practice
Disjoint and Complete Categories
P14. Of the 34,071,000 people in the United States
who fish, 1,847,000 fish in the Great Lakes,
27,913,000 fish in other fresh water, and
9,051,000 fish in salt water. [Source:U.S. Census
Bureau, Statistical Abstract of the United States, 2006,
Table 1241.]
a. In categorizing people who fish, are

these three categories disjoint? Are they
complete?
b. Suppose you randomly select a person
from among those who fish. Can you find
the probability that the person fishes in
salt water?
c. Suppose you randomly select a person
from among those who fish. Can you find
the probability that the person fishes in
fresh water?
d. The number of people who fish in fresh
water is 28,439,000. How many people
fish in both salt water and fresh water?
P15. Display 5.33 categorizes the child support
received by custodial parents with children
under age 21 in the United States.
Display 5.33 Custodial parents and court-ordered

child support, 2001. [Source: U.S. Census
2006, Table 558.]
a. Revise the table so that the categories

are complete and disjoint. Note that
one category wasnt included: custodial
parents with a child support agreement or
award who were not supposed to receive
payments in 2001 ( but maybe get them in
some other year ).
321
22-03-2009 21:03
Lesson
8 de 10
b. If one of the 13,383,000 custodial

parents is selected at random, what is the
probability that he or she was supposed
to receive payments and received the full
amount or received a partial amount?
P16. If you roll two dice, are these pairs of events
mutually exclusive? Explain.
a. doubles; sum is 8
b. doubles; sum is odd
c. a 3 on one die; sum is 10
d. a 3 on one die; doubles
P17. A researcher will select a student at random
from a school population where 33% of the
students are freshmen, 27% are sophomores,
25% are juniors, and 15% are seniors.
a. Is it appropriate to use the Addition Rule
for Disjoint Events to find the probability
that the student will be a junior or a
senior? Why or why not?
b. Find the probability that the student will
be a freshman or a sophomore.
P18. A tetrahedral die has the numbers 1, 2, 3,
and 4 on its faces. Suppose you roll a pair of
tetrahedral dice.
a. Make a table of all 16 possible outcomes
( or use the one you made in E4 on
page 299 ).
b. Use the Addition Rule for Disjoint Events
to find the probability that you get a sum
of 6 or a sum of 7.
c. Use the Addition Rule for Disjoint
Events to find the probability that you get
doubles or a sum of 7.
d. Why cant you use the Addition Rule for
Disjoint Events to find the probability
that you get doubles or a sum of 6?
The Addition Rule

P19. Display 5.34 gives information about all
reportable crashes on state-maintained roads
in North Carolina in a recent year.
Display 5.34 Information on reportable crashes in

North Carolina. [Source: North Carolina
Crash Data Query, www.hsrc.unc.edu.]
a. Are the events crash involved a teen driver

and crash was speed related mutually
exclusive? How can you tell?
b. Use numbers from the cells of the table to
compute the probability that a randomly
selected crash involved a teen driver or
was speed related.
c. Now use two of the marginal totals and
one number from a cell of the table to
compute the probability that a randomly
selected crash involved a teen driver or
was speed related.
P20. Use the Addition Rule to compute the
probability that if you roll two six-sided dice,
a. you get doubles or a sum of 4
b. you get doubles or a sum of 7
c. you get a 5 on the first die or you get a 5
on the second die
P21. Use the Addition Rule to compute the
probability that if you flip two fair coins, you
get heads on the first coin or you get heads
on the second coin.
P22. Use the Addition Rule to find the probability
that if you roll a pair of dice, you do not get
doubles or you get a sum of 8.
22-03-2009 21:03
Lesson
9 de 10
Exercises
E27. Suppose you flip a fair coin five times. Are
these pairs of events disjoint?
a. You get five heads; you get four heads and
one tails.
b. The first flip is heads; the second flip is
heads.
c. You get five heads; the second flip is tails.
d. You get three heads and two tails; the
second flip is tails.
e. The first four flips are heads; the first
three flips are heads.
f. Heads first occurs on the third flip; the
fourth flip is heads.
g. Heads first occurs on the third flip; three
of the flips are heads.
E28. Suppose you select two people at random
from a nearby health clinic. Are these pairs
of events disjoint?
a. One person has health insurance; the
other person has health insurance.
b. Both people have health insurance; only
one person has health insurance.
c. One person is over age 65; the other
person is under age 19.
d. One person is over age 95; the other
person is the first persons mother.
E29. Display 5.35 shows the U.S. population
categorized by age and sex, as reported by the
U.S. Bureau of the Census. You are working
for a polling organization that is about to
select a random sample of U.S. residents.
a. What is the probability that the first
person selected will be female? Female
and age 85 or older? Female or age 85 or
older?
b. What is the probability that the first
person selected will be male? Male and
under age 30? Male or under age 30?
c. What proportion of females are age 85 or

older? What proportion of people age 85
or older are female?
Display 5.35 Resident population of the United

States by age and sex ( excluding
the population living in institutions,
college dormitories, and other group
quarters ), 2004. [Source: U.S. Census
Bureau, 2004 American Community Survey,
Table B01001.]
E30. Display 5.36 shows the U.S. college

population categorized by age and sex.
Display 5.36 College population of the United

States by age and sex. [Source: U.S. Census
2006, Table 270.]
a. What percentage of college students are

female? Female and age 30 or older?
Female or age 30 or older?
b. What percentage of college students are
male? Male and under age 30? Male or
under age 30?
c. What proportion of female college
students are age 30 or older? What
proportion of college students age 30 or
older are females?
323
22-03-2009 21:03
Lesson
10 de 10
E31. Display 5.37 describes students in a large class.
Display 5.37 Table of laptop ownership and dog

possession.
a. Fill in the missing cells and marginal

totals.
b. What is the probability that a student
randomly selected from this class doesnt
have a dog or doesnt own a laptop?
E32. Display 5.38 classifies crashes in a recent year
in Virginia. Some cells are missing because
they werent given directly by the source.
the probability that the student carries both

a backpack and a wallet. ( Note that the word
only makes this exercise different from the
example on page 320. )
E35. In the United States, about 7,243,000 people
ages 16 to 24 have completed high school
and arent enrolled in college. There are
3,766,000 high school dropouts in this same
age group. Of the high school graduates,
71% are employed, 11% are unemployed,
and 18% are not in the labor force. Of the
high school dropouts, 53% are employed,
14% are unemployed, and 33% are not in the
labor force. If you randomly select a person
age 16 to 24 who isnt enrolled in college or
high school, what is the probability that he
or she is employed? Make a two-way table to
answer this question. [Source: U.S. Census Bureau,
E36. Polls of registered voters often report the

percentage of Democrats and the percentage
of Republicans who approve of the job the
president is doing. Suppose that in a poll
of 1500 randomly selected voters 860 are
Democrats and 640 are Republicans. Overall,
937 approve of the job the president is doing,
and 449 of these are Republicans. Assuming
Display 5.38 Characteristics of crashes in Virginia.
[Source: Virginia Department of Motor Vehicles,
the people in this poll are representative,
www.dmv.state.va.us.]
what percentage of registered voters are
Republicans or approve of the job the
a. Fill in the missing cells and marginal
president is doing? First answer this question
totals.
by using the Addition Rule. Then make a
b. What proportion of crashes involved a
two-way table showing the situation.
fatality and a traffic law violation?
E37. Jill computes the probability that she gets
c. What proportion of crashes involved a
heads at least once in two flips of a fair coin:
fatality or a traffic law violation?
E33. Suppose 80% of people can swim. Suppose
P( at least one heads in two flips )
70% of people can whistle. Suppose 55%
= P( heads on first flip or heads on second )
of people can do both. What percentage of
people can swim or whistle?
= P( heads on first ) + P( heads on second )
a. Construct a table to help you answer this
question.
b. Use the Addition Rule to answer this
question.
She defends her use of the Addition Rule
E34. In a local school, 80% of the students carry a
because getting heads on the first flip and
backpack ( B ) or a wallet ( W ). Forty percent
heads on the second flip are mutually
carry only a backpack, and 30% carry only a
exclusiveboth cant happen at the same
time. What would you say to her?
wallet. If a student is selected at random, find
22-03-2009 21:03
Lesson
1 de 14
E38. Jill computes the probability that she gets

heads exactly once in two flips of a fair
coin:
P(exactly one heads in two flips)
= P((tails on first flip and heads on second)
or (heads on first flip and tails on
second))
= P(tails on first flip and heads on second)
+ P( heads on first flip and tails on
second )
5.4
She defends her use of the Addition Rule

because HT and TH are mutually exclusive.
What would you say to her?
E39. When is this statement true?
P( A or B or C ) = P( A ) + P( B ) + P(C)
E40. Suppose events A, B, and C are three events
where P(A and B) 0, P(A and C) 0,
P( B and C ) 0, and P( A and B and C ) 0.
a. Draw a Venn diagram to illustrate this
situation.
b. Use the Venn diagram from part a to
help you write a rule for computing
P( A or B or C ).
Conditional Probability
The Titanic sank in 1912 without enough lifeboats for the passengers and crew.
Almost 1500 people died, most of them men. Was that because a man was
less likely than a woman to survive, or did more men die simply because men
outnumbered women by more than 3 to 1 on the Titanic? You might turn to an
old campfire song to decide what might have happened. One line is about loading
the lifeboats: And the captain shouted, Women and children first. Statisticians
arent opposed to songs. But statisticians also know that stories and theories, even
stories set to music, are no substitute for data. Do the data in Display 5.39 support
the lyrics of the song?
Display 5.39 Titanic survival data. [Source: Journal of Statistics Education 3,

no. 3 (1995).]
Although the numbers alone cant tell you who got to go first on the lifeboats,
they do show that
or roughly 73%, of the females survived, while only
or roughly 21%, of the males survived. Thus, the data are fully consistent with the
hypothesis that the song explains what happened.
Overall,
or approximately 32.3%, of the people survived, but the survival
rate for females was much higher and that for males much lower. The chance of
surviving depended on the condition of whether the person was male or female.
This commonsense notion that probability can change if you are given additional
information is called conditional probability.
325
22-03-2009 21:05
Lesson
2 de 14
Conditional Probability from the Sample Space

Often you can calculate conditional probabilities directly from the sample space.
The symbol for the

conditional probability
that A happens given
that B happens is P( A | B ).
Example: The Titanic and Conditional Probability

Suppose you pick a person at random from the list of people aboard the Titanic.
Let S be the event that this person survived, and let F be the event that the person
was female. Find P(S | F), read the probability of S given that F is known to have
happened.
Solution
The conditional probability P(S | F ) is the probability that the person survived
given the condition that the person was female. To find P( S | F ), restrict your
sample space to only the 470 females (the outcomes for whom the condition
is true ). Then compute the probability of survival as the number of favorable
outcomes344, the women survivingdivided by the total number of
outcomes470in the restricted sample space.
P( S | F ) = P( survived | female ) =
0.732
Overall, only 32.3% of the people survived, but 73.2% of the females survived.
Example: Sampling Without Replacement

When you sample without replacement from a small population, the probabilities
for the second draw depend on what happens on the first draw. For example,
imagine a population of four recent graduates ( that is, N = 4 ), two from the west
( W ) and two from the east ( E ). Suppose you randomly choose one person from
the population and he or she is from the west. What is the probability that the
second person selected is also from the west? In symbols, find
P( W selected 2nd | W selected 1st )
Solution
If you start with a population {W, W, E, E} and the first selection is W, that leaves
the restricted sample space { W, E, E } from which to choose on the second draw.
The conditional probability that the second selection is from the west is
P(W selected 2nd | W selected 1st ) =
Conditional Probability from the Sample Space

D18. When you compare sampling with and without replacement, how does
the size of the population affect the comparison? Conditional probability
lets you answer the question quantitatively. Imagine two populations of
students, one large ( N = 100 ) and one small ( N = 4 ), with half the students
22-03-2009 21:05
Lesson
3 de 14
in each population male ( M ). Draw random samples of size n = 2 from each

population.
a. First, consider the small population. Find P( 2nd is M | 1st is M ), assuming
you sample without replacement. Then calculate the probability again,
this time with replacement.
b. Do the same for the larger population.
c. Based on these calculations, how would you describe the effect of
population size on the difference between the two sampling methods?
The Multiplication Rule for P(A and B)

You will now investigate a general rule for finding P( A and B ) for any two events
A and B. Suppose, for example, you choose one person at random from the
passenger list of the Titanic. What is the probability that the person was female
( F ) and survived ( S )? You already know how to answer this question directly by
using the data in Display 5.39. Out of 2201 passengers, 344 were women and
survived, so
P( F and S ) =
You can also find the answer using conditional probability. It sometimes helps
to show the order of events in a tree diagram, such as the one in Display 5.40.
Display 5.40 Tree diagram for the Titanic data.
You can interpret the top branch of the tree diagram, for example, as
P( F ) P( S | F ) =
= P( F and S )
The tree diagram leads to a rule for finding P( A and B ), shown in the box on
the next page.
327
22-03-2009 21:05
Lesson
4 de 14
P( A and B ) is
sometimes written
as P( A B ) and
sometimes as P( AB ).
The Multiplication Rule

The probability that event A and event B both happen is given by
P( A and B ) = P( A ) P( B | A )
Using set notation,

P( A B ) = P( A ) ( B | A )
or P( A B ) =
P( B ) ( A | B ).
Alternatively,
P( A and B ) = P( B ) P( A | B )
Display 5.41 shows the Multiplication Rule on the branches of a tree
diagram.
From past work with fractions, you know that of translates to multiplication.
Three-fourths of 20 means 20, or 15. The Multiplication Rule is simply an
extension of this same logic.
Display 5.41 The general Multiplication Rule, shown on a tree

diagram.
Example: Females on the Titanic

Use the Multiplication Rule to compute the percentage of passengers on the
Titanic who were females and survived.
Solution
Twenty-one percent of all Titanic passengers were female, and 73% of female
passengers survived. So 73% of 21%, or 15.33%, of all passengers were female and
survived.
P( female and survived ) = ( 0.21 )( 0.73 ) = P( female ) P( survived | female )
Example: Coincidences
This article appeared in the Los Angeles Times on July 18, 1978.
Man, Wife Beat Odds in Moose-Hunt Draw
The Washington State Game Department conducted a public drawing
last week in Olympia for three moose hunting permits. There were
2,898 application cards in the wire mesh barrel. It was cranked around
22-03-2009 21:05
Lesson
5 de 14
several times before the first name was drawn: Judy Schneider of East
Wenatchee, Wash. The barrel was spun again. Second name drawn: Bill
Schneider, Judys husband. Result: groans of disbelief from hopeful moose
hunters in the auditorium.
a. What is the probability that Judys name would be the first drawn and her
husbands name would be second?
b. What assumptions are you making when you answer this question?
c. Comment on why coincidences like this seem to happen so often.
Solution
a. P( Judy first and Bill second ) = P( Judy first ) P( Bill second | Judy first )
0.000000119
b. The assumptions are that Bill and Judy had only one card each in the barrel
and that the cards were well mixed.
c. The chance that Judys name and then her husbands name will be called is
about one chance in 10 million. However, there is also the chance that Bills
name will be called first and then Judys. This doubles their chances of being
the first two names drawn to 0.000000238only about one chance in
5 million. However, suppose all the names in the barrel were those of
couples. Then the probability that the first two names drawn will be those
of a couple is
or about 0.000345, because the first name can be anyone
and that persons partner is one of the 2897 left for the second draw. This
probability is still small, but now it is 3 chances in 10,000. Finally, thousands
of lotteries take place in the United States every year, so it is virtually certain
that coincidences like this will happen occasionally and be reported in the
newspaper.

D19. Use the tree diagram for the Titanic data in Display 5.40 on page 327 to find
P( male and survived ) and P( male and died ).
D20. Use the Titanic data in Display 5.39 on page 325 to show that
P( M and S ) = P( S ) P( M | S ) = P( M ) P( S | M )
D21. Use the frequencies in the table below to show that
P( A and B ) = P( A) P( B | A ) = P( B ) P( A | B )
is true in general.
329
22-03-2009 21:05
Lesson
6 de 14
The Definition of Conditional Probability

Its time to define conditional probability formally.
Using set notation,
this definition
is written
For any two events A and B such that P( B ) > 0,
The Multiplication Rule is a consequence of the formal definition of

conditional probability. Simply solve the equation in the definition for
P( A and B ).
Example: Conditional Probability

Suppose you roll two dice. Use the definition of conditional probability to find the
probability that you get a sum of 8 given that you rolled doubles.
Solution
Thinking in terms of the restricted sample space, you already know that the
answer is because there are six equally likely ways to roll doubles and only
one of these is a sum of 8. Using the definition of conditional probability, the
computation looks like this.
Conditional Probability and Medical Tests

In medicine, screening tests give a quick indication of whether a person is likely to
have a particular disease. (For example, the ELISA test is a screening test for HIV,
and a chest X ray is a screening test for lung cancer.) Because screening tests are
intended to be relatively quick and noninvasive, they often are not as accurate as
other tests that take longer or are more invasive. (For example, a biopsy is more
accurate than a chest X ray if you want to know whether a lung tumor is cancerous.)
A two-way table like the one in Display 5.42 is often used to show the four
possible outcomes of a screening test.
Display 5.42 Possible results of a screening test.
22-03-2009 21:05
Lesson
7 de 14
The effectiveness of screening tests is judged using conditional probabilities.

These four terms are commonly used. In each case, higher values are better.
Positive predictive value ( PPV ) = P( disease | test positive ) =
Negative predictive value (NPV ) = P( no disease present | test negative ) =
Sensitivity = P( test positive | disease present ) =
Specificity = P( test negative | no disease ) =
Example: A Relatively Rare Disease

Relatively rare diseases have tables similar to Display 5.43. This test has 50 false
positives; that is, 50 people who dont have the disease tested positive. It has
only 1 false negativea person with the disease who tested negative. Find the
sensitivity, specificity, PPV, and NPV for this test.
Display 5.43 Hypothetical results of a screening test for a rare disease.
Solution
In some ways, this is a pretty good test. It finds 9 out of the 10 people who have
the disease, for a sensitivity of 0.9. It correctly categorizes 9940 out of the 9990
people who dont have the disease, for a specificity of 0.99. The NPV is 9940 out
of 9941, or 0.9999. However, notice that only 9 out of 59, or 15%, of the people
who test positive for the disease actually have it! The PPV is quite low because
there are so many false positives.
The previous example shows what can happen when the population being
screened is mostly disease free, even with a test of high specificity. Most of the
people who test positive do not, in fact, have the disease. On the other hand, if the
population being screened has a high incidence of the disease, then there tend to
be many false negatives and the negative predictive value tends to be low.
Because the positive predictive value and the negative predictive value depend
on the population as well as the screening test, statisticians prefer to judge a test
based on the other pair of conditional probabilities: sensitivity and specificity.

D22. Most patients who take a screening test find the positive predictive value
and the negative predictive value easier to interpret than sensitivity and
specificity. Explain why.
331
22-03-2009 21:05
Lesson
8 de 14
Conditional Probability and Statistical Inference

To calculate a probability, you must work from a model. For example, suppose
you want to calculate the probability of observing an even number of dots on a
roll of a die. Your model is that the die is fair. Using this model, you know that
P( even number | fair die ) =
Now suppose you have a die that you suspect isnt fair. If you roll it and get
an even number, you cannot calculate P( fair die | even number ). So how can you
discredit a model?
If you roll the die 20 times and you get an even number every time, you can
compute (by the method of the next section)
P( even number on all 20 rolls | fair die ) =
0.000001
This outcome is so unlikely that you can feel justified in abandoning the model
that the die is fair. But you still didntand cantcompute
P( fair die | even number on all 20 rolls )
The dialogue that follows is invented and did not actually occur in the
Westvaco case (Chapter 1), but it is based on real conversations one of your
authors had on several occasions with a number of different lawyers as they
grappled with conditional probabilities.
Statistician: Suppose you draw three workers at random from the set of ten
hourly workers. This establishes random sampling as the model
for the study.
Lawyer: Okay.
, or 120, possible samples of size 3,
Statistician: It turns out that there are
and only 6 of them give an average age of 58 or more.
Lawyer: So the probability is
, or 0.05.
Statistician: Right.
Lawyer: Theres only a 5% chance that the company didnt discriminate and
a 95% chance that it did.
Statistician: No, thats not true.
Lawyer: But you said . . .
Statistician: I said that if the age-neutral model of random draws is correct,
then theres only a 5% chance of getting an average age of 58 or
more.
Lawyer: So the chance that the company is guilty must be 95%.
22-03-2009 21:05
Lesson
9 de 14
Statistician: Slow down. If you start by assuming the model is true, you can
compute the chances of various results. But youre trying to start
from the results and compute the chance that the model is right
or wrong. You cant do that.
Here is the analysis: The lawyer is having trouble with conditional
probabilities. The statistician has computed
P(average age
58 | random draws) = 0.05
The lawyer wants to know

P(no discrimination | average age = 58)
The statistician has computed P(data | model). The lawyer wants to know
P(data | model). Finding the probability that there was no discrimination given
that the average age was 58 is not a problem that statistics can solve. A model
is needed in order to compute a probability.
Conditional Probability and Statistical Inference

D23. Theres a parallel between statistical testing and medical testing. Write
expressions for conditional probabilitiesin terms of finding a company
guilty or not guilty of age discriminationthat correspond to a PPV and an
NPV. (You might think of the true medical status as comparable to the true
state of affairs within the company.)
Summary 5.4: Conditional Probability

In this section, you saw these important uses of conditional probability in statistics:
to compare sampling with and without replacement
to study the effectiveness of medical screening tests
to describe probabilities of the sort used in statistical tests of hypotheses
You have learned the definition of conditional probability and the
Multiplication Rule and how to use these two key concepts in solving practical
problems. When you are solving conditional probability problems, by using the
Multiplication Rule or otherwise, tree diagrams and two-way tables help you
organize the data and see the structure of the problem.
The definition of conditional probability is that for any two events A and B,
where P(B) > 0, the probability of A given the condition B is
The Multiplication Rule is
5.4 Conditional Probability 333
22-03-2009 21:05
Lesson
10 de 14
Practice
Conditional Probability from the
Sample Space
P23. For the Titanic data in Display 5.39 on
page 325, let S be the event a person survived
and F be the event a person was female. Find
and interpret these probabilities.
b. P(F | S)
a. P(F)
c. P(not F)
d. P(not F | S) e. P(S | not F)
P24. Display 5.44 gives the hourly workers in
the United States, classified by race and
by whether they were paid at or below
minimum wage or above minimum wage.
You select an hourly worker at random.
c. P(3rd draw is blue | 1st draw is red and

2nd draw is blue)
d. P(3rd draw is red | 1st draw is red and
2nd draw is red)
P26. Suppose Jill draws a card from a standard
52-card deck. Find the probability that
a. it is a club, given that it is black
b. it is a jack, given that it is a heart
c. it is a heart, given that it is a jack

P27. Look again at the Titanic data in Display 5.39
on page 325.
a. Make a tree diagram to illustrate this
situation, this time branching first on
whether the person survived.
b. Write these probabilities as unreduced
fractions.
i. P(survived)
ii. P( female | survived)
Display 5.44 Hourly wages of workers (in
iii. P(survived and female)
thousands), by race. [Source: U.S. Bureau
of Labor Statistics, Employment and Earnings,
c. Now write a formula that tells how the
January 2005; U.S. Census Bureau, Statistical
three probabilities in part b are related.
Abstract of the United States, 2006, Table 636.]
Compare it to the computation on
page 328.
a. Find P(worker is paid at or below
minimum wage).
d. Write two formulas involving conditional
probability to compute P(male and
b. Find P(worker is paid at or below
survived).
minimum wage | worker is white).
P28. Use the Multiplication Rule to find the
c. What does a comparison of the two
probability that if you draw two cards from
probabilities in parts a and b tell you?
a deck without replacing the first before
d. Find P(worker is black).
drawing the second, both cards will be
e. Find P(worker is black | worker is paid at
hearts. What is the probability if you replace
or below minimum wage).
the first card before drawing the second?
f. What does a comparison of the two
P29.
Suppose you take a random sample of
probabilities in parts d and e tell you?
size n = 2, without replacement, from
P25. Suppose Jack draws marbles at random,
the population {W, W, M, M}. Find these
without replacement, from a bag containing
probabilities: P(W chosen 1st) and
three red and two blue marbles. Find these
P(W chosen 2nd | W chosen 1st). Now
conditional probabilities.
find P(W chosen 1st and W chosen 2nd).
a. P(2nd draw is red | 1st draw is red)
P30. Use the Multiplication Rule to find the
probability of getting a sum of 8 and doubles
b. P(2nd draw is red | 1st draw is blue)
when you roll two dice.
22-03-2009 21:05
Lesson
11 de 14

P31. Suppose you roll two dice. Use the
definition of conditional probability
to find P(doubles | sum 8). Compare this
probability with P(sum 8 | doubles).
P32. Suppose you know that, in a class of 30
students, 10 students have blue eyes and 20
have brown eyes. Twenty-four of the students
are right-handed, and 6 are left-handed. Of
the left-handers, 2 have blue eyes. Make and
fill in a table showing this situation. Then use
the definition of conditional probability to
find the probability that a student randomly
selected from this class is right-handed,
given that the student has brown eyes.
P33. As of July 1 of a recent season, the Los
Angeles Dodgers had won 53% of their
games. Eighteen percent of their games had
been played against left-handed starting
pitchers. The Dodgers won 36% of the games
played against left-handed starting pitchers.
What percentage of their games against
right-handed starting pitchers did they win?
same degree of contamination. Suppose the

technician makes the correct decision 90% of
the time.
a. Make a two-way table showing what you
would expect to happen.
b. Compute and interpret her PPV.
c. Compute and interpret her NPV.
d. How would these rates change if she
were given 100 samples with
50 contaminated?
P35. Display 5.45 presents hypothetical data
on 130 people taking an inkblot test for
bureaucratic pomposity disorder (BPD). Use
the data to compute the four conditional
probabilities defined for screening tests on
page 331. Is the test a good one, in your
judgment? Why or why not?
[Source: Los Angeles Times, July 1, 2000, page D8.]

Display 5.45 Results of a test for bureaucratic
pomposity disorder.
Conditional Probability and Statistical

Inference
P36. Rosa has two pennies. One is an ordinary
fair penny, but the other came from a magic
shop and has two heads. Rosa chooses one
of these coins, but she does not choose it
randomly. She also doesnt tell you how she
chose it, so you dont know the probabilities
P(coin is two-headed) or P(coin is fair). She
flips the coin once. For each probability, give
a numerical value if it is possible to find one.
If it is not possible to compute a probability,
explain why.
P34. A laboratory technician is being tested on
her ability to detect contaminated blood
samples. Among 100 samples given to her,
20 are contaminated, each with about the
a.
b.
c.
d.
e.
f.
P(coin lands heads | coin is fair)

P(coin lands heads | coin is two-headed)
P(coin is fair | coin lands heads)
P(coin is two-headed | coin lands heads)
P(coin is fair | coin lands tails)
P(coin is two-headed | coin lands tails)
22-03-2009 21:05
Lesson
12 de 14
Exercises
E41. Display 5.46 (at the bottom of this page)
gives a breakdown of the U.S. population age
16 and older by age and whether the person
volunteers time to charitable activities.
You are working for a polling organization
that is about to select a random sample of
U.S. residents age 16 and older. Find these
probabilities for the first person selected in
the random sample.
a. P(person is a volunteer)
b. P(person is a volunteer | person is
age 16 to 24)
c. P(person is age 16 to 24 | person is a
volunteer)
d. P(person is age 16 to 34 | person is not
a volunteer)
e. Which age group has the highest
percentage of volunteers?
E42. Joseph Lister (18271912), British surgeon at
the Glasgow Royal Infirmary, was one of the
first to believe in Louis Pasteurs germ theory
of infection. He experimented with using
carbolic acid to disinfect operating rooms
during amputations. When carbolic acid was
used, 34 of 40 patients lived. When carbolic
acid was not used, 19 of 35 patients lived.
Suppose a patient is selected at random.
Make a two-way table and find
a. P(patient died | carbolic acid used)
b. P(carbolic acid used | patient died)
c. P(carbolic acid used and patient died)
d. P(carbolic acid used or patient died)
E43. A class collects information on students

gender and handedness. The table gives the
proportion of class members in each category.
a. If possible, fill in the rest of this table.
b. If a student in this class is chosen at

random, what is the probability that the
student is left-handed, given that the
student is male?
c. What is the probability that the
randomly selected student is female
and left-handed?
E44. Suppose the only information you have
about the Titanic disaster appears in this
two-way table. Is it possible to fill in the
rest of the table? Is it possible to find the
probability that a randomly selected person
aboard the Titanic was female and survived?
Either calculate the numbers, or explain why
it is not possible to do so without additional
information and tell what information you
would need.
Display 5.46 Persons (in thousands) who performed unpaid volunteer activities in the last
year. [Source: U.S. Census Bureau, Statistical Abstract of the United States, 2006, Table 575.]
22-03-2009 21:05
Lesson
13 de 14
E45. As you have seen, a useful way to work

with the Multiplication Rule is to record the
probabilities along the branches of a tree
diagram. To practice this, suppose you draw
two marbles at random, without replacement,
from a bucket containing three red (R)
and two blue (B) marbles. Compute the
probabilities in parts af and then use these
probabilities to fill in and label the branches
in a copy of the tree diagram in Display 5.47.
a. P(R on 1st draw)
b. P(B on 1st draw)
c. P(R on 2nd draw | R on 1st draw)
d. P(B on 2nd draw | R on 1st draw)
e. P(R on 2nd draw | B on 1st draw)
f. P(B on 2nd draw | B on 1st draw)
E47. Suppose you draw two cards from a 52-card

deck.
a. What is the probability that both cards
are aces if you replace the first card before
drawing the second?
b. What is the probability that both cards
are aces if you dont replace the first card
before drawing the second?
c. What is the probability that the first card
is an ace and the second is a king if you
replace the first card before drawing the
second?
d. What is the probability that the first card
is an ace and the second is a king if you
dont replace the first card before drawing
the second?
e. What is the probability that both cards
are the same suit if you replace the first
card before drawing the second?
f. What is the probability that both cards
are the same suit if you dont replace the
first card before drawing the second?
E48. A Harris poll found that 36% of all adults
said they would be interested in going to
Mars. Of those who were interested in going,
62% were under age 25. Write a question
for which the solution would be to compute
(0.36)(0.62). [Source: Harris, 1997.]
E49. Suppose that if rain is predicted, there is a
60% chance that it will actually rain. If rain
Display 5.47 Tree diagram for drawing marbles.
is not predicted, there is a 20% chance that it
will rain. Rain is predicted on 10% of days.
E46. Your sock drawer hasnt been organized in a
while. It contains six identical brown socks,
a. On what percentage of days does it rain?
three identical white socks, and five identical
b. Describe a simulation that you could run
black socks. You draw two socks at random
to check your answer.
(without replacement).
E50. In 1999, Elizabeth Dole was a candidate to
a. Make a tree diagram similar to that in
become the first woman president in U.S.
E45 to show the possible results of your
history, and many observers assumed that
draws.
she would show particular strength among
b. Find P(first sock is white).
women voters. According to a Gallup poll,
She did slightly better among Republican
c. Find P(second sock is white | first sock
women than among Republican men,
is white).
but this strength was not nearly enough
d. Find P(second sock is white | first sock
to enable her to challenge Bush. In the
is brown).
October poll, Dole received the vote of 16%
e. What is the probability that you get two
of Republican women, compared to 7%
socks that match?
22-03-2009 21:05
Lesson
14 de 14
of Republican men. Given the additional

information that the Republican party is
about 60% male, find the probability that
a Republican randomly selected from the
October survey would have voted for Dole.
[Source: www.gallup.com, 2000.]
E51. A laboratory screening test for the detection

of a certain disease gives a positive result
6% of the time for people who do not have
the disease. The test gives a negative result
0.5% of the time for people who do have
the disease. Large-scale studies have shown
that the disease occurs in about 3% of the
population.
a. Fill in this two-way table, showing the
results expected for every 100,000 people.
b. What is the probability that a person

selected at random tests positive for
this disease?
c. What is the probability that a person
selected at random who tests positive for
the disease does not have the disease?
E52. A screening test for bureaucratic pomposity
disorder (BPD) has reasonably good
sensitivity and specificity, say 90% for both.
(This means that, on average, nine out of
every ten people who have BPD test positive
and nine out of every ten who dont have
BPD test negative.) Consider testing two
different populations.
a. Officiousville has a population of 1000.
Half the people in Officiousville have
BPD. Suppose everyone in Officiousville
gets tested. Fill out a table like Display
5.42 on page 330. Then use it to compute
the PPV and NPV for this test.
b. Mellowville also has a population of 1000,
but only 10 people in the whole town
have BPD. Fill out a table to show how
the test would perform if you used it to
screen Mellowville. Then compute the

PPV and NPV for this test.
E53. A new test for HIV, the virus that causes
AIDS, uses a mouth swab instead of a blood
test. The president of the company that
produces this test was quoted as saying that
overall the test was more than 99% accurate.
The president claimed that, in the last year,
the company received complaints of only
107 false positives out of 28,436 tests. [Source:
New York Times, December 10, 2005.]
a. Can you determine the positive predictive

value (PPV) for this test? Explain.
b. The article gave no information about
false negatives. Why do you think no
information was given?
c. How many false negatives would there
have to be in order for the test to produce
correct results in 99% of the 28,436 tests?
E54. Make a table and do a computation to
illustrate that if the population being
screened has a high percentage of people
who have a certain disease, the NPV will
tend to be low.
E55. Return to the situation in P36 in which
Rosa has two pennies. One is an ordinary
fair penny, but the other came from a magic
shop and has two heads. Rosa chooses one
of these coins, this time at random. She flips
the coin once. For each probability, give a
numerical value if it is possible to find one.
If it is not possible, explain why not.
a. P(coin lands heads | coin is fair)
b. P(coin lands heads | coin is two-headed)
c. P(coin is fair | coin lands heads)
d. P(coin is two-headed | coin lands heads)
e. P(coin is fair | coin lands tails)
f. P(coin is two-headed | coin lands tails)
E56. One of these two statements is true, and
one is false. Which is which? Make up an
example using hypothetical data to illustrate
your decision.
P(A) = P(A | B) + P(A | not B)
P(A) = P(A and B) + P(A and not B)
22-03-2009 21:05
Lesson
1 de 10
5.5
Independent Events
Jack and Jill have finished conducting taste tests with 100 adults from their
neighborhood. They found that 60 of them correctly identified the tap water, but
how well a person did depended on whether that person regularly uses bottled
water for drinking. Of people who regularly drink bottled water,
or 80%,
correctly identified the tap water. Of people who drink tap water, only
or
51.4%, correctly identified the tap water. (See Display 5.48.)
Display 5.48 Results of taste test depend on whether a person

regularly drinks bottled water.
On the other hand, as Display 5.49 shows, men and women did equally well
in identifying the tap water. In each case, 60% correctly identified the tap water.
Display 5.49 Results of taste test dont depend on gender.
Suppose one person is selected at random from the 100 who completed
the taste test. Based on these data, the events drinks bottled water and correctly
identifies tap water are dependent events. If the person drinks bottled water,
he or she is more likely to correctly identify tap water. The events is a male and
correctly identifies tap water are independent events. Being male doesnt change
the probability that the person correctly identifies tap water.
More formally, events A and B are independent if the probability of event A
doesnt depend on whether event B happens.that is, knowing that B happens
doesnt change the probability that A happens.
Definition of Independent Events
Suppose P(A) > 0 and P(B) > 0. Then events A and B are independent events
if and only if P(A | B) = P(A) or, equivalently, P(B | A) = P(B).
5.5 Independent Events 339
22-03-2009 21:05
Lesson
2 de 10
In E73, you will show that P(A | B) = P(A) if and only if P(B | A) = P(B). In
E74, you will show that if events A and B are independent, then so are events A
and not B (event not B is the complement of event B, consisting of all possible
outcomes that are not in B).
Example: Water, Gender, and Independence

Show that the events is a male and correctly identifies tap water are independent
and that the events drinks bottled water and correctly identifies tap water are not
independent.
Solution
Suppose a person is selected at random from the 100 completing the water taste
test. First, consider the events is a male and correctly identifies tap water.
P(correctly identifies tap water) =
P(correctly identifies tap water | is a male) =
= 0.60
Because these two probabilities are equal, the events are independent.
Next, consider the events drinks bottled water and correctly identifies tap water.
P(correctly identifies tap water) =
= 0.60
P(correctly identifies tap water | drinks botted water) =
= 0.80
Because these two probabilities are not equal, the events are not independent.
Independent Events
D24. Suppose a person is selected at random from the 100 completing the water
taste test. Show in another way that the events is a male and correctly
identifies tap water are independent and that the events drinks bottled water
and correctly identifies tap water are not independent. Specifically, use
correctly identifies tap water as the condition in both cases.
D25. Suppose you choose a student at random from your school. In each case,
does knowing that event A happened increase the probability of event
B, decrease the probability of event B, or leave the probability of event B
unchanged?
a. A: The student is a football player; B: The student weighs less than 120 lb.
b. A: The student has long fingernails; B: The student is female.
c. A: The student is a freshman; B: The student is male.
d. A: The student is a freshman; B: The student is a senior.
22-03-2009 21:05
Lesson
3 de 10
Multiplication Rule for Independent Events

In Section 5.4, you learned how to use the Multiplication Rule, P(A and B) =
P(A) P(B | A). If A and B are independent, then P(B | A) = P(B), so the
Multiplication Rule reduces to a simple product.
Two events A and B are independent if and only if
P(A and B) = P(A) P(B)
More generally, events A1, A2, . . . , An are independent if and only if
P(A1 and A2 and . . . and An) = P(A1) P(A2.) . . P(An)
You can also use this rule to decide whether two events are independent.
If P(A and B) = P(A) P(B), then A and B are independent.
Example: Four Flips

If you flip a fair coin four times, what is the probability that all four flips result
in heads?
Solution
The outcomes of the first flips dont change the probabilities on the remaining
flips, so the flips are independent. For a sequence of four flips of a fair coin,
P(heads on 1st flip and heads on 2nd and heads on 3rd and heads on 4th)
= P(heads on 1st) P(heads on 2nd) P(heads on 3rd) P(heads on 4th)
Example: Multiplication Rule and Health Insurance

According to Health Highlights, published by the U.S. Department of Health and
Human Services, about 30% of young American adults ages 19 to 29 dont have
health insurance. [Source: www.healthfinder.gov, May 4, 2005.] What is the chance that if you
choose two American adults from this age group at random, the first has health
insurance and the second doesnt?
Solution
Because the sample is selected from the large number of young adults in the
United States, you can model the outcomes of the two trials as independent
events. Then
P(1st has insurance and 2nd doesnt have insurance) = (0.3) (0.7) = 0.21
22-03-2009 21:05
Lesson
4 de 10
You can conveniently list the probabilities of all four possible outcomes for two
young adults in a multiplication table, as in Display 5.50.
Display 5.50 Two-way table for the health insurance example.
Example: Computing the Probability of At Least One

About 30% of young American adults ages 19 to 29 dont have health insurance.
Suppose you take a random sample of ten American adults in this age group.
What is the probability that at least one of them doesnt have health insurance?
Solution
This question is asking for the probability that the first young adult doesnt have
health insurance or the second young adult doesnt have health insurance or . . . or
the tenth young adult doesnt have health insurance. You cannot use the Addition
Rule for Mutually Exclusive Events and add 0.3 + 0.3 + + 0.3, because the
events arent mutually exclusive (and note that youll get a probability larger than
1). However, you can easily compute the complement, the probability that all the
young adults have health insurance:
P(at least one doesnt have insurance)
= 1 P(all have insurance)
= 1 P(1st has insurance and 2nd has insurance . . . and 10th has insurance)
...
= 1 P(1st has insurance) P(2nd has insurance)
P(10th has insurance)
= 1 (0.7)(0.7) (0.7)
= 1 (0.7) 10
1 0.028
= 0.972
You can use the Multiplication Rule because the very large number of young
adults in the United States means that the probability of getting someone with
insurance does not change (by any noticeable amount) depending on who has
already been chosen.
In practice, computing P(A and B) for independent events is straightforward:
Use the definition, which tells you to multiply. Whats not always straightforward
in practice is recognizing whether independence gives a good model for a
real situation. The basic idea is to ask whether one event has a bearing on the
likelihood of another. This is something you try to decide by thinking about
22-03-2009 21:05
Lesson
5 de 10
the events in question. As always, models are simplifications of the real thing,
and youll often use independence as a simplifying assumption to enable you to
calculate probabilities, at least approximately.
Multiplication Rule
D26. Are these situations possible? Explain your answers.
a. Events A and B are disjoint and independent.
b. Events A and B are not disjoint and are independent.
c. Events A and B are disjoint and dependent.
d. Events A and B are not disjoint and are dependent.
D27. Suppose that you randomly and independently select three houses from
those sold recently in the Chicago area. Find P(all three are below the median
price). Find P(at least one is below the median price).
Independence with Real Data

You have seen that data often serve as a basis for establishing a probability model
or for checking whether an assumed model is reasonable. In using data to check
for independence, however, you have to be careful. For example, it would be fairly
unusual for Jack and Jill to come up with exactly the same percentage of men as
women who can identify tap water even if men and women are equally likely to be
able to identify tap water. That example was cooked up; its almost impossible to
find such perfection in practice.
Example: Independence and Baseball

In a recent season, by July 1 the Los Angeles Dodgers had won 41 games and lost
37 games. The breakdown by whether the game was played during the day or at
night is shown in Display 5.51. If one of these games is chosen at random, are the
events win and day game independent?
Display 5.51 Dodger record of wins and losses by time of game.

[Source: Los Angeles Times, July 1, 2000, page D8.]
Solution
First, check the probabilities:
22-03-2009 21:05
Lesson
6 de 10
Because these two probabilities arent exactly equal, you must conclude that the
events win and day game are not independent. Yet, given that the Dodgers played
21 day games, the percentage of day games won couldnt be any closer to 0.526.
A statistician would conclude that if these results can be considered a random
sample of their games, there isnt sufficient evidence to say that the two events
win and day game are not independent. This is the subject of Chapter 10.
Activity 5.5a further illustrates the difficulty of establishing independence
with real data.
Independence with Real Data
What youll need: one penny per student
1. Collect and study data on eyedness and
handedness.
a. Are you right-handed or left-handed?
b. Determine whether you are right-eyed or
left-eyed: Hold your hands together in front
of you at arms length. Make a space between
your hands that you can see through.
Through the space, look at an object at
least 15 ft away. Now close your right eye.
Can you still see the object? If so, you are
left-eyed. Now close your lefteye. Can you still see the object? If so, you
are right-eyed.
c. Would you expect being right-handed and right-eyed to be independent?
d. Complete a two-way table for members of your class, showing the
frequencies of eyedness and handedness.
e. For a randomly selected student in your class, are being right-handed
and right-eyed independent?
2. Collect and study data on the results of two coin flips.
a. With other members of your class, flip a coin twice until you have
100 pairs of flips. Place your results in a two-way table. (Save this data
for use again in Chapter 8.)
b. Would you expect the results of the flip and the second flip to be
independent?
c. From your data for two coin flips, do the results of the first flip and the
second flip appear to be independent?
3. Did you get the results you expected in steps 1 and 2? Explain.
If a person is selected at random from the general population, being

right-handed and being right-eyed arent independent, which is probably what
you found for your class in Activity 5.5a. On the other hand, two flips of a coin
are independent, even though for real dataactual flipsP(A) almost never
exactly equals P(A | B), as the definition of independence requires.
22-03-2009 21:05
Lesson
7 de 10
In Chapter 10, you will learn how to examine tables like the ones for your
coin flips in Activity 5.5a and for the Dodger record and then perform a statistical
test of independence. The key question, just as in Chapter 1, will be whether the
data are too extreme (too far from the strict definition of independence) to be
consistent with that hypothesis.
Summary 5.5: Independent Events

The word independent is used in several related ways in statistics. First, two
events are independent if knowledge that one event occurs does not affect
the probability that the other event occurs. Second, in sampling units from a
population with replacement, the outcome of the second selection is independent
of what happened on the first selection (and the same is true for any other pair
of selections). Sampling without replacement leads to dependent outcomes on
successive selections. (However, if the population is large compared to the sample,
you can compute probabilities as if the selections were independent without much
error.) In both situations, the same definition of independence applies: Two events
A and B are independent if and only if P(A) = P(A | B); the probability of A does
not change with knowledge that B has happened. If P(A) = P(A | B), then it is also
true that P(B) = P(B | A).
You learned in this section that the Multiplication Rule simplifies to
if and only if A and B are independent events.
Practice
Independent Events
P37. Suppose you select one person at random
from the Titanic passengers and crew in
Display 5.39 on page 325. Use the definition
of independent events to determine whether
the events didnt survive and male are
independent. Are any two events in this table
independent?
P38. Suppose you draw a card at random from
a standard deck. Use the definition of
independent events to determine which pairs
of events are independent.
a. getting a heart; getting a jack
b. getting a heart; getting a red card
c. getting a 7; getting a heart
P39. About 42% of people have type O blood.
Suppose you select two people at random
and check whether they have type O blood.
a. Make a table like Display 5.50 on page 342

to show all possible results.
b. What is the probability that exactly one
of the two people has type O blood?
c. Make a tree diagram that illustrates the
situation.
P40. Suppose you select ten people at random.
Using the information in P39, find the
probability that
a. at least one of them has type O blood
b. at least one of them doesnt have type O
blood
P41. After taking college placement tests,
freshmen sometimes are required to repeat
high school work. Such work is called
remediation and does not count toward
a college degree. About 11% of college
freshmen have to take a remedial course in
reading. Suppose you select two freshmen at
22-03-2009 21:05
Lesson
8 de 10
random and check to see if they have

to take remedial reading. [Source: C. Adelman,
Principal Indicators of Student Academic Histories in
Postsecondary Education, 19722000 (2004), Tables 7.1
and 7.2, preview.ed.gov. Data from U.S. Department of
Education, NCES, National Education Longitudinal Study
of 1988 (NELS:88/2000), Fourth Follow-up, 2000.]
b. Show how to use a table like Display

5.50 on page 342 to find the probability
that exactly one freshman needs to take
remedial reading.
c. Show how to use a tree diagram to answer
the question in part b.
a. Find the probability that both freshmen

have to take remedial reading.
Exercises
E57. Which of these pairs of events, A and B, do
you expect to be independent? Give a reason
for your answer.
a. For a test for tuberculosis antibodies:
A: The test is positive.
B: The person has a relative with
tuberculosis.
b. For a test for tuberculosis antibodies:
A: A persons test is positive.
B: The last digit of the persons Social
Security number is 3.
c. For a randomly chosen state in the United
States:
A: The state lies east of the Mississippi
River.
B: The states highest elevation is more
than 8000 ft.
E58. Use the definition of independent events to
determine which of these pairs of events are
independent when you roll two dice.
a. rolling doubles; rolling a sum of 8
b. rolling a sum of 8; getting a 2 on the first
die rolled
c. rolling a sum of 7; getting a 1 on the first
die rolled
d. rolling doubles; rolling a sum of 7
e. rolling a 1 on the first die; rolling a 1 on
the second die
E59. Display 5.52 gives the handedness and
eyedness of a randomly selected group of
100 people. Suppose you select a person
from this group at random.
Display 5.52 Eyedness and handedness of

100 people.
a. Find each probability.

i. P(left-handed)
ii. P(left-eyed)
iii. P(left-eyed | left-handed)
iv. P(left-handed | left-eyed)
b. Are being left-handed and being left-eyed
independent events?
c. Are being left-handed and being left-eyed
mutually exclusive events?
E60. Display 5.53 gives the decisions on all
applications to two of the largest graduate
programs at the University of California,
Berkeley, by gender of the applicant. Suppose
you pick an applicant at random.
Display 5.53 Application decisions for the two

largest graduate programs at the
University of California, Berkeley, by
gender. [Source: David Freedman, Robert
Pisani, and Roger Purves, Statistics, 3rd ed. (New
York: Norton, 1997); data from Graduate Division,
UC Berkeley.]
22-03-2009 21:05
Lesson
9 de 10
a. Find each probability.

i. P(admitted)
ii. P(admitted | woman)
iii. P(admitted | man)
b. Are being admitted and being a woman
independent events?
c. Are being admitted and being a woman
mutually exclusive events?
d. Is there evidence of possible discrimination
against women applicants? Explain.
Display 5.54 gives the admissions data
broken down by which program the
applicant applied to.
Display 5.54 Application decisions for two of the

largest graduate programs at the
University of California, Berkeley,
by program and gender.
e. Does Program A show evidence of

possible discrimination against women
applicants? Explain.
f. Does Program B show evidence of
possible discrimination against women
applicants? Explain.
g. These data illustrate Simpsons paradox.
What is the paradox?
E61. When a baby is expected, there are two
possible outcomes: a boy and a girl.
However, they arent equally likely. About
51% of all babies born are boys.
a. List the four possible outcomes for a
family that has two babies (no twins).
b. Which of the outcomes is most likely?
Which is least likely?
c. What is the probability that both children

will be boys in a two-child family? What
assumption are you making?
d. Sometimes people say that girls run in
our family or boys run in our family.
What do they mean? If they are right,
how would this affect your answers in
parts ac?
E62. Shondra has six shirts (blue, green, red,
yellow, and two identical white shirts) and
four pairs of pants (brown slacks, black
slacks, and two identical pairs of jeans).
Suppose Shondra selects a pair of pants at
random and a shirt at random. Make a tree
diagram showing Shondras choices. Use the
tree diagram to compute the probability that
she wears a white shirt with black pants.
E63. Forty-two percent of people in a town have
type O blood, and 5% are Rh-positive.
Assume blood type and Rh type are
independent.
a. Find the probability that a randomly
selected person has type O blood and is
Rh-positive.
b. Find the probability that a randomly
selected person has type O blood or is
Rh-positive.
c. Make a table that summarizes this
situation.
E64. Schizophrenia affects 1% of the U.S.
population and tends to first appear between
the ages of 18 and 25. Today, schizophrenia
accounts for the fifth highest number of
years of work lost due to disability among
Americans ages 1544. Fewer than 30%
of people with this illness are currently
employed. Suppose you select a person from
the U.S. population at random. State whether
each question can be answered using the
information given. If the question can be
answered, answer it. [Source: National Institutes of
Health, www.nih.gov.]
a. What is the probability that the person is

schizophrenic?
22-03-2009 21:05
Lesson
10 de 10
of success for surgery A are 85% and the

b. What is the probability that the person is
chances of success for surgery B are 90%,
unemployed?
what are the chances that both would fail?
c. What is the probability that the person
Write an answer to this question for Marilyn.
is unemployed and schizophrenic?
[Source: PARADE, November 27, 1994, page 13.]
d. What is the probability that the person is
E70.
Joaquin computes the probability that if
unemployed or schizophrenic?
two
dice are rolled, then exactly one die will
e. What is the probability that the person
show a 4 as
is unemployed given that he or she is
schizophrenic?
P(4 on one die and not 4 on the other) =
f. What is the probability that the person
is schizophrenic given that he or she is
unemployed?
Joaquin is wrong. What did he forget?
E65. A family has two girls and two boys. To
E71. A batter hits successfully 23% of the
select two children to do the dishes, the
time against right-handed pitchers but
mother writes the names of the children on
only 15% of the time against left-handed
separate pieces of paper, mixes them up,
pitchers.
Suppose the batter faces rightand draws two different pieces of paper at
handed
pitchers
in 80% of his at bats and
random. What is the probability that both
left-handed pitchers in 20% of his at bats.
are girls?
What percentage of the time does he hit
E66. A committee of three students is to be
successfully?
formed from six juniors and five seniors.
E72. In Dave Barry Talks Back, Dave relates that
If three names are drawn at random, what
he and his wife bought an Oriental rug in
is the probability that the committee will
a failed attempt to become tasteful. Before
consist of all juniors or all seniors?
going out to dinner, they admonished their
E67. As stated in P41, about 11% of college
dogs Earnest and Zippy to stay away from
freshmen have to take a remedial course in
the rug.
reading. Suppose you take a random sample
NO!! we told them approximately 75 times
of 12 college freshmen.
while looking very stern and pointing at the
a. What is the probability that all 12 will
rug. This proven training technique caused
have to take a remedial course in
them to slink around the way dogs do when
reading?
they feel tremendously guilty but have no
b. What is the probability that at least one
idea why. Satisfied, we went out to dinner.
will have to take a remedial course in
I later figured out, using an electronic
reading?
calculator, that this rug covers approximately
2 percent of the total square footage of
E68. As stated in E64, about 1% of the U.S.
our house, which means that if you (not
population are schizophrenic. Suppose you
you personally) were to have a random . . .
take a random sample of 50 U.S. residents.
[Daves word omitted because this is a
a. What is the probability that none will be
textbook] attack in our home, the odds
schizophrenic?
are approximately 49 to 1 against your
having it on our Oriental rug. The odds
will be schizophrenic?
against your having four random attacks
E69. In the Ask Marilyn column of PARADE
on this rug are more than five million to
magazine, this question appeared:
one. So we had to conclude that it was done
on purpose.
Suppose a person was having two surgeries
performed at the same time. If the chances
22-03-2009 21:05
Lesson
1 de 7
Do you agree with Daves arithmetic? Do you

agree with his logic? [Source: Dave Barry, Dave Barry
c. Prove that
Talks Back, 1991 by Dave Barry. Used by permission of

Crown Publishers, a division of Random House, Inc.]
P( B | A ) =
E73. Two events A and B are independent if any

of these statements holds true:
P(A) = P(A | B)
P(B) = P(B | A)
Use the definition of conditional probability
to show that these three conditions are
equivalent as long as P(A) and P(B) dont
equal 0. That is, show that if any one is true,
so are the other two.
E74. Show that if events A and B are independent,
then events A and not B are also independent.
E75.
sometimes is called the law of total
probability.
a. Prove the law of total probability.
b. Use this law to find P(Dodgers win)
from the information in Display 5.51
on page 343.
This result is called Bayes Theorem.

d. Use Bayes Theorem to find
P(Dodgers win | day game).
E76. Two screening tests are used on patients
chosen from a certain population. Test I, the
less expensive of the two, is used about 60%
of the time and produces false positives in
about 10% of cases. Test II produces false
positives in only about 5% of cases.
a. Fill in this table of percentages.
b. A false positive is known to have

occurred in a patient tested by one of
these two tests. Find the approximate
probability that Test I was used, first by
direct observation from the table and
then by use of the formula in E75, part c.
Chapter Summary
Probability is the study of random behavior. The probabilities used in statistical
investigations often come from a model based on equally likely outcomes. In other
cases, they come from a model based on observed data. While you do not know
for sure what the next outcome will be, a model based on many observations gives
you a pretty good idea of the possible outcomes and their probabilities.
Two important concepts were introduced in this chapter: disjoint (mutually
exclusive) events and independent events.
Two events are disjoint if they cant happen on the same opportunity. If A and
B are disjoint events, then P (A and B) is 0.
Two events are independent if the occurrence of one doesnt change the
probability that the other will happen. That is, events A and B are independent
if and only if P(A) = P(A | B).
Chapter Summary
349
22-03-2009 21:06
Lesson
2 de 7
You should know how to use each of these rules for computing probabilities:
The probability that event A doesnt happen is
P(not A) = 1 P(A)
This rule often is used to find the probability of at least one success:
P(at least one success) = 1 P(no successes)
The Addition Rule gives the probability that event A or event B or both occur:
P(A or B) = P(A) + P(B) P(A and B)
If the events A and B are disjoint, this rule simplifies to
P(A or B) = P(A) + P(B)
The Multiplication Rule gives the probability that events A and B both occur:
P(A and B) = P(A) P(B | A) or P(B) P(A | B)
If A and B are independent events, this rule simplifies to
Review Exercises
E77. Suppose you roll a fair four-sided
(tetrahedral) die and a fair six-sided die.
a. How many equally likely outcomes are
there?
b. Show all the outcomes in a table or in a
tree diagram.
c. What is the probability of getting
doubles?
d. What is the probability of getting a sum
of 3?
e. Are the events getting doubles and
getting a sum of 4 disjoint? Are they
independent?
f. Are the events getting a 2 on the
tetrahedral die and getting a 5 on the
six-sided die disjoint? Are they
independent?
E78. Backgammon is one of the worlds oldest
games. Players move counters around the
board in a race to get home first.The
number of spaces moved is determined by
a roll of two dice. If your counter is able
to land on your opponents counter, your

opponent has to move his or her counter
to the bar where it is trapped until your
opponent can free it. For example, suppose
your opponents counter is five spaces
ahead of yours. You can hit that counter
by rolling a sum of 5 with both dice or by
getting a 5 on either die.
a. Use the sample space for rolling a pair of
dice to find the probability of being able
to hit your opponents counter on your
next roll if his or her counter is five spaces
ahead of yours.
b. Can you use the Addition Rule for
Disjoint Events to compute the
probability that you roll a sum of 5 with
both dice or get a 5 on either die? Either
do the computation or explain why you
cant.
E79. Jorge has a CD player attached to his alarm
clock. He has set the CD player so that when
its time for him to wake up, it randomly
selects one song to play on the CD. Suppose
22-03-2009 21:06
Lesson
3 de 7
there are nine songs on the CD and Jorges

favorite song is the third one.
a. Describe a simulation to estimate the
solution to this probability problem:
What is the probability that Jorge will
hear his favorite song at least once in
the next week?
b. Compute this probability exactly.
E80. This problem appeared in the Ask Marilyn
column of PARADE magazine and stirred up
a lot of controversy. Marilyn devoted at least
four columns to it.
A woman and a man (unrelated) each have
two children. At least one of the womans
children is a boy, and the mans older child
is a boy. Do the chances that the woman has
two boys equal the chances that the man has
two boys? [Source: PARADE,July 27, 1997.]
a. Design a simulation to answer this
question. (The process of designing a
simulation will help you clarify your
assumptions.)
b. What is the answer?
E81. As you saw in the example on page 305,
about 75% of men and 90% of women wash
their hands after using a public restroom.
Suppose you observe one man and one
woman at random. Compute each of these
probabilities.
a. P(both the man and the woman wash
their hands)
b. P(neither washes his or her hands)
c. P(at least one washes his or her hands)
d. P(man washes his hands | woman washes
her hands)
E82. As you saw in E18 on page 312, about
25% of U.S. residents claim to believe in
astrology. Suppose you select two U.S.
residents at random. Compute each of these
probabilities.
a. P(both believe in astrology)
b. P(neither believes in astrology)
c. P(at least one believes in astrology)
d. P(the second person believes in astrology |
the first person believes in astrology)
E83. Among recent graduates of mathematics

departments, half intend to teach high
school. A random sample of size 2 is to
be selected from the population of recent
graduates.
a. If mathematics departments had only
four recent graduates total, what is the
chance that the sample will consist of
two graduates who intend to teach high
school?
b. If mathematics departments had 10
million recent graduates, what is the
chance that the sample will consist of
two graduates who intend to teach high
school?
c. Are the selections technically
independent in part a? Are they
technically independent in part b? In
which part can you assume independence
anyway? Why?
E84. If you select a student at random from the
school described, what is the probability that
the student has ridden a merry-go-round or
a roller coaster, given the information about
that school? Make a table to illustrate each
situation.
a. In a particular school, 80% of students
have ridden a merry-go-round and 45%
have ridden a roller coaster. Only 15%
have done neither.
b. In another school, 30% of students have
ridden a merry-go-round but not a roller
coaster. Forty-five percent have ridden a
roller coaster but not a merry-go-round.
Only 20% have done neither.
E85. A website at Central Michigan University
collects data about college students. Students
were asked their political party preference
and which candidate they voted for in the
last presidential election. Display 5.55 (on
the next page) shows the results for the first
42 students who were 22 years old or older
and had either Democrat or Republican
party preference. Suppose you pick one of
these 42 students at random.
Chapter Summary 351
22-03-2009 21:06
Lesson
4 de 7
E87. Consider a situation with these three

probabilities:
Display 5.55 Results of voting and political party

preference survey. [Source: Central Michigan
University, stat.cst.cmich.edu.]
a. Find each of these probabilities.

i. P(Republican)
ii. P(voted Republican)
iii. P(voted Republican and Republican)
iv. P(voted Republican or Republican)
v. P(Republican | voted Republican)
vi. P(voted Republican | Republican)
b. Are being Republican and voting
Republican independent events?
c. Are being Republican and voting
Democrat mutually exclusive events?
E86. Display 5.56 gives probabilities for a
randomly selected adult in the United
States.
Display 5.56 Cell phone ownership, by age.
Write an expression for the probability that a

randomly selected adult
a. has a cell phone
b. is 25 or older, given that the person has a
cell phone
c. has a cell phone, given that the person is
age 18 to 25
d. is 25 or older and has a cell phone
e. is 25 or older or has a cell phone
a. Fill in a two-way table that has A and not

A for one way and B and not B for the
other way.
b. Are events A and B independent?
E88. Jack and Jill found that 60 out of 100 people
correctly identified the tap water. They are
uncertain what to conclude in their report.
They are concerned that perhaps everyone
was simply guessing and that 60 of them
got the right answer simply by chance. So
Jack and Jill flipped a fair coin 100 times
and counted the number of heads. They did
this over and over again. Out of thousands
of trials, they got 60 or more heads about
2.5% of the time. Write a conclusion for
their report.
E89. Many people who have had a stroke get
depressed. These patients tend to have
poorer outcomes than stroke patients who
arent depressed. A standard, but lengthy,
test for depression was given to 79 stroke
patients and 43 were found to be clinically
depressed. Researchers wondered if a
single question, Do you often feel sad or
depressed? would be just as good as the
longer test. Of the 43 patients found to be
clinically depressed, 37 answered yes, and
6 answered no. Of the 36 patients classified
as not clinically depressed, 8 answered yes
and 28 answered no. [Source: C. Watkins et al.,
Accuracy of a Single Question in Screening for Depression
in a Cohort of Patients After Stroke: Comparative Study.
BMJ 2001;323:1159 (17 November).]
Does the single question appear to be a

good diagnostic test for clinical depression
in stroke patients? First make a table that
summarizes this situation. Then use the ideas
of sensitivity, specificity, positive predictive
value, and negative predictive value in your
answer.
22-03-2009 21:06
Lesson
5 de 7
E90. A telephone area code is a three-digit

number that begins with a digit from 2
through 9. The second digit can be any
number from 1 through 8 or 0. The third
number must be different from the second.
How many area codes are possible, according
to these rules?
E91. A few years ago, the U.S. Post Office added
the zip plus 4 to the original five-digit zip
codes. For example, the zip code for the
national headquarters of the American Red
Cross is now 20006-5310.
For this problem, assume that a zip code
can be any five-digit number except 00000
and that the four-digit plus 4 can be any
number except 0000.
a. Assume that the U.S. population is
roughly 300 million. If all possible
five-digit zip codes are used, what is the

average number of people per zip code?
b. If all possible zip plus 4 codes are used,
what is the average number of people per
zip plus 4?
E92. DNA contains coded information that tells
living organisms how to build proteins by
combining 20 different amino acids. The
code is based on the sequence of building
blocks called nucleotides. There are four
kinds of nucleotides grouped into triplets
called codons. Each codon corresponds to an
amino acid. In a codon, order doesnt matter,
and the same nucleotide can be used more
than once. So, for example, AAB is the same
as ABA. Show that there is exactly one codon
for each of the 20 amino acids.
Chapter Summary
353
22-03-2009 21:06
Lesson
6 de 7
AP1. A student argues that extraterrestrials

will either abduct her statistics teacher by
tomorrow or they will not, and therefore
theres a 1 out of 2 chance for each of these
two events. Which of the following best
explains why this reasoning is incorrect?
The two events are not independent.
The two events are mutually exclusive.
The two events are not equally probable.
The two events are complements.
There are more than two events that
need to be considered.
AP2. Suppose you roll two dice. Which of the
following are independent events?
getting a sum of 8; getting doubles
getting a 1 on the first die; getting a
sum of 5
getting a 1 on the first die; getting
doubles
of a successful day. How should the student

conduct one run?
Assign the digits 09 to the retirees.
Assign the integers 115 to the children.
Randomly choose a number from each
group, pairing up a child and a retiree.
Repeat, without replacement, until all
10 retirees are assigned a child. Record
whether the day is successful.
Randomly choose digits one at a time,
with replacement, until 10 different
integers are chosen. Record the number
of integers needed to get 10 different ones.
Assign the digits 09 to the retirees.
Randomly choose 15 digits, with
replacement, and record whether all
10 digits were chosen or not.
Randomly choose 10 of these integers,
without replacement, and record the
proportion that are less than 10.
Randomly choose 10 of these integers,
with replacement, and record whether
all 10 integers were different or not.
AP3. Suppose you roll two dice. Which of the

following are mutually exclusive (disjoint)
events?
AP5. In a statistics classroom, 50% of the students
are female and 30% of the students got
an A on the most recent test. What is the
probability that a student picked at random
getting a 1 on the first die; getting a
from this classroom is a female who got an
sum of 5
A on the most recent test?
getting a 1 on the first die; getting
doubles
0.15
AP4. Suppose 15 children visit a retirement home
0.20
at various times during one day, and each
0.40
child randomly chooses one of 10 retirees
0.65
to visit. A successful day is one in which
cannot be determined from the
all 10 retirees are visited by at least one of
information given
these children. A statistics student wants to
use simulation to estimate the probability
22-03-2009 21:06
Lesson
7 de 7
AP6. In the fictional country of Valorim, 4% of

adults are smokers who get lung cancer,
8% are nonsmokers who get lung cancer,
22% are smokers who do not get lung
cancer, and the remaining 66% are
nonsmokers who do not get lung cancer.
What is the probability that a randomly
selected adult smoker gets lung cancer?
4/100
4/30
4/26
4/22
4/12
AP7. Seventy percent of Valorites can answer the
previous question correctly. Suppose you
select three Valorites at random. What is the
probability that at least one of the three can
answer correctly?
Investigative Tasks
AP9. So they stuck them down below, where
theyd be the first to go. This line from
the song about the Titanic refers to the
third-class passengers. The first two-way
table in Display 5.57 gives the number of
surviving Titanic passengers in various
classes of travel. The members of the ships
crew have been omitted from this table.
The second and third tables break down
survival by class and by gender. Use the
ideas about conditional probability and
independence that you have learned to
analyze these tables.
0.100
0.343
0.900
0.973
2.100
AP8. A new space probe, the Mars Crashlander,
has a main antenna and a backup antenna.
The probe will arrive at Mars and crash into
the surface. The main antenna has a 50%
chance of surviving the crash. The backup
has a 90% chance of surviving if the main
antenna survives, and a 20% chance of
surviving if the main antenna fails. What is
the probability that the Crashlander will have
at least one working antenna after the crash?
10%
27.5%
55%
60%
70%
Display 5.57 Titanic survival data, by class.

[Source: Journal of Statistics Education 3,
no. 3 (1995).]
AP Sample Test 355
22-03-2009 21:06
Lesson
1 de 2
Forty percent of people

have type A blood. A
blood bank is in dire
need of a type A donor
today. How many
donors will they have
to test before finding
the first type A? A
probability distribution
can describe the
chances of the possible
outcomes.
22-03-2009 21:07
Lesson
2 de 2
A probability distribution describes the possible numerical outcomes of a

chance process and allows you to find the probability of any set of possible
outcomes. Sometimes a probability distribution is defined by a table, like the ones
Jack and Jill made in Chapter 5. Sometimes a probability distribution is defined
by a curve, like the normal curve in Chapter 2. If you select a male at random,
all possible heights he could have are given by the values on the x-axis, and the
probability of getting someone whose height is between two specifined x-values is
given by the area under the curve between those two x-values. As you will learn in
this chapter, sometimes a probability distribution is defined by a formula.
Probability distributions for practical use come about in two different
waysthrough data collection and through theory. If you want to know the
chance of a paper cup landing on its closed bottom when tossed, the best way
to find out is to toss the cup many times and collect data. On the other hand, if
you want the probability of an event that can be modeled by coin flipping, you
can use the fact that the probability of heads is always 0.5 and build from there.
This chapter begins with data collection but quickly moves to theory.
Some types of probability distributions occur so frequently in practice that
it is important to know their names and formulas. Among these are the binomial
and geometric distributions, which closely (but not perfectly) reflect many
real-world situations and so are used to model many applications that have
similar characteristics.

the terminology of probability distributions
to construct probability distributions from data or theory
the concept and uses of expected value
to recognize and apply the binomial distribution
to recognize and apply the geometric distribution
357
22-03-2009 21:07
Lesson
1 de 24
6.1
Sampling reasons
from a sample to
a population, and
probability reasons
from a population
to a sample.
Random Variables and Expected Value

In Chapter 4 you learned about sampling, and in Chapter 5 you learned about
probability. What is the connection between these two topics?
Sampling deals with estimating a characteristic of a population by selecting
data from that population. A random sample of households from your
community helps you estimate the percentage of all households in your
community that own a motor vehicle.
Probability deals with the chance of getting a specified outcome in a sample
when you know about the population. If you know 91% of households in
your community own a motor vehicle, then you can calculate the probability
that three randomly selected households all own a motor vehicle.
One of the most important questions statistics tries to answer is What might
an unknown population look like? A census, which would answer this question,
is usually impossible or impractical. Fortunately, early 20th-century statisticians
had this extraordinary insight: Use probability to learn what samples from
populations with known distributions tend to look like. Then, take a sample
from the unknown population and compare the results to those from known
populations. The populations for which your sample appears to fit right in are the
plausible homes for your sample. This chapter begins to formalize this profound
idea, which will be developed more fully in the remaining chapters of the book.
Probability Distributions from Data

Suppose you are working for a contractor who is building 500 new single-family
houses and wants to know how many parking spaces will be needed per house.
Your job is to sample 500 households for the purpose of estimating the number of
vehicles per household. Fortunately, before starting the work of taking a survey,
you find a distribution of the number of vehicles per household (Display 6.1)
from the U.S. Census Bureau. Assuming that these national results are typical of
your community, your work is done! You can report to your boss, for example,
that about 0.385 + 0.137 + 0.058, or 0.580 (more than half) of the households
will have two vehicles or more.
Your boss is pleased that you have so much extra time and gives you a new
project. Your construction company is also building 500 duplexes, and your boss
wants to know something about the total number of vehicles that will be parked
by the two households occupying a duplex.
Display 6.1 does not give you that information. You are beginning to think
you will have to take a survey when you realize that you can take two households
at random from the distribution in Display 6.1 and add their numbers. Repeating
this 500 times, you will get an approximation to the distribution your boss is
asking for. This approximation depends on the assumptions that households living
in duplexes mirror households in general with respect to vehicles per household,
and that the number of vehicles in one household is independent of the number of
vehicles in the neighboring household. How good the approximation is depends
on how well these assumptions hold up. This is typical of all modeling problems;
the accuracy of the result depends on the appropriateness of the model.
358 Chapter 6 Probability Distributions
22-03-2009 21:07
Lesson
2 de 24
Display 6.1
The number of motor vehicles per household. The

4 represents 4 or more, but the proportion of
households with more than four vehicles is very
small. [ Source: U.S. Census Bureau, American Community Survey,
2004, factfinder.census.gov.]
You begin by cutting up 1000 identical slips of paper, planning to write 0

on 88 of them, 1 on 332 of them, 2 on 385 of them, 3 on 137 of them, and 4 on
58 of them. You will then place the slips in a box, mix them up, and draw one at
random. You will have an
or 0.088, chance of getting a household with no
motor vehicles, a
or 0.332, chance of getting a household with one motor
vehicle, and so on. You will replace the slip, mix the slips up, and draw another.
Then you will add the two numbers to get the total number of vehicles for the first
duplex. You will need 500 pairs of draws to simulate the results for 500 duplexes.
After cutting 321 slips of paper, a quicker method occurs to youuse a table
of random digits! To use a list of random digits to select a household at random
from this distribution, you make the assignments as in Display 6.2. You need
three-digit numbers because the percentages have three decimal places.
Display 6.2 Assigning random digits to represent the motor

vehicle population.
For example, 88 out of 1000 (0.088) three-digit sequences are needed to represent
households with no motor vehicles. Here the 88 numbers 001 through 088 were
used, although you could use 000 through 087 or any other choice as long as
there are 88 numbers. The next 332 three-digit sequences (089420) represent
households with one vehicle, and so on. Now that youve assigned all the
359
22-03-2009 21:07
Lesson
3 de 24
three-digit sequences, youre ready to draw your pairs of households. Divide a

string of random digits into groups of three.
391 545 177 324 106 845 248
The first sequence, 391, represents a household with one motor vehicle
because 391 is in the interval 089420. The second sequence, 545, represents a
household with two motor vehicles. So for the first duplex, you have a total of
three motor vehicles. In sampling this way, you might come across a sequence that
you have already used. You should go ahead and use it again because each random
sequence represents many households, not just one, and you want to keep the
probabilities the same for each household selected.
Such a simulation for 500 duplexes (each consisting of randomly selecting two
values from the distribution in Display 6.1 and adding them) is shown in Display
6.3. This distribution differs considerably from the one in Display 6.1 and could
not be anticipated without some clever work with probabilities.
Display 6.3
Simulated distribution of total vehicles in two

households.
You can now report to your boss, for example, that there is an 0.088 chance
that a single-family household will have no vehicles but only an 0.008 chance that
the occupants of a duplex will have no vehicles.

D1. When you sample from distributions like the one in Display 6.1 using
random digits, explain why repeated random digits should be used again
rather than discarded.
D2. Compare the shape, center, and spread of the distribution in Display 6.3
with that in Display 6.1.
D3. Your boss finds it odd that there is a fairly good chance (0.137 0.058,
or 0.195) that a single household will need to park three or more vehicles
while there is little chance (0.064 0.010 0.002, or 0.076) that the two
households in a duplex will need to park six or more vehicles. Explain to
your boss why this is reasonable.
22-03-2009 21:07
Lesson
4 de 24
Probability Distributions from Theory

The language of probability distributions reflects the chance involved in random
outcomes. The variable of interest, X, is called a random variable because its
numerical values vary depending on the results of a particular trial. Thus, if
X represents the number of motor vehicles in a randomly selected household,
P(X = 2) 0.385, because about 38.5% of randomly selected households have
two motor vehicles.
You now have two ways to establish probability distributions from data.
Sometimes relative frequencies from data already collected can serve as a
probability distribution for a random variable to be studied in the future, as
with the distribution of motor vehicles per household from the Census Bureau.
At other times a simulation is a good way to build an approximate probability
distribution, as with the approximation of the probability distribution of the
total number of vehicles in two randomly selected households in a duplex. There
is a third way: Use the rules of theoretical probability to construct a probability
distribution from basic principles and assumptions. The next three examples
illustrate this approach.
Example: Probability Distributions from Rolling Two Dice

In Chapter 5, you constructed a table like Display 6.4, which lists all possible
outcomes when two dice are rolled. From this table, you can construct
different probability distributions, depending on the numerical summary
that interests you.
Display 6.4 The 36 outcomes when rolling two dice.
For example, if you are playing Monopoly, you need to find the sum of the two
dice. If you are playing backgammon, you might be more interested in the larger
of the two numbers. Construct each of these probability distributions. Then
find the probability that, when you roll two dice, the sum is 3. Then find the
probability that the larger number is 3.
Solution
Display 6.5 shows these two probability distributions, one of the sum of two dice
and one of the larger number on the two dice. (In the case of doubles, the larger
number and the smaller number are the same.)
361
22-03-2009 21:07
Lesson
5 de 24
Display 6.5 Two different probability distributions from rolling

two dice.
When you roll two dice, the probability that the sum of the numbers is 3 is
The probability that the larger number is 3 is
Example: Sampling Lung Cancer Patients

Display 6.6 shows data on the proportion of lung cancer cases caused by
smoking. Suppose two lung cancer patients are randomly selected from the large
population of people with that disease. Construct the probability distribution
of X, the number of patients with lung cancer caused by smoking.
Display 6.6 Lung cancer cases caused by smoking. [Source: Lung

Disease Data, American Lung Association, 2006.]
Solution
There are four possible outcomes for the two patients. With yes representing
caused by smoking and no representing not caused by smoking, the
possibilities are
no for 1st patient and no for 2nd patient
no for 1st patient and yes for 2nd patient
yes for 1st patient and no for 2nd patient
yes for 1st patient and yes for 2nd patient
22-03-2009 21:07
Lesson
6 de 24
The fact that the patients are selected at random from a large population implies
that the outcomes for the two selections may be considered independent. Using
the rules of Chapter 5,
P(no for 1st patient and no for 2nd patient) = P(no for 1st) P(no for 2nd)
= (0.1)(0.1) = 0.01
P(no for 1st patient and yes for 2nd patient) = (0.1)(0.9) = 0.09
P(yes for 1st patient and no for 2nd patient) = (0.9)(0.1) = 0.09
P(yes for 1st patient and yes for 2nd patient) = (0.9)(0.9) = 0.81
The first of these outcomes results in X = 0, the second and third each result in
X = 1, and the fourth results in X = 2. Because the second and third outcomes
are disjoint, their probabilities can be added. The probability distribution of X is
then given by this table:
Example: Sampling Hockey Teams

Display 6.7 gives the capacity of the home arenas of the fi ve teams in the Atlantic
Division of the National Hockey League. Suppose you pick two teams at random
to play a pair of exhibition games, with one game played in each home arena.
Construct the probability distribution of X, the total number of people who could
attend the two games. Then fi nd the probability that the total possible attendance
will be at least 36,000.
Display 6.7 Home arena capacities.
Solution
There are ten ways to pick two teams from the five:
= 10. These ten
pairs of teams, with the total possible attendance, are shown in Display 6.8.
363
22-03-2009 21:07
Lesson
7 de 24
Display 6.8 Capacities for pairs of exhibition games.
The probability distribution of the total possible attendance for two games is
given in Display 6.9.
Display 6.9
Probability distribution of total possible attendance

at two games.
The probability that the total attendance is at least 36,000 is 0.2 + 0.2, or 0.4.

D4. The boss of your construction company isnt quite satisfied with the
simulation in Display 6.3 on page 360 and would like to know the exact
probability that two randomly selected duplex households have a total of two
vehicles. Use the rules of probability from Chapter 5 and the information in
Display 6.1 on page 359 to compute this probability exactly. How does your
exact probability compare to the estimate in Display 6.3?
Expected Value and Standard Deviation

Your boss has looked at the table in Display 6.1 on page 359 and would like to
know the average number of vehicles that would need parking spaces for a
single-family house.
Assuming that your community is like the entire United States, you can find
the mean number of vehicles per household directly from Display 6.1. You expect
22-03-2009 21:07
Lesson
8 de 24
8.8% of households to have no vehicles, 33.2% to have one vehicle, and so on.
Thus, you expect the mean number of vehicles per household to be
= 0(0.088) + 1(0.332) + 2(0.385) + 3(0.137) + 4(0.058) = 1.745
Computing the mean can be easily accomplished by adding a third column to
the table, as in Display 6.10. Notice that the mean, 1.745, is the balance point of
the histogram in Display 6.1.
Display 6.10
Computing the mean from a probability

distribution.
The mean of a probability distribution has a special name, although

continuing to call it the mean is not wrong.
The mean of a probability distribution for the random variable X is called its
expected value and is usually denoted by , or E(X).
You can report to your boss that the expected number of vehicles per
household is 1.745. However, you realize that you should give your boss an
estimate of how much a typical household is likely to differ from this average.
To calculate the standard deviation of the number of vehicles per household, you
find the expected value of the square of the deviations from the mean, which
is called the variance of the probability distribution. As always, the standard
deviation is then the square root of the variance. The variance of the distribution
in Display 6.1 on page 359 is given by
= (0 1.745)2 (0.088) + (1 1.745)2(0.332) + (2 1.745)2(0.385)
+ (3 1.745)2(0.137) + (4 1.745)2(0.058)
0.9880
The standard deviation is

0.994. You tell your boss that the
neighborhood of single-family houses can expect to have 1.745 motor vehicles
per house, but quite often a house will have about one vehicle more or less than
this mean.
The formulas for the expected value (mean) and standard deviation of a
probability distribution listed in a table, shown in the following box, parallel those
for a relative frequency table.
365
22-03-2009 21:07
Lesson
9 de 24
Formulas for the Mean and Variance of a Probability Distribution

Listed in a Table
The expected value (mean)
listed in a table are
and variance
of a probability distribution
where pi is the probability that the random variable X takes on the specific
value xi. To get the standard deviation, take the square root of the variance.
Now your boss asks about the duplexes. For the duplexes, the simulated data
of Display 6.3 on page 360 should provide a good approximation of the actual
probability distribution of this random variable, call it Y. The mean, variance, and
standard deviation as calculated from the 500 simulated values in the frequency
The neighborhood of duplexes can expect to have 3.530 motor vehicles per duplex
but often will see up to 1.357 vehicles more or less than this.
Example: CDs Purchased by USC Students

The probability distribution in Display 6.11 was constructed from a survey at the
University of Southern California (USC). The random variable is the number of
CDs purchased by a student in the previous year. For example, if you select a USC
student at random, the probability is 0.06 that he or she purchased no CDs in the
previous 12 months. Estimate the expected value and standard deviation of this
distribution.
Display 6.11
CDs purchased by USC students. [Source: Mark Latonero,

Survey of Usage for MP3, the Online Music Service, June 2000,
www.learcenter.org.]
Solution
Student: I remember how to do this. First I have to make a reasonable
estimate of the center of each interval. For example, for the
22-03-2009 21:07
Lesson
10 de 24
interval of 15 CDs, Ill use 3. Then I simply use the formulas:
= 0(0.06) + 3(0.28) + 8(0.25) + 15.5(0.21)

+ 35.5(0.17) + 55(0.03)
= 13.78
and
= (0 13.78)2(0.06) + (3 13.78)2(0.28)
+ (8 13.78)2(0.25) + (15.5 13.78)2(0.21)
+ (35.5 13.78)2(0.17) + (55 13.78)2(0.03)
184.0766
So the standard deviation,
, is approximately 13.57.
Statistician: Your computations are perfect. Now, what is the expected number
of CDs purchased by a USC student?
Student: Easy. My computations give 13.78, but because we cant have part
of a CD, we say the expected number of CDs purchased by a
USC student is 13 or 14.
Dont round expected
values to whole
numbers.
Statistician: And you were doing so well. Its true you cant have part of a
CD, but expected value is the same thing as the average, and you
know that the average doesnt have to be one of the values in the
distribution. So you should say that the expected number of CDs
is 13.78. What does the fact that the mean and standard deviation
are about equal tell you about the shape of this distribution?
Student: Well, the number of purchases cant go below zero, which is only
one standard deviation below the mean. So, there must be a long
tail toward the larger values.
Statistician: You are rapidly becoming a statistical thinker!
You can use a calculator to quickly find the mean and variance of a probability
distribution listed in a table. The mean and variance for the data in Display 6.11
are shown here. [See Calculator Note 6A to learn how to calculate these statistics.]
367
22-03-2009 21:07
Lesson
11 de 24

D5. In Chapter 2, you used this formula to compute the mean, , of the values in
a frequency table, where x is the value, f is its frequency, and n = f:
a. Use this formula to compute the mean of the values in this frequency table.
b. Suppose the data had been given in a relative frequency table like this
one, which shows the proportion of times each value occurs. Fill in the
rest of the second column.
c. Show that you can find the mean, , using the formula
D6.
D7.
D8.
D9.
d. Discuss how the formula in part c relates to the formula for the expected
value, .
Compare the mean number of vehicles in single-family households to the
mean number in a duplex. Compare the variances. What do you notice?
Eighteen percent of high school boys and 10% of high school girls say they
rarely or never wear seat belts. Suppose one high school boy and one high
school girl are selected at random, with the random variable of interest
being the number in the pair who say they rarely or never wear seat belts.
Describe two ways of finding the expected value and standard deviation of
this random variable, at least approximately. (Use only the material that has
been presented in this chapter so far.)
Define a random variable for Display 6.4 on page 361 that is different from
the two random variables in Display 6.5. Give its probability distribution
and compute its expected value and standard deviation.
This sentence appeared in the British humor magazine Punch:
The figure of 2.2 children per adult female was felt to be in some respects
absurd, and a Royal Commission suggested that the middle classes be paid
money to increase the average to a rounder and more convenient number.
[Source: M. J. Moroney, Facts from Figures (Baltimore: Penguin Books, 1951).]
Who would find the figure absurdthe student or the statistician from the
dialogue on page 367?
22-03-2009 21:07
Lesson
12 de 24
Expected Value in Everyday Situations

In many real-life situations, you can determine the best course of action by
considering expected value.
Example: The Wisconsin Lottery

The Wisconsin lottery had a scratch-off game called Big Cat Cash. It cost $1 to
play. The probabilities of winning various amounts are listed in Display 6.12. Find
and interpret the expected value when playing one game.
Display 6.12 Probabilities for Wisconsin scratch-off game.

[Source: www.wilottery.com.]
Solution
First, note that the probabilities dont sum to 1its not even close. Thats because
the most likely outcome is winning nothing. So imagine another row with $0
for Winnings and 0.7793 for Probability. Using the expected-value formula,
you can verify that the expected value for this probability distribution is 0.6014.
This means that if you spend $1 to play this game, you expect to get 60.14
back in winnings. Of course, you cant get this amount on any one play, but over
the long run that would be the average return. Another way to understand this
is to imagine playing the game 1000 times. You expect to get back $601.40, but
you will have spent $1000. The standard deviation of the winnings per game
is $4.040, which is quite large because of the possibility of winning one of the
larger amounts.
The expected value may not be of much importance to an individual player
(unless he or she is going to play many times) but it is of great importance to the
State of Wisconsin, which can expect to pay out $601.40 for every $1,000 bet.
Example: Burglary Insurance

A person who lives in a large city comes to your insurance agency and asks you
to insure her household so that if it is burgled, you would pay her $5000. What
should you charge per year for such insurance?
Solution
How much you charge per year depends on how likely it is that her household
will be burgled. Insurance companies and the mathematicians they employ,
called actuaries, keep careful records of various crimes and disasters so they can
369
22-03-2009 21:07
Lesson
13 de 24
know the probabilities that these will occur. Unfortunately, you neglected to hire
an actuary, and all the information you can fi nd is the nationwide rate of 41.9
burglaries per 1000 urban households.
Display 6.13 gives the two possible outcomes, their payouts, and their
probabilities.
Display 6.13 Table of payouts and their probabilities. [ Source:

U.S. Bureau of Justice Statistics, Criminal Victimization, 2004,
www.ojp.usdoj.gov.]
The expected payout per policy is

E(X) =
= 0(0.9581) + 5000(0.0419) = 209.50
You expect to break even if you charge $209.50 for the insurance. But as a good
businessperson, you will actually charge more to cover your costs of doing
business and to give yourself some profit.

D10. A few years ago, a fast-food restaurant chain had a scratch-off card
promotion that contained four separate games. The player could play
only one of them and could choose which one. The probability of winning
game A was , and the prize was a drink worth $0.55. The probability of
winning game B was , and the prize was a food item worth $0.69. The
probability of winning game C was , and the food prize was worth $1.44.
The probability of winning game D was , and the food prize was worth
$1.99. Assuming you like all the prizes, which is the best game to play? This
question is open to different interpretations, so well be specific: On which
game are your expected winnings greatest?
You can organize the information for each game in a table. For example, for
game A, the table looks like this:
The expected winnings if you play game A are

= (0.55)(0.5) + (0)(0.5) = 0.275
or $0.275.
a. Compute the expected winnings from games B, C, and D. On which
game are your expected winnings greatest?
b. For what reasons would people choose to play the other games, even if
they knew the expected values were lower?
22-03-2009 21:07
Lesson
14 de 24
Rules for Means and Variances

Adding a constant
recenters, or shifts,
a distribution, while
multiplying by a
constant stretches
or shrinks it.
In Section 2.4, you learned that if you recenter a data setthat is, add the same
number c to all the valuesit doesnt change the shape or standard deviation but
adds c to the mean. If you rescale a data setthat is, multiply all the values by
the same nonzero number dit doesnt change the basic shape but multiplies the
mean by d and the standard deviation by |d| where |d| indicates the absolute value
of d. These rules for distributions of data also apply to probability distributions.
Linear Transformation Rule: The Effect of a Linear Transformation
and
of X on
Suppose you have a probability distribution with random variable X, mean X ,
and standard deviation . If you transform each value by multiplying it by d
and then adding c, where c and d are constants, then the mean and standard
deviation of the transformed values are given by
Example: Tripling the Wisconsin Lottery

Suppose that, for a special promotion, the prizes are tripled for the Wisconsin
scratch-off game Big Cat Cash described in Display 6.12 on page 369. What are
the expected winnings for a person who plays one game? What is the standard
deviation?
Solution
The most effi cient method of doing this problem is to transform the expected value,
0.6014, and the standard deviation, 4.040, for the original game using the rules of
linear transformation. Here, c = 0 and d = 3, so the expected value of the tripled
prize is 3(0.6014), or $1.804, with a standard deviation of 3(4.040), or $12.12.
A second method of doing this problem is to rewrite Display 6.12 showing
the winnings during the special promotion, as shown in Display 6.14.
Display 6.14
Winnings during special promotion.
371
22-03-2009 21:07
Lesson
15 de 24
Then use the formulas for the expected value and variance from page 366 as
before:
E(X) = 0(0.7793) + 3(1/10) + 6(1/14) + 9(1/24) + 54(1/200) + 150(1/389)
+ 450(1/20000) + 2700(1/120000)
1.804
Var(X) = (0 1.804)2 0.7793 + (3 1.804)2(1/10) + (6 1.804)2(1/14)
+ (9 1.804)2(1/24) + (54 1.804)2(1/200)
+ (150 1.804)2(1/389) + (450 1.804)2(1/20000)
+ (2700 1.804)2(1/120000)
146.89
The expected winnings are $1.804, with a standard deviation of $12.12.

Suppose that, instead of tripling the winnings, the special promotion is to
give a person three tickets instead of just one. What are the mean and standard
deviation of the winnings now? The rules in the next box will help you solve this
new problem.
Addition and Subtraction Rules for Random Variables
If X and Y are random variables, then
and if X and Y are independent, then
Example: Getting Three Lottery Tickets

Refer to the example about the Wisconsin lottery game with an expected value of
$0.6014 and standard deviation $4.040. Suppose a person is given three tickets.
What are the expected total winnings and the standard deviation of this total?
22-03-2009 21:07
Lesson
16 de 24
Solution
Let X + Y + Z represent the persons total winnings on three tickets. Then
is the expected value of the total winnings on three tickets. This problem requires
a generalization of the Addition Rule:
= 0.6014 + 0.6014 + 0.6014

=1.804
Because the winnings on the three tickets can be considered independent, the
variance of X + Y + Z is
= 4.040 2 + 4.040 2 + 4.040 2

48.965
The person expects winnings of $1.804 with standard deviation $6.997.
Note the difference in the previous two examples. In the first, there was one
randomly selected ticket and its value was tripled. In the second, there were three
randomly selected tickets and their values were added. The expected values are
the same, but there is more variability when the winnings from a single ticket are
tripled. The next example combines both sets of rules.
Example: Expected Savings

Suppose you earn $12 an hour for tutoring but spend $8 an hour for dance
lessons. You save the difference between what you earn and the cost of your
lessons. The number of hours you spend on each activity in a week varies
independently according to the probability distributions in Display 6.15. Find
your expected weekly savings and the standard deviation of your weekly savings.
Display 6.15 Probability distributions of the number of hours per

week taking dance lessons and tutoring.
Solution
Let X be the number of hours per week taking dance lessons and Y be the number
of hours per week tutoring. The expected number of hours you take dance
lessons, , is 0(0.4) + 1(0.3) + 2(0.3), or 0.9, with a standard deviation, , of
373
22-03-2009 21:07
Lesson
17 de 24
0.831 hour. Similarly, the expected number of hours you tutor, or , is 2.3, with a
standard deviation, , of 1.1 hours. You spend $8 for each hour of dance lessons,
so by the Linear Transformation Rule your mean weekly expenditure for dance
lessons is
or $7.20. Similarly, if you earn $12 for each hour of tutoring, then by the Linear
Transformation Rule, your mean weekly earnings from tutoring are
or $27.60. By the Subtraction Rule, your expected weekly savings are
or $20.40.
By the Linear Transformation Rule, the standard deviation of the cost of your
weekly lessons,
,is 8(0.831), or $6.648, and the standard deviation of your
weekly earnings from tutoring ,
, is 12(1.1), or $13.20. Because the amounts
are independent, you can now use the Subtraction Rule to fi nd that the variance
of your weekly savings is
Taking the square root, the standard deviation,
, is about $14.78.
For most students, the most surprising rule says that to get the variance of
the difference of two independently selected variables, you add the individual
variances. Why add and not subtract? Activity 6.1a will help you see the reason.
The Sum and Difference of Two Rolls of a Die

What youll need: a die, two partners
You will roll your die twice. One partner will record the sum of the numbers on
the two rolls. The other partner will record the differencethe number on the
first roll minus the number on the second roll. For example, if the first roll is 3
and the second roll is 5, the first partner records 8 and the second records 2.
1. What do you think the distribution of sums (first roll plus second roll) and
the distribution of differences (first roll minus second roll) will look like
after many rolls? Where will they be centered? Which will have the larger
spread?
2. Roll the die until your class has recorded 100 sums and 100 differences.
3. Plot the 100 sums and the 100 differences on separate dot plots. How do
the shapes compare? How do the spreads compare?
4. What is the mean of your distribution of sums? Of differences? How do
these means compare to the mean of the distribution of a single roll of a die?
(continued)
22-03-2009 21:07
Lesson
18 de 24
5. Compute the variance of your distribution of sums. Compute the

variance of your distribution of differences. How do the variances of your
distributions of sums and differences compare with the variance of the
outcomes from a single roll of a die? How do they compare with each
other?
6. Use the formulas in the box on page 366 to compute the mean and variance
of the distribution of sums. Use the same formulas to compute the mean
and variance of the distribution of differences. How close are your results
in step 5 to those you get using the formulas?
7. Explain why it should be the case that the spread in the differences is equal
to the spread in the sums.

D11. If you expect to work 10 hours next week with a standard deviation of
2 hours and you expect to study 15 hours next week with a standard
deviation of 3 hours, what is the total number of hours you expect to spend
working or studying? Use the rule for adding variances to estimate the
standard deviation of the total number of hours spent studying or working.
Is it reasonable to use the rule in this case? Explain.
D12. Provide an intuitive argument as to why the variability in a sum should be
larger than the variability in either variable making up the sum. Do the same
for the variability in a difference.
Summary 6.1: Random Variables and Expected Value

A probability distribution describes the possible numerical outcomes, x, of a
chance process and allows you to find the probability, P(x), of any set of possible
outcomes. A probability distribution table for a given random variable lists each
of these outcomes in one column and their associated probability in another
column. You can compute the expected value (mean) and variance of a probability
distribution table using the formulas
Estimating the expected value has many real-world applications. For example,
you can figure out your expected weekly savings or the break-even price for
insurance.
For random variables X and Y and constants c and d,
the mean and standard deviation of a linear transformation of X are given by
375
22-03-2009 21:07
Lesson
19 de 24
if you add or subtract random variables X and Y, the mean is given by
if X and Y are independent, then the variance is given by
Practice
P1. Refer to Display 6.2 on page 359. Use this
line of random digits to simulate selecting
two households at random and counting
the number of motor vehicles in both
households together. Then repeat for two
more households.
177324106845248
P2. As you saw in the example on page 362,
90% of lung cancer cases are caused by
smoking. How would you assign the
numbers in a random digit table so that they
represent the distribution in Display 6.6? Use
a random digit table (Table D on page 828)
to select a lung cancer patient at random and
then tell whether smoking was responsible
for the patients lung cancer.
P3. The distribution in Display 6.16 gives the
number of children per family in the United
States. Describe how to use this line from
a random digit table to find the number of
children in a randomly selected family:
48830994251773890817
Use your process to find the total number
of children in three randomly selected
families.
Display 6.16 The number of children per family.

[Source: U.S. Census Bureau, Statistical Abstract of
the United States, 20042005, www.census.gov.]
P4. Use the appropriate rule of probability from

Chapter 5 and the fact that there is an 0.088
chance that a single household will have
no vehicles to compute the probability that
the two households in a duplex will have
a total of zero vehicles. How close is your
computation to the estimate in Display 6.3
on page 360?
22-03-2009 21:07
Lesson
20 de 24

P5. Suppose you roll two dice. Construct the
probability distribution of the smaller of
the two numbers.
P6. The best estimate of the percentage of all
email that is spam (electronic junk mail)
is about 70%. Suppose you randomly
select three email messages. Construct
the probability distribution of the random
variable, X, defined as the number of
messages in your sample that are spam.
Begin by listing the outcomes, as was done
in the lung cancer example on page 362.
P7. Suppose five students (Claire, Charlotte,
Max, Alisa, and Shaun) belong to a club
and two of them (Alisa and Shaun) take
a bus to school. You randomly select
three students from the club to serve on
a committee that meets before school.
Your goal is to construct the probability
distribution of the random variable, X,
defined as the number of committee
members who take a bus to school.
a. List all possible random samples of size 3
from this group of five students.
b. Recalling that in simple random sampling
all samples of the same size have
equal probability of selection, assign a
probability to each sample listed in part a.
c. Based on parts a and b, construct the
probability distribution of the random
variable X.
P8. The distribution in Display 6.16 gives the
number of children per family in the United
States. Compute the mean and standard
deviation of this population. Count 4 or
more as 4.
P9. Describe the shape of the distributions of
the two random variables in Display 6.5 on
page 362. Find their expected values and
standard deviations.
P10. Suppose a quality-control inspection rule
calls for randomly sampling two computer
monitors from each group of six that comes

off the assembly line and inspecting them
for defects. On a particularly bad Monday
morning, the first six monitors coming off
the line include three monitors with defects.
a. List all possible samples of size 2 that
could be selected from the six monitors.
Then make a probability distribution
table for X, the number of sampled
monitors with defects.
b. Find the expected value and standard
deviation of X.
c. If either of the two sampled monitors
shows any defects, the companys policy
calls for shutting down the line for an
investigation of the cause of the defects.
What is the probability that the line will
be shut down after the inspection of two
randomly selected monitors from the first
six that come off the line this Monday
morning?
P11. For a raffle, 500 tickets will be sold.
a. What is the expected value of a ticket if
the only prize is worth $600?
b. What is the expected value of a ticket if
there is one prize worth $1000 and two
prizes worth $400 each?
P12. Refer to the example on burglary insurance
on page 369. The burglary rate for a
suburban location is 23.2 burglaries per
1000 households. If the home is in a
suburban area, what should you charge
for $5000 of burglary insurance in order
to break even?
P13. A neighborhood hardware store rents a
steam cleaner for 3 hours at a fee of $30.00
or for a day (up to 8 hours) at a fee of $50.00.
Sometimes the cleaner can be rented out
for two 3-hour periods in the same day.
By studying rental records, the manager
of the store comes up with approximate
probabilities for renting the cleaner on a
typical Saturday, as shown in Display 6.17
(on the next page).
6.1 Random Variables and Expected Value 377
22-03-2009 21:07
Lesson
21 de 24
b. You realize that a randomly selected

group of 200 households probably wont
have exactly the expected number of
vehicles. How far off is your estimate
likely to be?
Display 6.17 Steam cleaner rental probabilities.
c. What assumptions are you making in
your answers to parts a and b? Do you
a. Find the expected rental income for the
think they are reasonable for apartment
store on a typical Saturday.
dwellers?
b. Find the standard deviation of the
P17. Find the expected value and standard
expected rental income on a typical
deviation of the total number of vehicles
Saturday. Given that the standard
deviation is supposed to measure a
per duplex using only the information
typical deviation from the mean, does
from Display 6.1 on page 359. How do your
this value seem reasonable?
answers compare with the approximate
values found from Display 6.3 (recorded on
page 360)?
P14. Construct a probability distribution for the
P18. Display 6.18 shows the distribution of the
difference when a die is rolled twice (first roll
number of color televisions per household in
minus second roll).
the United States.
a. What is the shape of this distribution?
b. Compute the mean and the variance
of this distribution using the standard
formulas
c. Verify that the mean of this probability

distribution is equal to , or 0, and
the variance is
+ , or 2 where
Display 6.18 Color televisions per household.
and are the mean and standard
[Source: U.S. Census Bureau, Statistical Abstract
deviation of the distribution of a single
of the United States, 2006.]
roll of a die.
a. A random variable is defined as the
P15. Refer to the example on expected savings
number of color televisions observed
on page 373. Find the amount you expect to
in a single household selected at
save in a year and the standard deviation of
random from across the country. Find
your yearly savings.
the (approximate) expected value and
P16. Your construction company plans to build
standard deviation for the probability
distribution of this random variable.
apartment buildings, each with one common
parking lot. Your boss would like an estimate
Mark the expected value on a plot of this
of how many spaces would be needed for a
distribution.
building with 200 households.
b. Suppose the random variable of interest
a. Using the mean number of vehicles per
is the total number of color televisions
household, 1.745, from page 365, find how
in two households selected at random.
many vehicles you would expect for an
Compute the expected value and standard
apartment building with 200 households.
deviation of this new random variable.
22-03-2009 21:07
Lesson
22 de 24
Exercises
E1. A large class was assigned a diffcult
homework problem. Display 6.19 shows
the scores the students received and the
proportion of students who received each
score.
E4.
E5.
Display 6.19
Distribution of homework scores.
a. Compute the mean and standard

deviation of the scores.
b. Describe how to use this list of random
digits to take a sample of size 4 from
these scores:
30558 45957 36911 97199 08432
(Note: The digits are grouped in batches
of five to make them easier to read.
Theyre to be interpreted as a single string
of digits.)
E2. Display 6.20 gives ages of students enrolled
at the University of Texas, San Antonio.
E6.
E7.
Display 6.20
Ages of students. [Source: University of

Texas, www.utsa.edu.]
a. Make a reasonable estimate of the center

of each age group, and compute the mean
and standard deviation of the ages of
these students.
b. Describe how to use this list of random
digits to take a sample of six ages:
30558 45957 36911 97199 08432
E3. Refer to Display 6.1 on page 359. If your boss
provides two parking spaces per single-family
E8.
house, how many vehicles do you estimate

will have to be parked on the street in the
new neighborhood of 500 single-family
houses? What if your boss provides three
spaces per house?
Refer to Displays 6.1 and 6.3. Your boss
suggests providing two parking spaces for
each of the 500 single-family houses and,
to keep things fair, four spaces for each of
the 500 duplexes. If this plan is followed, in
which neighborhood will a larger proportion
of vehicles have to be parked on the street?
About 10% of high school girls reported
in a survey that they rarely or never wear
seat belts in motor vehicles. Suppose you
randomly sample three high school girls.
Construct the probability distribution of the
random variable, X, defined as the number
of girls in your sample who say they rarely
or never wear seat belts. Begin by listing the
outcomes as in the lung cancer example on
page 362. [Source: www.nhtsa.dot.gov.]
As stated in the example on page 362, 90%
of lung cancer cases are caused by smoking.
Suppose three lung cancer patients are
randomly selected from the large population
of people with that disease. Construct the
probability distribution of X, the number of
patients with cancer caused by smoking.
The report cited in E5 says that 18% of high
school boys report that they rarely or never
wear seat belts. Suppose you randomly
sample two high school boys. Construct
the probability distribution of the random
variable, X, defined as the number of
sampled boys who say they rarely or never
wear seat belts.
According to a recent government report,
73% of drivers now use seat belts regularly.
Suppose a police officer at a road check
randomly stops three cars to check for
seat belt usage. Construct the probability
distribution of X, the number of drivers
using seat belts. [Source: usgovinfo.about.com.]
379
22-03-2009 21:07
Lesson
23 de 24
E9. Refer to the burglary insurance example on

page 369.
a. If you charge an urban householder
$209.50 per year for insurance, what
is the largest profit you could earn on
that one policy for the year? What is the
largest possible loss for the year?
b. If you want to earn an expected yearly
profit of $5000 per 1000 urban customers,
how much should you charge per
customer?
c. Factors besides the location of the
household affect the probability of a
burglary. What other factors might an
insurance company take into account?
E10. The passenger vehicle with the highest theft
loss is the Cadillac Escalade EXT, with 20.2
claims for theft per 1000 insured vehicles
per year and an average payment of $14,939
per claim. How much would you charge an
owner of a Cadillac Escalade EXT for theft
insurance per year if you simply wanted to
expect to break even? [Source: Highway Loss Data
a. Construct the exact probability

distribution for X.
b. Find the expected value of X.
c. What assumptions are necessary for your
answer in part a?
E13. For each million tickets sold, the original
New York Lottery awarded one $50,000
prize, nine $5,000 prizes, ninety $500 prizes,
and nine hundred $50 prizes.
a. What was the expected value of a ticket?
b. The tickets sold for $0.50 each. How
much could the state of New York expect
to earn for every million tickets sold?
c. What percentage of the income from the
lottery was returned in prizes?
E14. A scratch-off card at a fast-food restaurant
chain gave the information contained
in Display 6.21. All cards were potential
winners. In order to actually win the prize,
the player had to scratch off a winning path
without making a false step. The probability
of doing this was given as
for the food
Institute news release, October 19, 2004, www.hwysafety.org.]
item and
for any nonfood item. So, for
example, if you got the card with the prize
E11. The boss of your construction company isnt
satisfied with the simulation in Display 6.3
worth $1 million, you still would have to
on page 360 and asks you to construct the
scratch off a path without making a mistake.
exact probability distribution for selecting
The probability of doing that and actually
two households at random and counting the
winning the $1 milion was
total number of motor vehicles.
a. Use the rules of probability from Chapter 5
and the information in Display 6.1 on page
359 to construct this distribution exactly.
b. How does your exact distribution
compare to the simulated distribution
in Display 6.3?
c. Find the mean value of this exact
distribution and compare it to the
expected value found in P17.
E12. Display 6.18 on page 378 provides
information on the number of color
televisions in U.S. households. Suppose a
duplex is occupied by two randomly selected
households and you are interested in the
Display 6.21 Values of prizes, along with frequencies.
random variable X, defined as the total
number of color televisions in a duplex.
22-03-2009 21:07
Lesson
24 de 24
a. What is the total value of all the potential

prizes?
b. What total amount did the restaurant
expect to pay out?
c. What is the expected value of a card?
E15. You are trying to decide whether to buy
Brand A or Brand B as your new dishwasher.
Each brand is expected to last about 10 years.
Brand A has a price of $950 and comes with
an unlimited number of repairs at $150 each.
Brand B has a higher price, $1200, but comes
with an unlimited number of free repairs.
Which dishwasher you buy depends on the
number of repairs you expect Brand A to
require. You investigate and find out the
information about Brand A in Display 6.22.
Display 6.22 Probability of number of repairs.
a. Find the expected cost of each brand,

including original price and repairs.
b. What would be the advantage of buying
Brand A? Brand B?
E16. Display 6.23 gives information about the
four most commonly used local service plans
for the New York /New Jersey calling area.
Display 6.23
New York cell phone plans. [Source: New

York City Department of Consumer Affairs, The
Buyers Guide to Better Cellular Service, April
2002; www.nyc.gov.]
If you talk for 500 minutes or less each

month, Voicestream will be the cheapest
service. But some months you have a lot to
talk about, and you estimate that the number

of minutes you talk each month follows the
probability distribution in Display 6.24.
Display 6.24
Probability of number of minutes.
Rank the service plans according to how

much you would expect them to cost per
month, knowing that your only calls will
be within the New York /New Jersey calling
area.
E17. Suppose you roll two tetrahedral dice, each
with faces numbered 1, 2, 3, and 4.
a. Make a probability distribution table for
the sum of the numbers on the two dice.
What is the probability that the sum is 3?
b. Make a probability distribution table for
the absolute value of the difference of the
numbers on the two dice. What is the
probability that the absolute value of
the difference is 3?
c. What is the expected value of the
probability distribution in part b?
E18. Construct the probability distribution table
for the situation of flipping a coin five times
and counting the number of heads. Compute
the mean and standard deviation of the
distribution, and construct a probability
distribution graph.
E19. Many mechanical and electronic systems
are built to include backups to major
components (much like the spare tire in
a car). If the main component fails, the
backup kicks in. Often, however, the backup
component is for emergency use only and
is not built to the same specifications as the
main component. Suppose a main pump in
a city water system works without failure
for 1 month with probability 0.1, 2 months
with probability 0.3, and 3 months with
probability 0.6. The pumps backup works
without failure for 1 month with probability
381
22-03-2009 21:07
Lesson
1 de 11
0.2 and for 2 months with probability 0.8.

The main pump is used alone until it fails;
then the backup pump kicks in and is used
alone until it fails. You can assume that
the main pump and its backup operate
independently of each other.
a. Find the probability distribution of the
total working time of the main pump
followed by its backup.
deviation of the total working time of the
main pump and its backup.
c. Show that the expected total time in part
b is the sum of the expected working
times of the main pump and of its
backup.
E20. You pay $1 to play Game A, which generates
a payoff $0, $1, $2, or $3 with respective
probabilities 0.4, 0.3, 0.2, and 0.1. You also
pay $2 to play Game B, which generates
a payoff $0, $2, or $4 with respective
probabilities 0.7, 0.2, and 0.1. The games are
operated independently of each other.
of your total gain from playing the two
games. (Your gain is your winnings minus
the cost of playing the game.)
deviation of your total gain.
c. Show that your expected total gain in part
b is the sum of your expected gains from
each of the two games.
6.2
E21. A regular die is rolled, and then a tetrahedral

die is rolled. The difference is calculated.
of the difference of the number on the
regular die and the number on the
tetrahedral die.
b. Verify that the mean of the probability
distribution is
and the variance
is
, where and are,
respectively, the mean and variance
of the distribution of the roll of a
are,
single six-sided die and and
respectively, the mean and variance of
the distribution of the roll of a single
four-sided die.
E22. A single die is rolled and the result used
twice, once as the first number and again
as the second number.
a. Construct the probability distribution of
the difference of the first number and the
second number.
b. Find the mean and variance of this
probability distribution using the
standard formulas
c. Is the mean of the differences equal to

?
d. Is the variance equal to
?
The Binomial Distribution

In Section 6.1, you used both theory and simulation to construct probability
distributions of random variables. You might have found this to be a lot of
work, even for relatively simple situations, and might have wondered if there
was a simpler way to do it. Well, good news is coming. For some of the most
commonly used random variables, mathematical formulas will give you the
distribution directly. These types of situations have a common structure that
makes a mathematical generalization possible. How can you find these magic
formulas? In this section, Jack and Jill come to the rescue and show you how
its done.
22-03-2009 21:08
Lesson
2 de 11
Binomial Probabilities
Many random variables seen in practice amount to counting the number of
successes in n independent trials, such as
the number of doubles in four rolls of a pair of dice (on each roll either you
get doubles or you dont),
the number of patients with type A blood in a random sample of ten patients
(a person either has type A blood or doesnt)
the number of defective items in a sample of 20 items (either an item is
defective or it isnt)
These situations are called binomial because each trial has two possible outcomes.
In Chapter 5, Jack and Jill were asking questions about a binomial situation
when they wanted to know the answer to a question like this: If we ask four
people which is the tap water and none of them can tell bottled water from tap
water, what is the probability that two of the four will guess correctly?
Jack: Thats exactly what we were doinglooking at n people and
counting the number who correctly identified the tap water.
Jill: Do they mean there is an easier way than listing all those outcomes?
Jack: It sounds like it. But I bet we can figure it out for ourselves now
that we know there is a formula. We presented each of four people
with bottled and tap water and asked them to identify the tap
water. We got a list of 16 possible outcomes, all equally likely if
none of the four can tell the difference and are merely guessing.
Lets call the outcomes C for correctly identifying the tap water
and I for incorrectly identifying it.
Jill: Yeah, look. Ive organized them in this table (Display 6.25). Lets
try to figure out the probability of getting exactly two people who
select the tap water.
Display 6.25 Jills table.
Jack: Using the Multiplication Rule for Independent Events, we know

that the probability of getting a particular outcome when two
, or . There are six
people guess correctly, say, CIIC,
outcomes where two people guess correctly. So the probability of
getting two people who correctly select tap water is . But I wish
we could find the numbers in the last column without listing all
the possible outcomes!
383
22-03-2009 21:08
Lesson
3 de 11
Jill: Ive seen those numbers beforein the fifth row of Pascals triangle.
Jack: Thats it! The number of ways we can choose exactly two people
out of four people to identify the tap water is
Pascals triangle
D13. Explain why each of the 16 outcomes in Display 6.25 has probability
Example: College Graduates

The proportion of adults age 25 and older in the United States with at least a
bachelors degree is 0.27. [Source: U.S. Census Bureau, Statistical Abstract of the United States, 2006,
Table 216.] Suppose you pick seven adults at random. What is the probability that
exactly three will have a bachelors degree or higher?
Solution
Jack: Now what do we do? Theyve changed the problem so that the
outcomes arent equally likely. That is so like them.
Jill: There is nothing about our method that requires equally likely
outcomesall that matters is independence and that the probability
stays the same. For example, one outcome with three grads is
grad
not
grad
grad
not
not
not
By the Multiplication Rule, the probability of this particular

outcome is
(0.27)(0.73)(0.27)(0.27)(0.73)(0.73)(0.73) = (0.27)3(0.73)4
Another outcome with three grads is
not
grad
grad
not
not
not
grad
The probability of this particular outcome is

(0.73)(0.27)(0.27)(0.73)(0.73)(0.73)(0.27) = (0.27)3(0.73)4
which is exactly the same.
Jack: Thats because the probabilities of the outcomes with exactly three
grads all have the same factors but in a different order.
Jill: And there are 35 of them because
Its a good thing we didnt have to list all of them!
22-03-2009 21:08
Lesson
4 de 11
Jack: So the probability of getting exactly three college grads is

(number of ways to get 3 grads) (probability of each way)
or
Jill: Yeah! Now we can do any problem they throw at us.

A summary of Jack and Jills method is given next.
The Binomial Probability Distribution
Suppose you have a series of trials that satisfy these conditions:
B: They are binomialeach trial must have one of two di erent outcomes,
one called a success and the other a failure.
I: Each trial is independent of the others; that is, the probability of a
success doesnt change depending on what has happened before.
Notation: p is the
probability of a
success on any one
trial, and n is the
number of trials.
N: There is a fixed number, n, of trials.

S: The probability, p, of a success is the same on each trial, with 0 < p < 1.
Then the distribution of the random variable X that counts the number of
successes is called a binomial distribution. Further, the probability that you
get exactly X = k successes is
Reminder:
where
is the binomial coefficient, calculated as
where n! = n(n 1)(n 2). . . 3 2 1.

You can use a calculator to calculate binomial probabilities quickly. For
example, the probability in the previous examplethat exactly three of seven
randomly chosen adults will have at least a bachelors degreeis shown here. [See
Calculator Note 6B to learn more about this calculator function.]
You can also calculate cumulative probabilities using a calculator. For

example, the probability that fewer than three (that is, zero, one, or two) of seven
385
22-03-2009 21:08
Lesson
5 de 11
adults will have a bachelors degree is shown here. [See Calculator Note 6C to learn
how to perform this operation.]
It is generally safe
to regard trials as
independent for all
practical purposes if the
condition n < 0.10N is
satisfied.
You might have noticed that the trials in Jack and Jills example arent
really independent. The first adult they selected who was age 25 or older has
probability 0.27 of being a college graduate. If that person is a college graduate,
the probability that the next person selected is a college graduate is a bit less.
However, the change in probability is so small that Jack and Jill can safely ignore
it. If there are 150,000,000 adults age 25 and older in the United States, then there
would be 40,500,000 college graduates. The probability that the first adult selected
, or 0.27. If that person turns out to be a college
is a college graduate is
graduate, the probability that the second adult selected is a college graduate
, or 0.2699999951, which is very close to 0.27.
is
You can treat your random sample as a binomial situation as long as the
sample size, n, is small compared to the population size, N. A simple guideline
that works well in practice is that n should be less than 10% of the size of the
population, or n < 0.10N.
In a famous
episode of I Love
Lucy, Lucille Ball
takes her passion
for chocolate to
comic extremes.
D14. Make a probability distribution table for the number of college graduates in
a random sample of seven adults if 27% of adults are college graduates.
D15. Make a probability distribution table for the number of people who
arent college graduates. Make a histogram of this distribution and of the
distribution in D14, and compare the two.
D16. Show why the population must be large in order to use the binomial
probability formula by pretending that the total adult population of the
United States is 12, of which 4 are chocoholics. Compute the exact
probability that in a random sample of 7 adults, none will be chocoholics.
Compare your results to those you get using the binomial formula.
D17. From a class of 20 students, half of whom are seniors, the teacher selects
10 students at random to check for completion of todays paper. If X denotes
the number of papers belonging to seniors among those checked, will X have
a binomial distribution? Explain why or why not.
In Activity 6.2a, youll conduct the tap water vs. bottled water experiment
yourself. This activity will give you practice in statistical decision-making as you
decide how many subjects will have to correctly select the tap water before you are
convinced that people can tell the difference.
22-03-2009 21:08
Lesson
6 de 11
Can People Identify the Tap Water?

What youll need: small paper or plastic cups, a
container of tap water, a container of bottled water,
about 20 volunteer subjects
1. Design a study to see whether people can
identify which of two cups of water contains
the tap water. Refer to the principles of good
experimental design in Chapter 4.
2. Suppose none of your 20 subjects actually
can tell the difference, so their choice is
equivalent to selecting one of the cups at
random. Construct a probability distribution
of the number who choose the tap water.
3. If only ten people correctly select the tap water, you have no reason to
conclude that people actually can tell the difference. Why?
4. Decide how many people will have to make the correct selection before
you are reasonably convinced that people can tell the difference.
5. Conduct your study and write a conclusion.
Shape, Center, and Spread of a Binomial Distribution

The fact that the binomial distribution is represented by such a simple and elegant
formula might lead you to believe that there is a simple formula for the expected
value and standard deviation as well. Jack and Jill to the rescue again! For Jack and
Jills tap water tasting experiment, the distribution of X, the number choosing tap
water out of four tasters, is shown in Display 6.26.
Display 6.26
Distribution of number of correct guesses.
The expected value for this distribution is

E(X) = 0(1/16) + 1(4/16) + 2(6/16) + 3(4/16) + 4(1/16) = 2
Notice that E(X) can be written as 4 1/2, which is np. In other words, you would
expect about half of the four to guess correctly given two equally likely choices.
Is this pattern just coincidence, or is it a general rule? Lets try one more
example to see if we have any hope of the latter. The discussion of college
387
22-03-2009 21:08
Lesson
7 de 11
graduates on page 384 stated that about 27% of U.S. adults have at least a
bachelors degree. If three adults are randomly sampled in an opinion poll, the
distribution of X, the number with at least a bachelors degree, will be as shown
in Display 6.27.
Display 6.27
Distribution of number of college graduates.
The expected value of this distribution is

E(X) = 0(0.39) + 1(0.43) + 2(0.16) + 3(0.02) = 0.81 = 3(0.27) = np
A general rule does seem to be a possibility, and that, indeed, is the case. As
a bonus, there is a general rule for the standard deviation that turns out to be just
about as simple as the one for the expected value.
The Characteristics of a Binomial Distribution
For a random variable X having a binomial distribution with n trials and
probability of success p, the mean (expected value) and standard deviation for
the distribution are given by
These formulas
hold for all sample
sizes n.
E(X) = X = np
and
The shape of the distribution becomes more normal as n increases.
Example: College Graduates

About 27% of U.S. adults have at least a bachelors degree. If you select 100 adults
at random from all adults in the United States, how many do you expect to have a
bachelors degree and with what standard deviation?
Solution
You expect to get E(X) = 100(0.27), or about 27 college graduates, with a
standard deviation of
, or 4.44.
Plots of binomial distributions for various values of n and p are shown in
Display 6.28.
22-03-2009 21:08
Lesson
8 de 11
Display 6.28 Binomial distributions for different values of n and p.
You can use a calculator to quickly graph and compare binomial distributions.
[See Calculator Note 6D to learn how.] The binomial distribution for n = 5 and
p = 0.7 is shown here. In D18 youll explore how binomial distributions change
for increased values of p or n.
Shape, Center, and Spread of a Binomial Distribution

D18. How do the shape, center, and spread of a binomial distribution change as p
increases for fixed values of n? How do they change as n increases for fixed
values of p?
D19. Ten CD players of a discontinued model are selling for $100 each, with a
double your money back guarantee if they fail in the first month of use.
Suppose the probability of such failure is 0.08. What is the expected net gain
for the retailer after all ten CD players are sold? Ignore the original cost of
the CD players to the retailer.
Summary 6.2: The Binomial Distribution

The random variable X is said to have a binomial distribution if X represents the
number of successes in n independent trials, where the probability of a success
is p on each trial. For example, X might represent the number of successes in a
random sample of size n from a large population, with probability of success p on
each selection.
389
22-03-2009 21:08
Lesson
9 de 11
A binomial distribution has these important characteristics:

The probability of getting exactly X = k successes is given by the formula
The expected value or mean is E(X) = X = np.

Practice
P19. Suppose Jack and Jill ask six people who
cant tell the difference between tap water
and bottled water to identify the tap water.
Use their method to make a probability
distribution table for six people. Then make
a graph of the distribution.
P20. Suppose you flip a coin eight times. What
is the probability that youll get exactly
3 heads? Exactly 25% heads? At least
7 heads?
P21. Suppose you roll a balanced die seven times.
What is the probability that you will get an
even number exactly two times? More than
half the time?
P22. About 8.8% of people ages 1424 are
dropouts, persons who are not in regular
school and who have not completed the
12th grade or received a general equivalency
degree. Suppose you pick five people at
random from this age group. [Source: U.S.
Census Bureau, Statistical Abstract of the United States, 2006,
Table 255.]
a. What is the probability that none of the

five are dropouts?
b. What is the probability that at least one is
a dropout?
c. Make a probability distribution table for
this situation.
P23. Describe how to use simulation to construct
an approximate binomial distribution for the
situation in P22. Conduct two trials of your
simulation, using these random digits:
89254 99538 18315 45716 36270 79665

49830 06226 88863 02322 36630 07176
P24. According to a recent government report,
73% of drivers now use seat belts regularly.
Suppose a police officer at a road check
randomly stops four cars to check for seat
belt usage. Find the probability distribution
of X, the number of drivers using seat belts.
[Source: National Highway Traffi c Safety Administration.]
Shape, Center, and Spread of a

Binomial Distribution
P25. The median annual household income for
U.S. households is about $44,400. [Source: U.S.
Census Bureau, www.census.gov.]
a. Among five randomly selected U.S.

households, find the probability that four
or more have incomes exceeding $44,400
per year.
b. Consider a random sample of 16 U.S.
households.
i. What is the expected number of
households with annual incomes
under $44,400?
ii. What is the standard deviation of the
number of households with annual
incomes under $44,400?
iii. What is the probability of getting at
least 10 out of the 16 households with
annual incomes under $44,400?
c. In a sample of 16 U.S. households,
suppose none had annual incomes under
$44,400. What might you suspect about
this sample?
22-03-2009 21:08
Lesson
10 de 11
P26. You neglected to study for a truefalse quiz

on the government of Botswana, so you
will have to guess on all the questions. You
need to have at least 60% of the answers
correct in order to pass. Would you rather

have a 5-question quiz or a 20-question
quiz? Explain your reasoning, referring to
Display 6.28.
Exercises
E23. If you roll a pair of dice five times, find the
probability of each outcome.
a. You get doubles exactly once.
b. You get exactly three sums of 7.
c. You get at least one sum of 7.
d. You get at most one sum of 7.
E24. Suppose you select five numbers at random
from 10 through 99, with repeats allowed.
Find the probability of each outcome.
a. Exactly three of the numbers are even.
b. Exactly one of the numbers has digits
that sum to a number greater than or
equal to 9.
c. At least one of the numbers has digits
that sum to a number greater than or
equal to 9.
E25. According to a recent Census Bureau
report, 37 million Americans, or 12.7%
of the population, live below the poverty
level. Suppose these figures hold true for
the region in which you live. You plan to
randomly sample 25 Americans from your
region. [Source: Current Population Survey, 2005 Annual
Social and Economic Supplement, www.census.gov.]
a. What is the probability that your sample

will include at least two people with
incomes below the poverty level?
b. What are the expected value and standard
deviation of the number of people in your
sample with incomes below the poverty
level?
E26. According to the U.S. Census Bureau, about
16% of residents have no health insurance.
You are to randomly sample 20 residents
for a survey on health insurance coverage.
[Source: U.S. Census Bureau, Current Population Survey,
March 2004.]
a. What is the probability that your sample

will include at least three people who do
not have health insurance?
b. What are the expected value and standard
deviation of the number of people in your
sample without health insurance?
E27. You buy 15 lottery tickets for $1 each.
With each ticket, you have a 0.06 chance of
winning $10. Taking into account the cost of
the tickets,
a. what is your expected gain (or loss) on
this purchase?
b. what is the probability that you will gain
$10 or more?
c. what is the standard deviation of your
gain?
E28. An oil exploration firm is to drill ten wells,
each in a different location. Each well has
a probability of 0.1 of producing oil. It will
cost the firm $60,000 to drill each well.
A successful well will bring in oil worth
$1 million. Taking into account the cost
of drilling,
a. what is the firms expected gain from the
ten wells?
b. what is the standard deviation of the
firms gain for the ten wells?
c. what is the probability that the firm will
lose money on the ten wells?
d. what is the probability that the firm will
gain $1.5 million or more from the ten
wells?
E29. A home alarm system has one detector for
each of the n zones of the house. Suppose the
probability is 0.7 that the detector sounds
an alarm when an intruder passes through
its zone and that this probability is the
391
22-03-2009 21:08
Lesson
11 de 11
same for each detector. The alarms operate

independently. An intruder enters the house
and passes through all the zones.
a. What is the probability that at least one
alarm sounds if n = 3?
alarm sounds if n = 6? Is the probability
from part a doubled?
c. How large must n be in order for the
probability of at least one alarm sounding
to be about 0.99?
E30. Complex electronic systems for which failure
could be catastrophic (such as airplane
electronic systems) are constructed so that
key components have a number of parallel
backup components that automatically take
over in case of failure. Suppose that at a
certain key juncture there are n electronic
switches (the main one and n 1 backups)
that operate independently of one another
under standard conditions. The probability
that any one switch works properly is 0.92.
a. What is the probability that at least one
switch works properly if n = 2?
switch works properly if n = 3?
c. An engineer wants the probability that
at least one switch works properly to be
1 105, or 0.99999. How many backup
switches should she place in the system?
E31. A potential buyer will sample DVDs from a
large lot of new DVDs. If she finds at least
one defective DVD, shell reject the entire
lot. In each case, find the sample size n for
which the probability of detecting at least
one defective DVD is 0.50.
a. Ten percent of the DVDs are defective.
b. Four percent of the DVDs are defective.
E32. Scientists often study the conditions of
populations of fish (and other animals) by
tagging some and then recapturing a few
tagged ones, along with some untagged
ones, at a later date. Suppose a certain
percentage of a large population of fish are
tagged and then a random sample of n fish
is to be selected from that population. Find

n such that the probability of recapturing
at least one tagged fish is 0.80 if it is
estimated that
a. 5% of the population was tagged
b. 2% of the population was tagged
E33. For a random sample of three adults, the

expected number of college graduates
is (3)(0.27), or 0.81, less than one whole
person. Provide a meaningful interpretation
of this number.
E34. The sum of the probabilities in a binomial
distribution must equal 1. To show that
this is true, first let q = 1 p so that
p + q = 1. The table shows that for a
binomial distribution with n = 1 trial, the
sum of the probabilities is 1.
a. Show that the sum of the probabilities

when n = 2 is p2 + 2pq + q2. Factor this
sum to show that it is equal to 1.
b. Find the sum of the probabilities when
n = 3. Factor this sum to show that it is
equal to 1.
c. Show that the terms of the expansion
of (p + q)n are the probabilities for
k = 0, 1, 2, . . . , n in the binomial
distribution with n trials and probability
of success p.
22-03-2009 21:08
Lesson
1 de 9
6.3
In waiting-time
problems, you count the
number of trials it takes
to get the first success.
The Geometric Distribution

Your favorite sports team has made the finals. You have a good estimate of the
probability that it will win any one game against its opponent. In the previous
section, you answered questions of the type What is the probability that the
team will win exactly three out of seven games?The number of trials was fixed.
In this section, you will answer questions like What is the probability that
your team will get its first win in the third game? Problems like this are called
waiting-time problems because the variable in question is the number of trials
you have to wait before the event of interest happens. The number of trials isnt
fixedyou simply count the number of trials until you get the first success.
The goal of Activity 6.3a is to design a simulation for a waiting-time situation
and use it to construct an approximate probability distribution.
Waiting for Type A Blood

What youll need: a device for generating random outcomes
In the general population, about 40% of people have
type A blood. [Source: American Red Cross, September 2006,
www.givelife2.org.] A technician in a blood bank needs
type A blood today and wants to know about how
many randomly selected blood donations he might
have to process to find the first donation that is type A.
1. Describe how to use random digits to simulate the
blood type of the first donation he checks.
2. Simulate the outcome for the first donation he
checks. Was it type A blood? In other words, was
his waiting time to success just one trial?
3. If he was successful on the first donation, then stop this run of the
simulation. If he did not have success on the first trial, then continue
generating outcomes until he gets his first success. Record the number
of the trial on which the first success occurred.
4. Repeat steps 2 and 3 at least ten times, each time recording the number of
trials needed.
5. Combine your results with others from the class and construct a plot to
represent the simulated distribution of the number of donations that must
be checked. Describe the shape of this distribution, and find its mean.
6. What is your estimated probability that
a. the first donation with type A blood is the second one he checks?
b. he will have to check five or fewer donations to find the first that is
type A?
c. he has to check at least two donations to get one that is type A?
d. he has to check at most four donations to get one that is type A?
7. On which trial is he most likely to find the first donation that is type A?
393
22-03-2009 21:08
Lesson
2 de 9
The probability distribution you constructed in the activity is called a

geometric distribution. Display 6.29 shows geometric distributions for p equal
to 0.1, 0.3, 0.5, and 0.8, where p is the probability of success on each trial. For
example, when p = 0.3, the probability of getting a success on the first trial is 0.3,
as the height of the first bar indicates. The probability of getting the first success
on the second trial is 0.21, the height of the second bar. In this case, you must fail
on the first try and then succeed on the second, and this probability is
P( fail on first try) P(succeed on second try) = (0.7)(0.3) = 0.21
Note that in each graph, each bar is the same fixed proportion of the height of
the bar on its left.
Display 6.29 Geometric distributions for different probabilities of

a success.
[See Calculator Note 6E to learn how to graph geometric distributions.]
Waiting-Time Situations
D20. Refer to the graphs in Display 6.29.
a. What is the shortest possible waiting time? What is the longest? What is
the most likely waiting time? Why is this the case?
b. What are the exact heights of the first and second bars in the graph for
p = 0.8? Explain.
c. How are the shapes of the four distributions in Display 6.29 similar? How
are they different?
d. What real-life situation could the graph for p = 0.5 represent?
22-03-2009 21:08
Lesson
3 de 9
The Formula for a Geometric Distribution

As you might already have discovered, you can derive a formula for the probability
distribution of a geometric (waiting-time) random variable quite easily.
Jack: Yep, we can do it. This is easier than our problem from Section 6.2
about the binomial distribution.
Jill: It sure is. Lets do it for the blood bank example, where about 40%
of people have type A blood. There the probability that the first
donation checked is type A is 0.4.
Jack: Right. The probability that the waiting time, X, is 1 is P(X = 1) = 0.4.
Jill: Now for P(X = 2). For the second donation to be the first that is
type A . . .
Jack: What kind of nonsense is that? For the second donation to be
the first . . . ?
Jill: Sorry, let me say it with more words. Suppose the first donation
checked isnt type A and the second donation is. Then we have our
first success with the second donation.
Jack: Thats better. The probability of this sequence of outcomes is
P( first donation isnt type A and second is type A) = (0.6)(0.4),
or 0.24.
Always check conditions!
Jill: But you get to multiply like that only if the events are independent.
What if a whole family had donated blood? Then their blood types
might not be independent.
Jack: Yeah, we will have to be careful about things like that. The
probability cant change depending on who else has come in.
Jill: If we can assume independence, the probability that the
third donation checked will be the first that is type A is
P(X = 3) = (0.6)(0.6)(0.4).
Jack: Because we have to have two failures and then our first success.
Jill: This could go on forever. So we better get started and make a table.
Display 6.30 Jills table.
Jack: All this work is making me thirsty. Five rows is enough. Lets go up
the hill and get some water.
395
22-03-2009 21:08
Lesson
4 de 9
Display 6.31 summarizes Jack and Jills reasoning, this time for a general
geometric distribution with probability of success p and probability of failure
q = 1 p.
Display 6.31 Table for the general geometric distribution.
The Geometric Probability Distribution

Suppose you have a series of trials that satisfy these conditions:
They are binomialthat is, each trial must have one of two different
outcomes, one called a success and the other a failure.
Each trial is independent of the others; that is, the probability of a
success doesnt change depending on what has happened before.
The trials continue until the first success.
The probability, p, of a success is the same on each trial, 0 < p < 1.
Then the distribution of the random variable X that counts the number of
trials needed until the first success is called a geometric distribution. The
probability that the first success occurs on the X = kth trial is
P(X = k) = (1 p)
k1
for k 1, 2, 3, . . . .
Both the binomial and geometric random variables start with a sequence
of independent trials with two outcomes and constant probability of success.
The binomial counts the number of successes in a fixed number of trials; the
geometric counts the number of the trial on which the first success occurs. [See
Calculator Notes 6Fand 6G to learn how to calculate geometric probabilities and
cumulative probabilities of a geometric distribution.]

D21. About 10% of the U.S. population has type B blood. [Source: American Red Cross,
September 2006, www.givelife2.org.] Suppose our technician is checking donations
that may be considered independent with respect to blood type.
22-03-2009 21:08
Lesson
5 de 9
a. What is the probability that the first donation that is type B is the third
one checked?
b. What is the probability that he will have to check at most three donations
to get the first that is type B? At least three donations?
c. Suppose that the technician is looking for a type B donation and has
already tested three donations without success. What is the probability
that he has to test at least three more donations to find one that is type B?
d. Part c refers to the memoryless property of the geometric distribution.
Explain why that term is appropriate.
D22. In Activity 3.5a, Copper Flippers, you started by flipping 200 pennies. You
then removed the ones that died (landed tails) and continued flipping with
a decreasing number of live copper flippers. Make a plot with the number
of the flip on the horizontal axis and the theoretical number of copper
flippers left after that flip on the vertical axis. Compare this plot with your
results from Activity 3.5a if you still have them.
D23. Give at least two ways to find the probability that it will take three or more
rolls of a pair of dice to get doubles.

In Activity 6.3b, you will discover the formula for the mean (expected value) of a
geometric distribution.
The Expected Value of a Geometric Distribution

Divide your class into nine groups. Each group will be assigned a different
value of p, for p = 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, and 0.9.
1. Make a table that shows the probabilities of the values of X from 1 through
30 for the geometric distribution with your groups assigned value of p.
2. Use your table to compute an approximation of the expected value of this
distribution.
3. Will your approximation in step 2 be a bit larger or a bit smaller than the
theoretical expected value? Explain.
4. Using the results from all nine groups, make a scatterplot of the expected
values plotted against the values of p. What is the shape of this plot?
5. Now make a scatterplot of the expected values plotted against the
reciprocals of the values of p. State a hypothesis about the formula for the
expected value of a geometric distribution.
Like the formulas for the binomial distribution, those for the expected value
and standard deviation of the geometric distribution turn out to be surprisingly
simple. Although using the mean as the measure of center and the standard
deviation as the measure of spread seemed complicated in Chapter 2, it pays
off now.
397
22-03-2009 21:08
Lesson
6 de 9
Characteristics of a Geometric Distribution

A random variable X that has a geometric distribution with probability of
success p has an expected value (mean) and standard deviation of
and
The shape of the distribution is skewed right, with bars decreasing
proportionally.
You can find a proof of the formula for the expected value in E42. The proof
of the formula for the standard deviation is a bit more involved.
Example: Mean and Standard Deviation for Blood Donations

In the scenario described in Activity 6.3a, find the expected value and standard
deviation of the number of blood donations the technician would have to check
in order to get the first donation with type A blood.
Solution
The probability that a random blood donation is type A is 0.4. Thus, the expected
number of blood donations that the technician would have to check in order to
get the first one that is type A is
with standard deviation
The expected value is

the expected number of
trials needed to get the
first success.
Note that the expected value is the expected number of donations that must
be checked to get one that is type A. It is not the expected number checked
before getting one that is type A. In this case, the expected number includes
1.5 donations that arent type A and one donation that is.
Example: The Expected Number for the Second Success

In the scenario described in Activity 6.3a, find the expected value of the number
of blood donations the technician would have to check in order to get two that are
type A blood.
Solution
The expected number of blood donations that must be checked to obtain two
of type A is 2(2.5), or 5, because the technician must check, on average, 2.5
donations to get the first success. Then the process starts again to get the second
success, again checking an average of 2.5 donations, or five in all.
22-03-2009 21:08
Lesson
7 de 9

D24. The phone line for a 24-hour ticket office is busy about 70% of the time.
a. If you dial at random times throughout the week, what is the expected
number of tries it will take you to get through? What is the standard
deviation of the number of tries?
b. What is the expected total number of times you will have to dial if you
forget to ask a question after getting through the first time and have to
call again?
c. Suppose you have been trying for 3 days, so your friend starts dialing as
wellat random timeswhile you continue trying. Who do you expect
to get through first, you or your friend?
D25. If p doubles, how does the expected value of a geometric random variable
change?
Summary 6.3: The Geometric Distribution

For a sequence of independent trials in which p, the probability of success, stays
the same for each trial, the random variable X that counts the number of trials
until the first success occurs has a geometric distribution.
The probabilities for this distribution are given by the formula
P(X = k) = (1 p)k1p
for k = 1, 2, 3, . . . and 0 < p < 1.
The mean (expected value) of the distribution is
E(X) =
The expected number of trials until the nth success is n .

Practice
Waiting-Time Situations
P27. Suppose you are rolling a pair of dice and P28. About 85% of Americans over age 25
waiting for a sum of 7 to occur.
have graduated from high school. You are
randomly sampling and interviewing adults
a. What is the probability that you get a
one at a time for an opinion poll that applies
sum of 7 for the first time on your first
only to high school graduates. [Source: U.S. Census
roll? On your second roll?
Bureau, Statistical Abstract of the United States, 2006.]
b. Using the graphs in Display 6.29 as a
a. What is the probability that you get your
guide, sketch an approximate graph of the
first high school graduate on the third
probability distribution of this situation.
person you interview?
399
22-03-2009 21:08
Lesson
8 de 9
b. What is the probability that you get your

first high school graduate sometime after
the second person you interview?
c. Sketch an approximate distribution of the
number of the interview on which you
get the first high school graduate.
P29. Suppose 9% of the engines manufactured
on a certain assembly line have at least one
defect. Engines are randomly sampled from
this line one at a time and tested. What is the
probability that the first nondefective engine
is found
a. on the third trial?
b. before the fourth trial?
P30. About 70% of the time, the telephone lines

coming into a concert ticket agency are all
busy. Suppose you are calling this agency.
a. What is the probability that it takes you
only one try to get through?
b. What is the probability that it takes you
two tries?
c. What is the probability that it takes you
four tries?
d. What assumption are you making in

computing these probabilities?
P31. The probability that a random blood
donation is type B is 0.1.
a. What is the expected number of
donations that must be checked to obtain
the first that is type B?
b. What is the standard deviation of the
number of donations that must be
checked to obtain the first of type B?
c. What is the expected number of
donations that must be checked to
obtain two that are type B? To obtain
three that are type B?
P32. About 85% of Americans over age 25
have graduated from high school. You are
randomly sampling and interviewing adults
one at a time for an opinion poll that applies
only to high school graduates.
a. What is the expected number of
interviews you have to conduct to get
the first high school graduate?
number of interviews you have to
conduct to get the first high school
graduate?
c. What is the expected number of
interviews you have to conduct to get
the first ten high school graduates?
d. What is the standard deviation of the
number of interviews you have to
conduct to get the first ten high school
graduates?
Exercises
E35. You are participating in a question bee in
history class. You remain in the game until
you give your first incorrect answer. The
questions are all multiple choice, each with
four possible answers exactly one of which is
correct. Unfortunately, you have not studied
for this bee, and you simply guess randomly
on each question.
a. What is the probability that you are still

in the bee after the first round?
b. What is the probability that you are still
in the bee after the third round?
c. What is the expected number of rounds
you will be in the bee?
d. If an entire class of 32 is simply guessing
on each question, how many students
22-03-2009 21:08
Lesson
9 de 9
do you expect to be left after the third

round?
E36. A certain type of insect has a 0.2 chance of
dying on any one day and this probability
remains relatively constant for a few days.
a. What is the probability that an insect of
this type survives its first day?
b. What is the probability that an insect of
this type survives its first three days?
c. What is the probability that an insect of
this type dies on its fourth day?
d. How many days do you expect an insect
of this type to survive?
e. If 200 insects of this type are being
studied in a laboratory, how many do you
expect to be alive after the third day?
E37. Suppose 12% of the engines manufactured
on a certain assembly line have at least one
defect. Engines are randomly sampled from
this line one at a time and tested.
a. What is the expected number of engines
that need to be tested to find the first
engine withouta defect? The third?
number of engines that need to be tested
to find the first engine without a defect?
c. If it costs $100 to test one engine, what
are the expected value and standard
deviation of the cost of inspection up to
and including the first engine without a
defect?
d. Will the cost of inspection often exceed
$200? Explain.
E38. An oil exploration firm is to drill wells at
a particular site until it finds one that will
produce oil. Each well has probability 0.1 of
producing oil. It costs $60,000 to drill a well.
a. What is the expected number of wells to
be drilled?
number of wells to be drilled?
c. What are the expected value and standard
deviation of the cost of drilling to get the
first successful well?
d. What is the probability that it will take at

least five tries to get the first successful
well? At least 15 tries?
E39. In a certain company, an employee must
take an aptitude test before being promoted
to the next job level, and the test cannot
be taken until the employee has 6 months
of experience in the company. Aptitude
tests, however, can be designed to have an
approximate probability, p, that any one
employee with 6 months or more experience
will pass. What value of p should be the
target of the test designer if the goal is
to have 80% of the employees pass in no
more than two tries? What assumption is
necessary for your answer to be correct?
E40. An engineer is designing a machine to
find tiny cracks in airplane wings that are
difficult to detect. The specifications state
that a section of wing known to have these
cracks may be tested by the machine up to
five times, but the probability of locating
the cracks in five trials or fewer must be at
least 0.99. In order to meet this specification,
what must be the probability of locating
cracks on any one trial?
E41. A geometric sequence is a sequence of the
form
a, ar, ar2, ar3, ar4, . . .
where a is the first term and r is the
common ratio.
a. Explain why the probability distribution
developed in this section is called the
geometric distribution.
b. The sum of the probabilities in a
geometric distribution form an infinite
series:
p + pq + pq2 + pq3 + pq4 +
(See Display 6.31). Use the next formula,
for the sum of an infinite geometric
series, to prove that the sum of the
probabilities of a geometric distribution is
equal to 1 as long as | r | < 1:
a + ar + ar2 + ar3 + . . . =
401
22-03-2009 21:08
Lesson
1 de 6
E42. In this exercise, you will use the last formula

in E41 twice to prove that the expected
value of the geometric distribution is .
a. Use the formula for the expected value
of a probability distribution to write the
formula for the mean of a geometric
distribution with probability of success p.
This will be an infinite series.
b. Show that you can write the series from
part a in this form:
E(X) = p (1 + q + q2 + q3 + . . .
+ q + q2 + q3 + . . .
+ q2 + q3 + . . .
+ q3 + . . .
+ )
c. Ignore p for now and sum each row using
the last formula in E41.
d. Find the sum of all the rows, again using
the last formula in E41.
Chapter Summary
In this chapter, you have learned that random variables are variables with
a probability attached to each possible numerical outcome. Probability
distributions, like data distributions, are characterized by their shape, center,
and spread. You can use similar formulas for the mean and standard deviation.
The binomial distribution is a model for the situation in which you count
the number of successes in a random sample of size n from a large population.
A typical question is If you perform 20 trials, what is the probability of getting
exactly 6 successes?
The geometric distribution is a model for the situation in which you count the
number of trials needed to get your first success. A typical question is What is
the probability that it will take you exactly five trials to get the first success?
Review Exercises
E43. Suppose you roll a dodecahedral (12-sided)
die twice. Your summary statistic will be the
sum of the two rolls.
a. What is the probability that the sum is 3
or less?
b. Compute the mean and standard
deviation of the distribution of outcomes
of a single roll of one die.
c. Compute the mean and standard
deviation of the probability distribution
of the sum of two rolls.
E44. Suppose you and your partner each roll two
dice. Each of you computes the average of
your two rolls. The summary statistic is the
difference between your average and your
partners average. Describe the sampling
distribution of these differences.
E45. Two different pumping systems on levees

have pumps numbered 1, 2, 3 and 4,
configured as in Display 6.32. In System
I, water will flow from A to B only if both
pumps are working. In System II, water
will flow from A to B if either pump is
working. Assume that each pump has
this lifelength distribution: 1 month with
probability 0.1, 2 months with probability
0.3, and 3 months with probability 0.6.
Lifelength refers to the length of time the
pump will work continuously without
repair. Assume that the pumps operate
independently of each other. (You are
interested only in whether water will flow,
not the amount of water flowing.)
22-03-2009 21:09
Lesson
2 de 6
Display 6.32 Two pumping systems.
a. Find the distribution of the length of time

water will flow for System I.
b. Find the distribution of the length of time
water will flow for System II.
c. Find the expected time that water will
flow for each system and compare. Which
system would you recommend?
E46. Display 6.33 gives the chances that an adult
(age 18 or older) living in the United States
will become a victim of the given events.
Display 6.33 Rates of insurable causes of accidents

or death per 1000 adults per year.
[Source: U.S. Bureau of Justice Statistics, 2001,
ag.ca.gov.]
E47. A blood bank knows that only about 10% of

its regular donors have type B blood.
a. The technician will check ten donations
today. What is the chance that at least one
will be type B? What assumptions are you
making?
b. The technician will check 100 donations
this month. What is the approximate
probability that at least 10% of them will
be type B?
c. This bank needs 16 type B donations.
If the technician checks 100 donations,
does the blood bank have a good chance
of getting the amount of type B blood it
needs? What recommendation would you
have for the blood bank managers?
d. What is the probability that the technician
will have to check at least four donations
before getting the first that is type B?
E48. One in ten high school graduates in the
state of Florida sends an application to the
University of Florida, says the director of
admissions there. If Florida has approximately
120,000 high school graduates next year, what
are the mean and standard deviation of the
number of applications the university will
receive from Florida high school students?
Do you see any possible weaknesses in using
the binomial model here?
a. What should a randomly selected U.S.

adult be charged yearly for a $100,000
life insurance policy if the insurance
company expects to break even?
b. What should a randomly selected U.S.
adult be charged yearly for a policy that
pays him or her $10,000 if he or she is the
victim of a violent crime?
c. An insurance company determines that
people are willing to pay $50 a year for
a policy that pays them a set amount if
they are injured by fire. What should this
set amount be if the insurance company
expects to break even?
Chapter Summary
403
22-03-2009 21:09
Lesson
3 de 6
E49. Girls Get Higher Grades is a headline

from USA Today. Eighty percent of the girls
surveyed said it was important to them to
do their best in all classes, whereas only 65%
of boys responded this way. Suppose you
take another random sample of 20 female
students and another sample of 20 male
students in your region of the country.
[Source: USA Today, August 12, 1998.]
a. What is the probability that more

than half the girls in the sample want
to do their best in every class? What
assumptions are you making?
b. What is the probability that more
than half the boys in the sample want
to do their best in every class? What
assumptions are you making?
c. Suppose you are to take one sample of
40 students at random, instead of 20 girls
and 20 boys. What is the approximate
probability that more than half of those
sampled will want to do their best in
every class? What assumptions are you
making?
b. If it costs $40 to interview a respondent

with health insurance and $20 to
interview a respondent without health
insurance, what is the expected cost of
conducting the 500 interviews?
c. If the interviews are conducted
sequentially, what is the expected number
of interviews to be conducted before the
second person without health insurance
is found?
E51. Airlines routinely overbook popular flights
because they know that not all ticket holders
show up. If more passengers show up than
there are seats available, an airline offers
passengers $100 and a seat on the next flight.
A particular 120-seat commuter flight has a
10% no-show rate.
a. If the airline sells 130 tickets, how much
money does it expect to have to pay out
per flight?
b. How many tickets should the airline sell
per flight if it wants the chance of giving
any $100 payments to be about 0.05?
E52. Suppose random variable X has a geometric
distribution with probability of success
denoted by p.
a. Find the conditional probability that
X = 6 given that X > 5.
b. Find the probability that X > 6 given that
X > 5.
c. Find the probability that X exceeds
k + m given that it exceeds m. That is,
Find an expression for
P(X > k + m | X > m)
E50. It is estimated that 16% of Americans have

no health insurance. A polling organization
randomly samples 500 Americans to ask
questions about their health. [Source: U.S. Census
Bureau, Current Population Survey, March 2004.]
a. What is the probability that more than

420 of those sampled will have health
insurance?
This is referred to as the memoryless

property of the geometric distribution.
E53. This question demonstrates one reason why
the mean and variance are considered so
important. Suppose you select one book at
random to read from List A and one from
22-03-2009 21:09
Lesson
4 de 6
List B. The numbers given are the number of

pages in each book.
a. Find the total number of pages in each of

the nine different possible pairs of books.
Compute the standard deviation, , and
variance, , of these nine totals.
b. Compute the variance of the number
of pages in the three books in List A.
Compute the variance of the number of
pages in the three books in List B. What
relationship do you see between these two
variances and the variance in part a?
c. Does the same relationship as in part b
hold for the three standard deviations?
d. Can you add the mean of the numbers in
List A to the mean of the numbers in List
B to get the mean of the nine totals?
e. Can you add the median of the numbers
in List A to the median of the numbers
in List B to get the median of the nine
totals?
E54. Each box of a certain brand of cereal
contains a poster of a famous athlete. There
are five different posters, representing five
different athletes.
a. Suppose you are interested in a particular
athlete. Set up a simulation that would
produce an approximate probability
distribution of the number of boxes of
cereal you would have to buy in order to
get one poster of that athlete. (You stop
buying cereal when you get the poster
you want.)
b. What is the probability that you would
get the poster of that athlete in four or
fewer boxes of cereal? (You stop buying
cereal when you get the poster you want.)
c. What is the expected number of boxes of

cereal you would have to buy to get the
one poster you want?
d. Suppose you want to get two particular
posters out of the five available. What is
the expected number of boxes of cereal
you would have to buy to get the two
specific ones? (Hint: At the outset, the
probability of getting a poster you want
in any one box is . After you get one of
them, what is the probability of getting
the other poster you want? Think of the
expected number of boxes of cereal being
purchased in terms of these two stages.)
e. What is the expected number of boxes
of cereal you would have to buy to get
a full set of all five posters? Set up a
simulation for this event, and compare
your theoretical result to the one obtained
from the simulation.
E55. Refer to the copper flipper description in
D22 on page 397. This time suppose you
start with 10 pennies. As in Activity 3.5a,
you flip all 10, remove the ones that have
died (landed tails), and continue flipping
with a decreasing number of pennies.
a. On average, how many times do you
have to flip a specified copper flipper
until it dies?
b. If you flip 10 copper flippers at once, what
is the probability that at least one dies?
c. On average, how many flips of the entire
set of 10 copper flippers does it take to get
at least one copper flipper that dies? Is the
answer more or less than the answer for
part a? Why?
d. Design and carry out a simulation to
answer this question: On average, how
many sets of flips will it take for all
10 copper flippers to die?
Chapter Summary
405
22-03-2009 21:09
Lesson
5 de 6
AP1. This table gives the percentage of women who

ultimately have a given number of children.
For example, 19% of women ultimately have
3 children. What is the probability that two
randomly selected women will have a total of
exactly 2 children?
0.0289
0.1549
0.17
0.34
0.70
AP2. In a raffle, 100 tickets are sold and 10 of the
tickets, selected at random, win a $5 prize.
A participant buys one ticket. What is the
standard deviation of the amount won?
$0.50
$1.50
$2.25
$2.50
$4.50
AP3. A machine producing a large spool of
cable puts out 2.5 ft at a time. After each
2.5 ft of cable, theres a 1/1000 chance that
the machine will cut the cable instead
of continuing production. What is the
expected length of the next cable the
machine produces?
0.0025 ft
400 ft
2,500 ft
6,250 ft
2,500,000 ft
AP4. Russells parents buy two apples each

weekend, and put them in his lunch on
two randomly selected weekdays of the
following week. What is the probability that
Russell gets an apple on exactly one day that
is a Monday or Tuesday?
AP5. Russells parents roll a fair six-sided die

every morning, and if the result is a 1 or 2,
they pack yogurt in his lunch. What is the
probability that he gets yogurt on exactly
2 of the 5 school days next week?
0.111
0.400
0.165
0.329
0.671
AP6. Dawna walks 10 miles starting at 1:00 p.m.
every day. On average she finishes at
3:50 p.m., with a standard deviation of
10 minutes. Jeanne leaves 30 minutes later,
and every day she runs 10 miles exactly
twice as fast as Dawna walks on that day.
What is the mean time Jeanne will finish
and the standard deviation of the number
of minutes it takes Jeanne to finish?
mean 2:25 p.m., standard deviation
5 minutes
20 minutes
22-03-2009 21:09
Lesson
6 de 6

35 minutes
5 minutes
35 minutes
AP7. According to the World Bank, in the year
2000 approximately 35% of all people in
India spent less than $1 per day. Suppose
that you repeatedly take random samples
of 100 Indians and record the number who
spent less than $1 per day. What are the
approximate mean and standard deviation
of this distribution?
mean 0.35, standard deviation 0.2275
mean 35, standard deviation 0.2275
AP8. In a gas station promotion, customers
receive a game piece every time they fill
their tank. With each game piece, the
customer has a 1 in 1250 chance of winning
a $10 gasoline card. Josie decides to fill her
tank at that gas station until she wins the
gasoline card. What is the probability that
she fills her tank there exactly 5 times?
0.0008
0.0040
0.9960
4.9800
7.9744
Investigative Tasks
AP9. Random sampling from a large lot of a
manufactured product yields a number
of defectives, X, with an approximate
binomial distribution, with p being the
true proportion of defectives in the lot.
A sampling plan consists of specifying the
number, n, of items to be sampled and an
acceptance number, a. After n items are
inspected, the lot is accepted if X a and
is rejected if X > a.
a. For n = 5 and a = 0, calculate the
probability of accepting the lot for values
of p equal to 0, 0.1, 0.3, 0.5, and 1.
b. Graph the probability of lot acceptance
as a function of p for this plan. This is
the operating characteristic curve for the
sampling plan.
Now, a quality-control engineer is
considering two different lot acceptance
sampling plans: (n = 5, a = 1) and
(n = 25, a = 5).
c. Construct operating characteristic
curves for both plans.
d. If you were a seller producing lots with
proportions of defectives between 0%
and 10%, which plan would you prefer?
e. If you were a buyer wishing high
protection against accepting lots with
proportions of defectives over 30%,
which plan would you prefer?
AP Sample Test
407
22-03-2009 21:09
Lesson
1 de 2
Sampling Distributions
What would happen

if you could take
random samples over
and over again from
your population?
Sampling distributions
show how much your
results might vary from
sample to sample,
as when estimating
the mean number of
departures from U.S.
airports for a given
period of time.
22-03-2009 21:10
Lesson
2 de 2
You have studied methods that are good for exploration and description, but for
inferencegoing beyond the data in hand to conclusions about the population
or the process that created the datayou need to collect the data by using a
random sample (a survey) or by randomly assigning treatments to subjects (an
experiment). The promise of Chapter 4 was that if you used those methods to
produce a data set, you then could use your data to draw sound conclusions.
Randomized data production not only protects against bias and confounding but
also makes it possible to imagine repeating the data production process so you
can estimate how the summary statistic you compute from the data would vary
from sample to sample. To oversimplify, but only a little, a sampling distribution is
what you get by repeating the process of producing the data and computing the
summary statistic many times.

how to use simulation to generate approximate sampling distributions
of common summary statistics such as the sample mean and the sample
proportion
to describe the shape, center, and spread of the sampling distributions of
common summary statistics without actually generating them
to use the sampling distribution to determine which results are reasonably
likely and which would be considered rare
22-03-2009 21:10
Lesson
1 de 16
7.1
A sampling distribution
answers the question
How does my summary
statistic behave when
I repeat the process
many times?
Generating Sampling Distributions

Sometimes it is possible to imagine repeating the process of data collection. A
sampling distribution is the distribution of the summary statistics you get from
taking repeated random samples. For example, in Chapter 1 (Display 1.10), you
created a simulated sampling distribution of the mean age from random samples
of three employees who could have been laid off at Westvaco. This distribution is
shown again in Display 7.1. There were ten workers who could have been laid off;
their ages were
25 33 35 38 48 55 55 55 56 64
You randomly selected three of these workerswithout replacement, because
Westvaco couldnt have laid o the same worker twice.
Your summary statistic was the mean (average) age of the three workers you
chose at random. You were able to repeat this process many times in order to
generate a distribution of possible values of the summary statistic.
Using this simulated sampling distribution, you could make a decision about
whether it was reasonable to assume that employees were selected for layoff
without respect to their age. The ages of the three workers actually laid offwere
55, 55, and 64, for an average age of 58. As you can see from Display 7.1, its
rather hard to get an average age that large just by chance. So Westvaco had
some explaining to do.
Display 7.1 A simulated sampling distribution of the mean age

from random samples of three people who could
have been laid off at Westvaco.
Random sample
Summary statistic
Repetition
Distribution
In the Westvaco analysis, you went through four steps in using simulation to
generate an approximate sampling distribution of the mean age of three workers:
1. Take a random sample of a fixed size n from a population.
2. Compute a summary statistic.
3. Repeat steps 1 and 2 many times.
4. Display the distribution of the summary statistics.
410 Chapter 7 Sampling Distributions
22-03-2009 21:10
Lesson
2 de 16
These steps describe how to use simulation to make an approximate sampling

distribution or simulated sampling distribution. If you had listed all possible samples
of 3 workers who could have been laid off from the 10 workers and computed the
mean age for all 10C3 = 120 samples, you would have constructed the exact or
theoretical sampling distribution, called simply the sampling distribution.
Shape, Center, and Spread: Now and Forever

In Chapter 4, you sampled from a population of 100 rectangles. Display 7.2 shows
their areas, which have mean 7.4 and standard deviation 5.2.
Display 7.2 The population of rectangle areas.
In Activity 4.2a on pages 232235, you generated an approximate sampling

distribution when you selected five rectangles at random and calculated their
mean area, doing this over and over again. You probably got a distribution
something like that in Display 7.3. Each of the 1000 dots in Display 7.3 represents
the mean area of a random sample of five rectangles.
Display 7.3
A simulated sampling distribution of the sample

mean of five rectangles.
A good description of a sampling distribution like that in Display 7.3 is still

the trio
Shape Center Spread
411
22-03-2009 21:10
Lesson
3 de 16
Two standard deviations:

population standard
deviation and standard
error of the sampling
distribution.
Two events: reasonably

likely and rare
This simulated sampling distribution is approximately normal, with a hint of a

skew to the right, with mean 7.4 and standard deviation 2.3.
Compare Display 7.3 with the population of the areas of all 100 rectangles
in Display 7.2. The two distributions both have mean 7.4. However, there are
two quite different standard deviations in the picturethe standard deviation
of the population, 5.2, and the smaller standard deviation of the sampling
distribution, 2.3. To tell them apart, these two types of standard deviations have
different names. The standard deviation of the population, , is called, naturally
enough, the population standard deviation. The standard deviation of the
sampling distribution is commonly called the standard error, or SE for short.
If you select one of the rectangles at random and get one with an area of 16,
you shouldnt be at all surprised. You can see from Display 7.2 that 15 of the
100 rectangles, or 15%, have an area of 16 or larger. Values like this that lie in the
middle 95% of a sampling distribution are called reasonably likely events. Values
that lie in the outer 5% of a sampling distribution are called rare events. In a
normal distribution, rare events are those values that lie more than two standard
deviations from the mean.
Notice that if you take a random sample of five rectangles and their mean
area is 16, you should be very surprised. You can see from the simulation in
Display 7.3 that a mean this large happened only once in 1000 runs and so would
be a rare event.
Shape, Center, and Spread

D1. Compare Displays 7.2 and 7.3.
a. Why is the largest value in Display 7.2 larger than the largest value in
Display 7.3? Why is the smallest value smaller?
b. Would you expect the means of the distributions in Displays 7.2 and 7.3
to be equal? Why or why not?
c. Would you expect the two standard deviations to be equal? Why or
why not?
d. How does the shape of the sampling distribution for samples of size 5
compare to the shape of the distribution of the population of the areas of
all 100 rectangles?
D2. Approximately what values of the sample mean for samples of size 5 would
be reasonably likely? Which would be rare events?
Once youre used to the steps involved in generating sampling distributions
and have a calculator or computer that lets you do simulations easily, there is no
end to the variety of summary statistics you can study. [See Calculator Note 7A to
learn about the kinds of sampling distributions you can generate with a calculator.]
In Activity 7.1a, you will use the 100 rectangles from Display 4.5 on page 233
to generate simulated sampling distributions of two summary statistics other than
the mean: the sample median and the sample maximum.
22-03-2009 21:10
Lesson
4 de 16
The Return of the Random Rectangles

What youll need: a copy of Display 4.5, a method of producing random digits
1. Generate five distinct random numbers between 00 and 99.
2. Find the rectangles in Display 4.5 on page 233 that correspond to your
random numbers. (The rectangle numbered 100 can be called 00.) This is
your random sample of five rectangles.
3. Determine the areas of the rectangles in your random sample and find the
sample median. (Save these five areas for step 7.)
4. Combine your sample median with those of other students in the class and
repeat steps 1 through 3 until you have 200 sample medians. Construct a
plot showing the distribution of the 200 sample medians.
5. Describe the shape, mean, and standard error of this plot of sample
medians. How do these compare to the shape, mean, and standard
deviation of the population of areas in Display 7.2?
6. How does this sampling distribution compare to the simulated sampling
distribution of the sample mean in Display 7.3?
7. Repeat steps 4 and 5, but this time use the maximum area of the five
rectangles in each sample as the summary statistic. For example, if the areas
in your sample are 1, 8, 4, 8, and 3, the maximum area is 8. (You can use
the same samples as before.) You will also need to keep your data for E13.
8. Compare the population in Display 7.2 to the sampling distributions you
plotted.
a. What is the median of the population of 100 rectangle areas? Is the sample
median a good estimate of the median of the areas of all 100 rectangles?
b. What is the maximum of the population of 100 rectangle areas? Is
the sample maximum a good estimate of the largest area of the
100 rectangles?
9. Reasonably likely and rare events. Use the plots you constructed to answer
these questions.
a. What values of the sample median are reasonably likely? What values
would be rare events?
b. What values of the sample maximum are reasonably likely? What values
would be rare events?
Exact Sampling Distributions

When the population size is very small, you can construct sampling distributions
exactly by listing all possible samples.
Simulation isnt
necessary here.
Example: Mapping National Parks

Utah has five national parks. Your company has been hired to make maps of
two of these parks, which will be selected at random. Use the information in
Display 7.4 to construct the sampling distribution for the total number of square
413
22-03-2009 21:10
Lesson
5 de 16
miles you would map. Then find the probability that you will have to map more
than 600 square miles.
Display 7.4 Sizes of Utahs national parks.
Solution
There are 5C2, or 10, equally likely ways to select two national parks, as shown in
Display 7.5. The relative frequency histogram shows the sampling distribution
of the sum. The probability that you will have to map more than 600 square
miles is
Display 7.5
Total number of square miles in two of Utahs

national parks.
22-03-2009 21:10
Lesson
6 de 16
Properties of Point Estimators

Sampling distributions are the connecting link between the collection of data
through designed studies based on random sampling or random assignment
of treatments to subjects, and statistical inference, the process of drawing
defensible conclusions from the data. When inferences are made from the sample
to the population, the sample mean is viewed as an estimator of the mean of
the population from which the sample was selected. Similarly, the proportion
of successes in a sample is an estimator of the proportion of successes in the
population. These two statistics are called point estimators for the reason that, for
a given sample, they provide only one point (number) as a plausible value of the
corresponding population parameter. (In the coming chapters you will learn how
to construct an interval of plausible values of the population parameter.)
Suppose you want to know the mean area of the rectangles whose areas are
plotted in Display 7.2. Because of time or money constraints, you cannot measure
them all. You know what to do: Take a random sample from the population
of rectangles; compute the sample mean area, ; use as your estimator of the
population mean area, .
This process seems so natural that it is easy to overlook just how powerful it
is. Two very fortunate things are happening in the background:
Although each individual sample mean usually is smaller or larger than
the population mean , in the long run, over many samples, the average value
of turns out to be exactly equal to .
The value of tends to be pretty close to the value of , and, as you will see in
the next section, the larger the sample size, the closer to it tends to be.
Even though in practice you cant repeat the sampling process, it is nice to
know whether the estimator you are using generally works out well. For example,
from Display 7.3, you can see that while a single estimate from a sample of size 5
might be too small or too large, the estimates are centered around the population
mean, 7.4. Display 7.3 also gives you an idea of how close your estimate is likely
to be to 7.4 if you sample only five rectangles. Some desirable properties of
estimators are generalized in the box.
When you use a summary statistic from a sample to estimate a parameter of a
population, there are two properties that you would like the summary statistic
to have.
The summary statistic should be unbiased. That is, the mean of
the sampling distribution is equal to the value you would get if you
computed the summary statistic using the entire population. More
formally, an estimator is unbiased if its expected value equals the
parameter being estimated.
The summary statistic should have as little variability as possible (be
more precise than other estimates) and should have a standard error
that decreases as the sample size increases.
415
22-03-2009 21:10
Lesson
7 de 16
Example: Estimating the Distance of the Farthest Galaxy

Your observatory is given the names of 31 galaxies. Your job is to estimate the
maximum distance from Earth among the 31 galaxies. You have the resources
to measure the distance of only 10 galaxies, which you will select at random.
Unbeknownst to you, the distances, in megaparsecs (Mpc), are
0.008 0.76 0.81 7.2 7.5 9.7 10.6 11.2 11.4 11.6 11.7 13.2 15 15 15.3 15.5
15.7 16.1 16.1 16.8 16.8 20.9 22.9 22.9 24.1 25.9 26.2 29.2 31.6 58.7 93
The maximum distance, 93 Mpc, is the population parameter you hope to
estimate.
Because you want to estimate the maximum distance from Earth, you decide
to use the maximum distance in the sample as your summary statistic. For
example, if your sample of size 10 is
0.81 7.5 11.6 11.7 13.2 16.1 16.8 24.1 26.2 58.7
you will estimate the population maximum to be 58.7. This is smaller than the
maximum in the population, so you have underestimated.
Display 7.6 shows a simulated sampling distribution of the maximum of
samples of size 10. What will happen when you use the maximum of the sample
as an estimator of the maximum of this population?
Display 7.6
The sample maximum

can never be too big!
Simulated sampling distribution of the maximum

of samples of size 10 taken from the population of
31 galaxies.
Solution
The population maximum is 93, but the mean of the sampling distribution is only
56. If you could repeat your process of estimating the population maximum by
using the maximum in the sample, on average your estimate would be too small.
In other words, the sample maximum is a biased estimator of the population
maximum. Thats not too surprising, because the maximum of a sample can
never be larger than the maximum of its population.
22-03-2009 21:10
Lesson
8 de 16
Further, there is a great deal of variability in the sample maximum. Your

estimate of the population maximum might be as high as 93 Mpc (the population
maximum), but about half the time your estimate will be less than 32, way too
small. Using the sample maximum as an estimator of the population maximum
doesnt work out very well.

D3. In what way are the ideas biased estimator and bias in sampling similar?
In what way are they different?
Summary 7.1: Sampling Distributions
Random sample
Summary statistic
Repetition
Distribution
A sampling distribution answers the question How would my summary statistic

behave if I could repeat the process of collecting data using random samples?
By showing you the summary statistics from random samples, a sampling
distribution can help you decide whether the same process created the result you
got from your actual sample.
You can generate a simulated sampling distribution of any sample statistic by
following these steps:
1. Take a random sample of size n from a population.
2. Compute a summary statistic.
You can sometimes create an exact sampling distribution by listing all possible
samples. In many more situations, you will be able to describe the shape, mean,
and standard error of the sampling distribution without simulation and without
listing samples. You will learn how to do this for sample means and sample
proportions in Sections 7.2 and 7.3.
Here are several key facts about sampling distributions:
A simulated, or approximate, sampling distribution is the distribution of the
sample statistic for a large number of repeated random samples.
Sampling distributions, like data distributions, are best described by shape,
center, and spread.
The standard deviation of a sampling distribution is called the standard error.
Many, but not all, sampling distributions are approximately normal. For
normal distributions, reasonably likely outcomes are those that fall within
approximately two standard errors of the mean.
If the mean of the sampling distribution is equal to the population parameter
being estimated, then the summary statistic you are using is an unbiased
estimator of that parameter.
417
22-03-2009 21:10
Lesson
9 de 16
Practice
Shape, Center, and Spread
P1. Display 7.7 gives the number of moons for
each planet in our solar system. For a term
paper in your astronomy class, you will
be given three different planets selected at
random and be asked to write a 50-word
description of each of the planets moons.
P2. Display 7.8 shows the distribution of exam

scores for 192 students in an introductory
statistics course at the University of Florida.
Display 7.8 A distribution of exam scores.
Display 7.7
Number of moons for the planets in

our solar system. [Source: NASA, solarsystem
.nasa.gov.]
a. What is the smallest number of words

you might have to write? The largest?
b. Describe how to generate a simulated
sampling distribution of the total number
of moons you must describe.
c. Generate 20 values for a simulated
sampling distribution and make a plot.
d. Use your plot from part c to estimate the
probability that you will have to write
150 words or fewer.
a. Match each histogram in Display 7.9 to

its description.
I. the individual scores for one random
sample of 30 students
II. a simulated sampling distribution of
the mean of the scores of 100 random
samples of 4 students
III. a simulated sampling distribution of
the mean of the scores of 100 random
samples of 30 students
Display 7.9 Histograms for exam scores.
Jupiter and its four planet-sized moons
b. The second-hour class of 30 students,

whose exam scores are part of the data
set, had a class average of 86 on this
exam. The instructor says that this class
is just a random assortment of typical
students taking this course. Do you
agree?
22-03-2009 21:10
Lesson
10 de 16
P3. Using the exam scores of P2, simulated

sampling distributions were generated for
the minimum, maximum, median, and lower
quartile for samples of size 10.
a. Match each histogram in Display 7.10 to
its summary statistic.
I. the minimum
II. the maximum
III. the median
IV. the lower quartile
b. Which sampling distribution has the
largest standard error? The next largest?
Exact Sampling Distributions

P4. Every year, Forbes magazine releases a list of
the top-earning dead celebrities. In 2005, the
top six and their yearly earnings were
1. Elvis Presley, $45 million
2. Charles M. Schulz, $35 million
3. John Lennon, $22 million
4. Andy Warhol, $16 million
5. Theodor Dr. Seuss Geisel, $10 million
6. Marlon Brando, $9 million
[ Source: www.forbes.com.]
Your talent agency gets an opportunity to

represent two of these dead celebrities, to be
selected at random. You will be paid 10% of
their earnings.
a. What is the most you could be paid? The
least?
b. Construct the sampling distribution of
your total possible earnings.
c. What is the probability that you will be
paid $3 million or more?
P5. The areas of the five national parks in Utah
Display 7.11 Areas of the five national parks in Utah.
a. What is the mean area of this population

of parks? The standard deviation?
b. Complete the table in Display 7.12 (on
the next page) and use it to construct the
sampling distribution of the mean area
of a random sample of two parks.
Display 7.10 Histograms of different summary
statistics from random samples of
size 10 of statistics exam scores.
419
22-03-2009 21:10
Lesson
11 de 16
a. What is the value of the population

parameter that you are estimating?
b. Refer to Display 7.11 and complete a copy
of the table in Display 7.13.
c. Construct a plot of the exact sampling
distribution of the sample range. What is
the mean of this sampling distribution?
d. Is the sample range an unbiased estimator
of the population range? Explain.
e. Is there much variability in the estimates?
Display 7.12 Mean area of two randomly chosen
national parks in Utah.
c. Verify that the sample mean is an

unbiased estimator of the population
mean. (In other words, verify that the
mean of the sampling distribution
is exactly equal to the mean of the
population.)
d. Verify that the standard error of the
sampling distribution is smaller than the
standard deviation of the population.
P6. Refer to the example on page 413. Your
companys contract has changed. You now
are to estimate the range (largest minus
smallest) of the areas of Utahs national
parks. Your procedure will be to select three
of the ve parks at random and use the range
of their three areas as the estimate.
Display 7.13 Range of areas of three randomly

chosen national parks in Utah.
Exercises
E1. Random samples are taken from the
population of random digits 0 through 9,
with replacement.
a. Each histogram in Display 7.14 is a
simulated sampling distribution of the
sample mean. Match each sampling
distribution to the sample size used:
1, 2, 20, or 50.
b. Compare the means. How does the mean
of the sampling distribution depend on
the sample size?
c. Compare the spreads. How does the

spread of the sampling distribution
depend on the sample size?
E2. Three very small populations are given, each
with a mean of 30.
A. 10 50
B. 10 20 30 40 50
C. 20 30 40
Match each population to the sampling
distribution of the sample mean (Display
7.15) for a sample of size 2 (taken with
replacement).
22-03-2009 21:10
Lesson
12 de 16
Display 7.14 Histograms of sample means of random digits.
Display 7.15 Sampling distribution of sample means for n = 2.
E3. The mean and the median are only two of

many possible measures of center. Another
is the midrange, defined as the midpoint
between the minimum and the maximum in
a data set. The rectangles in Display 4.5 on
page 233 have areas ranging from 1 to 18, so
the population midrange is
, or 9.5. The
histogram in Display 7.16 shows a simulated
sampling distribution of the midrange
based on 1000 random samples of size five
from the population of 100 rectangles. The
mean of this distribution is 8.05, and the
estimated standard error is 2.17.
a. Take a random sample of five rectangles;
compute the midrange of their areas.
Display 7.16 A sampling distribution of the sample

midrange (n = 5).
b. Describe the sampling distribution of the

midrange. How does it compare to the
sampling distribution of the sample mean
in Display 7.3 on page 411?
c. Would you use the midrange as a
measure of center for the rectangle area
population? Why or why not?
421
22-03-2009 21:10
Lesson
13 de 16
E4. The areas of all 50 U.S. states, in millions of

acres, are given in Display 7.17.
a. Describe the shape of the distribution.

b. Use a table of random digits to select a
random sample of five areas. Do these
areas appear to be representative of the
population? What is the mean of your
five areas?
c. The plot in Display 7.18 shows the mean
area for 25 random samples, each of size
5, just like the sample you generated in
part b. Describe this simulated sampling
distribution.
Display 7.18 A simulated sampling distribution of

the sample mean of 5 state areas.
d. From what particular sample could the

largest value in the plot in Display 7.18
have come?
E5. The five tennis balls in a can have diameters
62, 63, 64, 64, and 65 mm. Suppose you
select two of the tennis balls at random,
without replacing the first before selecting
the second.
a. Construct a dot plot of the five
population values.
b. List all possible sets of size 2 that can be
chosen from the five balls. There are 5C2
(5 choose 2), or 10, possible sets of
two balls.
c. List all 10 possible sample means, and
construct a dot plot of the sampling
distribution of the sample mean.
Compute the mean and standard error
of this distribution. Compare these to
the mean and standard deviation of the
population.
Display 7.17 Areas of the 50 U.S. states in millions

of acres (not including inland bodies
of water). [Source: World Almanac and Book of
Facts 2001, p. 621.]
22-03-2009 21:10
Lesson
14 de 16
d. Construct a dot plot of the sampling

distribution of the maximum diameter of
a sample of size 2.
e. Construct a dot plot of the sampling
distribution of the range (maximum
minus minimum) of a sample of size 2.
E6. As part of a statistics project at Iowa State
University, a student tested how well a bike
with treaded tires stopped on concrete. In
six trials, these lengths of skid marks (in
centimeters) were produced: 365, 374, 376,
391, 401, 402. Suppose instead that the
student had done only three trials, which
could have been any three selected from this
population of six, each with equal probability
of being selected. [Source: Stephen B. Vardeman,
Statistics for Engineering Problem Solving (Boston:
Prindle, Weber & Schmidt, 1994), p. 349.]
a. Construct a dot plot of the six population

values.
b. List all possible sets of three skid lengths
the student could get. There are 6C3, or
20, possible sets.
c. Construct a dot plot of the sampling
distribution of the sample mean. Find
its mean and standard error. Is the mean
equal to the mean of the population?
d. Construct a dot plot of the sampling
distribution of the sample median. Find
its median. Is it equal to the median of
the population?
e. Construct a dot plot of the sampling
distribution of the sample minimum.
Describe the shape of this distribution. Is
the sample minimum a good estimator

of the population minimum of all six
lengths? Explain.
E7. Refer to your exact sampling distributions
in E5. For each summary statistic here,
explain why the summary statistic is or is
not an unbiased estimator of the population
parameter. If it is a biased estimator, does it
tend to be too big or too small, on average?
a. sample mean (see part c)
b. sample maximum (see part d)
c. sample range (see part e)
E8. Refer to your exact sampling distributions
in E6. For each summary statistic here,
explain why the summary statistic is or is
not an unbiased estimator of the population
parameter. If it is a biased estimator, does it
tend to be too big or too small, on average?
a. sample mean (see part c)
b. sample median (see part d)
c. sample minimum (see part e)
E9. An inspector is called to see how much
damage has been caused to a warehouse
full of frozen fish by a recent power failure.
The fish are stored in cartons containing
24 one-pound packages. There are hundreds
of cartons in the warehouse.
The inspector decides he needs a sample of
48 one-pound packages of fish in order to
assess the damage. He selects two cartons
at random and treats this as a sample of
48 one-pound units.
The inspector doesnt realize it, but if one
package of fish in a carton spoils, very
rapidly the whole carton spoils. So it is safe
to assume that each carton in the warehouse
is either completely spoiled or not spoiled
at all.
a. If the inspector is using pounds of spoiled
fish in the sample as a statistic, what will
the sampling distribution of the statistic
look like? Describe this distribution as
completely as you can.
b. Did the inspector choose a good
sampling plan? If not, explain how he
could have chosen a better one.
423
22-03-2009 21:10
Lesson
15 de 16
E10. Joel knows that the sample range is a biased

estimator of the population range, so he has
a plan to use a different estimator. Instead
of using the sample range to estimate the
population range, he will double the sample
interquartile range (IQR) and use that as an
estimate of the population range. He will
try this out using the population of numbers
{0, 1, 2, 3, . . . , 98, 99, 100} and a sample size
of 8, taken with replacement.
a. What is the value of the population
parameter Joel is trying to estimate?
b. Explain why the sample range is a biased
estimator of the population range. You
can use Display 7.19 in your explanation.
c. To test his estimator, Joel does a

simulation. He takes 10,000 samples of
size 8, computes the IQR of each one,
and doubles it. His results are shown in
Display 7.20. Does Joels estimator of the
range appear to be biased? Do you have
any other concerns about it?
Display 7.20 Measures of 2 IRQ from sample of

range 0 to 100.
For E11E12: In Chapter 2, you might have

wondered why the denominator of the sample
standard deviation is n 1 rather than n. Now
that you know about sampling distributions,
you can use them to answer this question in true
statistical fashion. If a sample standard deviation is
to be a good estimator of the population standard
Display 7.19 Simulated sampling distribution of the
sample range.
22-03-2009 21:10
Lesson
16 de 16
deviation , its sampling distribution should be

centered at or at least near that value. In E11
and E12, you will see whether that turns out to
be the case.
E11. Suppose you have a population that consists
of only the three numbers 2, 4, and 6. You
take a sample of size 2, replacing the first
number before you select the second.
Use parts ae to complete the table in
Display 7.21.
Display 7.21 Variance calculations.
a. In the first column, list all nine possible

samples. (Count 2, 4 as different from
4, 2; that is, order matters.)
b. Compute the mean of each sample and
write it in the second column.
c. Compute the variance of each sample,
dividing by n = 2, and enter it in the
third column. (You should be able to do
this in your head.)
d. Compute the variance of each sample,
dividing by n = 1, or 1, and enter it in the
fourth column.
e. Compute the average of each column.
f. Compute the variance of the population
{2, 4, 6} using the formula on page 366
with the probabilities of selection being
equal for the three values. Compare it to
your answers in part e.
g. Do you get an unbiased estimate of
the variance of the population if you
divide by n or by n = 1? Explain how
you know.
h. If you divide by n rather than by n 1,
does the variance tend to be too large or
too small, on average?
E12. Look at the rectangle areas plotted in

Display 7.2 on page 411. Take a random
sample of five rectangles. Compute the
standard deviation of the areas of these
rectangles using the regular formula,
dividing by n 1, or 4. Then compute the
standard deviation again, but divide by n,
or 5. Which standard deviation is larger?
The mean area of the population is 7.42, and
the population standard deviation, , is 5.20.
The first histogram in Display 7.22 shows the
sample standard deviations from the same
samples of size 5 whose means are plotted
in Display 7.3 (page 411). The mean of these
standard deviations is 5.02.
If you divide by 5 rather than by 4 (that is, by
n instead of n = 1), the results produce the
distribution shown in the second histogram.
The mean of this distribution is 4.49. Explain
why dividing by n = 1 gives a better estimate
of the population standard deviation.
Display 7.22 Simulated sampling distributions of

sample standard deviations (n = 5).
425
22-03-2009 21:10
Lesson
1 de 20
E13. As you found in Activity 7.1a, step 7, if you

use the sample maximum as an estimate of the
population maximum, it tends to be too small.
Here is one possible rule for adjusting it:
b. Where would you expect the maximum

of a sample of size 4 to be?
c. If you know the sample maximum of a
sample of size 4 taken from a uniform
distribution on [0, N], what would you
(population maximum)
(sample maximum)
expect N to be?
d. Test how well this rule works using your
As always, n is the sample size. This rule
results from Activity 7.1a, step 7.
works best when the population is uniform.
E14.
You
have seen a number of examples
a. Suppose you take a random sample
in which the variation in the sampling
of four numbers from the uniform
distribution of the sample mean is
distribution on [0, N], where N is the
smaller than the variation in the sampling
population maximum. On average, what
distribution of the sample median. However,
size gap would you expect between two
these examples arose from populations that
adjacent numbers in your sample? The
were symmetric, or nearly so. Suppose the
dot plot in Display 7.23 might help. It
distribution for your population is highly
shows a sample of size 4 taken from a
skewed (for example, the areas of the U.S.
uniform distribution with minimum 0
states given in E4). Will the sampling
and maximum N.
distribution of the sample median have less
variation than the sampling distribution of
the sample mean? Use simulation to provide
evidence for your answer.
Display 7.23 Four numbers randomly chosen from a
uniform distribution with minimum 1.
7.2
For a finite population,
the sampling distribution
of the sample mean is
the set of means from
all possible samples of a
specific size.
Sampling Distribution of the Sample Mean

You have seen that the sample mean (average) is a very important statistic,
especially for symmetric distributions. In fact, you knew that the mean is
important before you ever took a statistics course because you see averages used
all around youaverage exam score, average income, average age, batting average,
and on and on. In this section, you will learn to predict the mean, standard
deviation, and shape of the sampling distribution of the sample mean. Because
its properties are so well understood, the sample mean will be used almost
exclusively as the measure of center in the remainder of this book.
Some of the distributions of data that you have studied so far have had a
roughly normal shape, but many others were not at all normal. Part of what makes
the normal distribution important is its tendency to emerge when you create
sampling distributions. Activity 7.2a shows how the normal distribution comes
up unexpectedly when you work with the sampling distribution of a mean.
22-03-2009 21:11
Lesson
2 de 20
Cents and Center

What youll need: 25 pennies collected from recent day-to-day change
1. If you were to construct a histogram
of the ages of all the pennies from
all the students in your class, what
do you think the shape of the
distribution would look like?
2. Find the age of each of your pennies
by subtracting the date on the penny
from the current year. Construct a
histogram of the ages of all the pennies in the class.
3. Estimate the mean and standard deviation of the distribution. Then
confirm these estimates by actual computation.
4. Take a random sample of size 5 from the ages of your classs pennies, and
compute the mean age of your sample.
5. If you were to construct a histogram of all the mean ages computed by the
students in your class in step 4, do you think the mean of the values in this
histogram would be larger than, smaller than, or the same as the mean for
the population of the ages of all the pennies? Regardless of your choice, try
to make an argument to support each choice. Estimate what the standard
deviation of the distribution of the mean ages will be.
6. Construct the histogram of the mean ages, using nickels if you have them,
and determine the mean and standard deviation. Which of the three
choices in step 5 appears to be correct?
7. Repeat steps 46 for samples of size 10, using dimes, and size 25, using
quarters.
8. Look at the four histograms that your class has constructed. What can you
say about the shape of the histogram as n increases? About the center?
About the spread?
In the rest of this section, youll see if the results from Activity 7.2a, step 8, are
true for other distributions.
Shape, Center, and Spread of the Sampling

Distribution of
Display 7.24 shows the distribution of the number of children per family in
the United States. This isnt a sampling distribution; rather, its the population
distribution you will use to build a sampling distribution.
427
22-03-2009 21:11
Lesson
3 de 20
Display 7.24 The number of children per family in the United

States. [Source: U.S. Census Bureau, Statistical Abstract of the
United States, 2004-2005, www.census.gov.]
If you count all families with four or more children as having four children,
this highly skewed population has a mean of about 0.9 and a standard deviation of
about 1.1. [See Calculator Note 6A to review how to calculate the mean and standard
deviation of a probability distribution.]
Random sample
Suppose you are working for a video game company that wants to sample
families to study interests of children. What will the sampling distributions of the
mean number of children per family look like? These four steps review how to
construct a simulated sampling distribution of the mean for samples of size 4.
1. Take a random sample from a population.
In your model, there should be a 0.524 chance that a randomly selected family
will have no children, and so on. Use a table of random digits, as you did in
Section 5.2, or use a calculators random number generator to select a random
sample of four families from this distribution.
22-03-2009 21:11
Lesson
4 de 20
Summary statistic
Repetition
Distribution
2. Compute a summary statistic for your random sample.

Suppose your random sample represents families with two, two, one, and two
children. The sample mean is then

The five sampling distributions of the sample mean in Display 7.25 are for
samples of size 1, 4, 10, 20, and 40. Note that the exact sampling distribution of
the mean for samples of size 1 is identical to the population distribution.
Display 7.25 Sampling distributions of the sample mean for samples of size 1, 4, 10, 20, and 40.
429
22-03-2009 21:11
Lesson
5 de 20
The plots and summary table of these sampling distributions reveal some
interesting patterns:
Thee population started out with only five different values and a large skew
toward the larger numbers. The sampling distribution for samples of size
1 looks almost identical to the distribution of the population, as it should.
The sampling distribution for samples of size 4 is still somewhat skewed but
has more values. The sampling distribution for samples of size 10 is even
less skewed.The sampling distributions for samples of size 20 and 40 look
nearly normal. ( This progression toward normality will turn out to be very
important.)
The means of all five sampling distributions are equal, 0.9. This is the mean of
the population.
The standard errors of the five sampling distributions decrease from 1.1 for
samples of size 1 (which is the same as the population standard deviation) to
0.55 for samples of size 4, to 0.35 for samples of size 10, and so on. Just as you
would predict, the sample mean tends to be closer to the population mean
with larger samples than it is with smaller samples.
It is time to distill the information you have worked with into a set of rules
for describing the behavior of sampling distributions of the sample mean. The
symbols commonly used are organized in this table.
Common symbols
Properties of the Sampling Distribution of the Sample Mean
Center
If a random sample of size n is selected from a population with mean and

standard deviation , then
the mean, - of the sampling distribution of x equals the mean of the
population, :
Spread
the standard deviation, , of the sampling distribution of x, sometimes

called the standard error of the mean, equals the standard deviation of
the population, , divided by the square root of the sample size n:
Shape
the shape of the sampling distribution will be approximately normal

if the population is approximately normal; for other populations,
the sampling distribution becomes more normal as n increases (This
property is called the Central Limit Theorem. )
22-03-2009 21:11
Lesson
6 de 20
All three properties are of great importance in statistics, and all three depend
says that the means of random
on random samples. First property,
samples are centered at the population mean. This might seem obvious, and it is
obvious for symmetric populations. But as you saw in Activity 7.2a, its also true
for skewed populations.
is the main reason for using the standard
The second property,
=
deviation to measure spreadyou can nd the standard error of the sample
mean without simulation . This result validates our intuitive feeling of why large
samples are better: The larger the sample, the closer its mean tends to be to the
population mean.
The third property, the Central Limit Theorem, deals with the shape of the
sampling distribution and will be central to your work in the rest of this course. It
helps you decide which outcomes are reasonably likely and which are not.
_
Shape, Center, and Spread of the Sampling Distribution of x
D4. Why is it the case that the sampling distribution of the mean for samples of
size 1 is identical to the population distribution?
D5. The scatterplot in Display 7.26 shows the standard error, SE, plotted
against the sample size, n, for the table in Display 7.25 on page 429. Find a
transformation that linearizes these points. Use the transformation and the
equation of the resulting least squares line to justify the rule SE =
Display 7.26 SE plotted against n.
D6. Justify the comment Large samples are better, because the sample mean
tends to be closer to the population mean.
Finding Probabilities Involving Sample Means

As you will see in the next example, you can solve problems involving the sample
mean by using the three properties of the sampling distribution in combination
with what you learned about normal distributions in Chapter 2.
431
22-03-2009 21:11
Lesson
7 de 20
Example: Average Number of Children

What is the probability that a random sample of 20 families in the United States
will have an average of 1.5 children or fewer?
Solution
By looking at the sampling distribution in Display 7.25 for n = 20, you can see
that the sampling distribution of the sample mean is approximately normal. The
mean of the population is 0.9, and the standard deviation of the population is 1.1.
So, for the sampling distribution, we have
Display 7.27 shows a sketch of this situation.
_
= 0.9,
Display 7.27 Sampling distribution of ,when
= 0.25, and n = 20. The shaded area shows the
probability that the sample mean is less than 1.5.
The z-score for the value 1.5 is
You can use a table or a calculator to fi nd that the area under a standard
normal curve to the left of 1.5 (a z-score of 2.4) is about 0.9918, which is the
probability that the sample mean will fall below 1.5. In a random sample of
20 families, it is almost certain that the average number of children per family
will be less than 1.5.
22-03-2009 21:11
Lesson
8 de 20
<
Example: Reasonably Likely Averages

What average numbers of children are reasonably likely in a random sample of
20 families?
Solution
From the preceding example, the distribution of the mean number of children
is approximately normally distributed with mean 0.9 and standard error 0.25.
Reasonably likely outcomes are those within approximately two standard errors
of the mean. This interval is
0.9 1.96(0.25)
which gives an average number of children between 0.41 and 1.39.
Finding Probabilities Involving Sample Means

D7. What is the probability that a random sample of nine U.S. families will
have an average of 1.5 children or fewer? Can this problem be done like the
example for 20 families? Why or why not?
Using the Properties of the Sampling Distribution

of the Mean
At this stage, you must have a few questions about how to use the properties in
the box on page 430.
1. When can I use the property that the mean of the sampling distribution of the
mean is equal to the mean of the population,
Good news. In random sampling, it is always true that
. The shape
of the population doesnt matter, nor does how large the sample is, how
large the population is, or whether you sample with replacement or without
replacement.
2. When can I use the property that the standard error of the sampling distribution
of the mean, , is equal to
?
Good news here too, mostly. You can use this formula with a population of
any shape and with any sample size as long as you randomly sample with
replacement or you randomly sample without replacement and the sample
size is less than 10% of the population size. Almost always, a random sample
is taken without replacement but the population size is more than 10 times
the sample size. So, almost always, you can use this formula. However, in the
rare instance when you are taking a random sample without replacement
from a small population, you should use the formula given in E30.
3. To compute probabilities, I find a z-score and then use the table of standard
normal probabilities on page 824 to get the probability. This doesnt work unless
the sampling distribution of the mean is approximately normal. When can I
assume that the sampling distribution is approximately normal?
Heres where the trouble starts. As you saw in Display 7.25, for example,
when the population is skewed, the sampling distribution also is skewed for
433
22-03-2009 21:11
Lesson
9 de 20
small sample sizes. With a large enough sample, the sampling distribution is
approximately normal. The problem is that there is no hard-and-fast rule to
follow in knowing what sample size is large enough. You will learn some
guidelines in Section 9.3.
For now, if you are told that the population is approximately normally
distributed (or mound-shaped or bell-shaped), you can assume that all
sampling distributions of the mean are approximately normal too, no matter
what the sample size. If you are told that the sample size is very large, its
safe to assume that the sampling distribution of the mean is approximately
normal. If the population looks like the one in Display 7.25, you know the
sampling distribution is approximately normal for samples of size 40 or
larger, and a normal approximation wouldnt be too bad for samples of size
20 or larger.
4. Isnt the size of the population really important? Surely a random sample of
200 households would provide more information about a city of size 20,000
households than about a city with 2 million households.
Not really. As long as the sample was randomly selected and as long as the
population is much larger than the sample, it doesnt matter how large the
population is. Perhaps an extreme example will help. Suppose you want to
know the percentage of all Skittles that are yellow. You start sampling Skittles.
After checking 100,000 pieces, you find that about 20% are yellow. With such
a large random sample, you should have a lot of faith that this estimate is
close to the real population percentage. You shouldnt have less faith in this
estimate if you know that 100 million Skittles have been produced than if you
know 10 million pieces have been produced.
Using the Properties of the Sampling Distribution

of the Mean
D8. If you select 100 households at random, would the standard error of the
sampling distribution of the mean number of children be larger if the
population size, N, is 1,000 or if it is 10,000?
D9. Consider the sampling distribution of the mean of a random sample of size
n taken from a population of size N with mean and standard deviation .
a. For a fixed N, how does the mean of the sampling distribution change as
n increases?
b. For a fixed N, how does the standard error change as n increases?
c. If N = n, what are the mean and standard error of the sampling
distribution?
Finding Probabilities Involving Sample Totals

Sometimes situations are stated in terms of the total number in the sample rather
than the average number: What is the probability that there are 30 or fewer
22-03-2009 21:11
Lesson
10 de 20
children in a random sample of 20 families in the United States? You have the
choice of two equivalent ways of doing this problem.
Method I: Find the equivalent average number of children, x, by dividing the
total number of children, 30, by the sample size, 20:
Then you can use the same formulas and procedure as in the previous
examples.
Method II: Convert the formulas from the previous examples into equivalent
formulas for the sum, and then proceed as described here.
The sum (or total) of a sample is equal to the sample size times the sample
mean:
You can get the mean and the standard error of the sampling distribution
of the sum from those of the sampling distribution of the mean simply by
multiplying by n.
Properties of the Sampling Distribution of the Sum of a Sample

If a random sample of size n is selected from a distribution with mean and
standard deviation , then
the mean of the sampling distribution of the sum is
the standard error of the sampling distribution of the sum is
the shape of the sampling distribution will be approximately normal

if the population is approximately normally distributed; and for other
populations the sampling distribution will become more normal as
n increases
Example: The Probability of 30 or Fewer Children

What is the probability that a random sample of 20 families in the United States
will have a total of 30 or fewer children?
435
22-03-2009 21:11
Lesson
11 de 20
Solution
The sampling distribution of the sum is approximately normal because it has the
same shape as the sampling distribution of the mean. It has mean sum = n =
20(0.9), or 18, and standard error
, or approximately
5. The z-score for a total of 30 children is
Using a calculator or Table A on page 824, the probability that z is less than 2.40 is
about 0.9918.
Example: Reasonably Likely Totals

In a random sample of 20 families, what total numbers of children are reasonably
likely?
Solution
From the preceding example, the sampling distribution of the total number
of children is approximately normal with mean 18 and standard error 5. The
reasonably likely outcomes are those within approximately two standard errors of
the mean. This is the interval
or a total number of children between 8.2 and 27.8.
Finding Probabilities Involving Sample Totals

D10. How do the shapes of the sampling distributions of the sample mean and
of the sample total compare? Show how to convert the plots for the sample
mean in Display 7.25 on page 429 to plots for the sample total.
Summary 7.2: Sampling Distribution of the

Sample Mean
Because averages are so important, it seems fortunate that the mean and standard
error of the sampling distribution of have such simple formulas. It seems doubly
fortunate that, for large sample sizes, the shape is approximately normal. This
allows you to use z-scores and the standard normal distribution to determine
whether a given sample mean is reasonably likely or a rare event. But fortune has
nothing to do with it! Its precisely because these formulas are so simple that the
mean and standard deviation are so important.
Here is a summary of the characteristics of the sampling distribution of
They apply only if the samples are selected at random.
22-03-2009 21:11
Lesson
12 de 20
The mean of the sampling distribution equals the mean of the population, or
and this fact does not depend on the shape of the population or on
the sample size.
The standard error of the sampling distribution equals the population
standard deviation divided by the square root of the sample size, or
and this fact does not depend on the shape of the population
or on the sample size as long as you sample with replacement or your sample
size is no more than 10% of the size of the population.
If the population is normally distributed, the sampling distribution is
normally distributed for all sample sizes n. If the population is not normally
distributed, then as the sample size increases, the sampling distribution will
become more normal. This is called the Central Limit Theorem.
If the population is at least ten times the size of the sample, whether you
sample with or without replacement is of little consequence.
The population size does not have much effect on the analysis unless the
sample size is greater than 10% of the population size.
If you have a situation involving the sum or total in a sample, you can find a
sample mean by dividing the total by n. Alternatively, the sampling distribution of
the sample sum has mean
and standard error
What rule can you use to decide whether the shape of your sampling
distribution is approximately normal? Unfortunately, there isnt onethe required
size of n depends on how close the population itself is to normal and how accurate
you want your approximations to be. You will see more on this topic later.
Practice
Shape, Center,_ and Spread of the Sampling
Distribution of x
P7. The distribution of the population of the
number of motor vehicles per household
and sampling distributions of the mean
for samples of size 4 and 10 are shown in

Display 7.28.
a. Which distribution is which? Make a
rough estimate of the mean and standard
deviation of each distribution.
Display 7.28 Three distributions.
7.2 Sampling Distribution of the Sample Mean 437
22-03-2009 21:11
Lesson
13 de 20
b. The mean of the population is about 1.7.

Theoretically, what is the mean of the
sampling distribution for samples of size
4? Of size 10? Are these computed means
consistent with your estimates of the
means in the histograms?
c. The standard deviation of the population
is about 1. Theoretically, what is the SE
of the sampling distribution for samples
of size 4? Of size 10? Are these computed
SEs consistent with your estimates of the
means in the histograms?
Finding Probabilities Involving

Sample Means
P9. Refer to Display 7.24 on page 428. Suppose a
television network selects a random sample
of 1000 families in the United States for a
survey on TV viewing habits.
a. Describe the distribution of the possible
values of the average number of children
per family.
b. What average numbers of children are
reasonably likely?
c.
What is the probability that the average
d. Compare the shapes of the three
number of children per family will be
distributions. Are the shapes consistent
0.8 or less?
with the Central Limit Theorem?
P10.
Last
January 1, Jenny thought about
P8. From 1910 through 1919, the single-season
buying
individual stocks. Over the next
batting averages of individual Major League
year, the mean of the percentage increases
Baseball players had a distribution that
in individual stock prices is 6.5% and the
was approximately normal, with mean .266
standard deviation of these percentage
and standard deviation .037. Suppose you
increases is 12.8%. The distribution of price
construct the sampling distribution of the
mean batting average for random samples of
increases is approximately normal.
15 players. What are the shape, mean, and
a. If Jenny had picked one stock at random,
standard error of this distribution? [Source:
what is the probability that it would have
Stephen Jay Gould, The Spread of Excellence from Plato to
gone down in price?
Darwin (New York: Harmony Books, 1996).]
b. If Jenny had picked four stocks at
random, what is the probability that
their mean percentage increase would be
negative?
c. If Jenny had picked eight stocks at
random, what is the probability that
their mean percentage increase would be
between 8% and 10%?
d. If Jenny had picked eight stocks at
random, what mean percentage changes
in price are reasonably likely? (In other
words, find the interval that contains the
middle 95% of possible mean percentage
increases in value.)
Using the Properties of the Sampling
Distribution of the Mean
P11. In this problem, you will review sampling
with and without replacement and the rules
of probability that you learned in Chapter 5.
Suppose you take a random sample of two
families from the population in Display 7.29.
22-03-2009 21:11
Lesson
14 de 20
Display 7.29 Population of families and their

numbers of children.
a. If you sample with replacement, what is

the probability that both families have
exactly one child?
b. If you sample without replacement, what
is the probability that both families have
exactly one child?
c. If you sample with replacement, what is
the probability that neither family has
exactly one child?
d. If you sample without replacement, what
is the probability that neither family has
exactly one child?
e. If you sample with replacement, what
is the probability that at least one of the
families has exactly one child?
f. If you sample without replacement, what
is the probability that at least one of the
families has exactly one child?
P12. A typical political opinion poll in the United
States questions about 1500 people. How
would the analysis change if this were a
random sample of 1500 people from Arizona
rather than from the entire United States?
Finding Probabilities Involving
Sample Totals
P13. The distribution of the number of motor
vehicles per household in the United States
is roughly symmetric, with mean 1.7 and
standard deviation 1.0.
a. If you pick 15 households at random,
what is the probability that they have at
least 30 motor vehicles among them?
b. If you pick 20 households at random,

what is the probability that they have
between 25 and 30 motor vehicles among
them?
P14. Refer to Display 7.24 on page 428. Suppose a
television network selects a random sample
of 1000 families in the United States for a
survey on TV viewing habits.
a. Do you think it is reasonably likely that a
sample of 1000 households will produce
at least 1000 children? Explain your
reasoning.
b. Suppose the network changes to a
random sample of 1200 households. Does
this dramatically improve the chances
of seeing at least 1000 children in the
sampled households? Explain.
P15. Suppose the network in P14 repeatedly uses

one of these four sample sizes and computes
the mean number of children per family.
What interval should contain the middle
95% of the sample means? That is, for each
sample size, determine what sample means
are reasonably likely.
a. 25
b. 100
c. 1000
d. 4000 (This is the approximate sample size
for the Nielsen television ratings.)
439
22-03-2009 21:11
Lesson
15 de 20
Exercises
E15. Consider the skewed population shown in
Display 7.30. Display 7.31 shows the three
simulated sampling distributions of the
sample mean for samples of size 2, 4, and 25.
Display 7.30
A skewed population.
a. Match the theoretical summary

information with the correct simulated
sampling distribution (A, B, or C) and
the correct sample size (2, 4, or 25).
I. mean 2.50; standard error 0.48
II. mean 2.50; standard error 1.20
III. mean 2.50; standard error 1.70
b. Does the rule for computing the standard
error of the mean from the standard
deviation of the population appear to
hold in all three situations?
c. Does the sampling distribution appear
to be approximately normal in all cases?
If not, explain how the given shape came
about.
d. For which sample sizes would it be
reasonable to use the rule stating that
about 95% of all sample means lie within
approximately two standard errors of the
population mean?
Display 7.31 Three sampling distributions of the

sample mean (for n = 2, 4, and 25, not
necessarily in that order).
22-03-2009 21:11
Lesson
16 de 20
E16. Consider the M-shaped population in

Display 7.32. Display 7.33 shows three
sample mean for samples of size 2, 4, and 25.
Display 7.32 An M-shaped population.
a. Match the theoretical summary

information with the correct simulated
sampling distribution (A, B, or C) and the
correct sample size (2, 4, or 25).
I. mean 4.50; standard error 1.75
II. mean 4.50; standard error 0.70
III. mean 4.50; standard error 2.47
b. Does the rule for computing the standard
error of the mean from the standard
deviation of the population appear to
hold in all three situations?
c. Does the sampling distribution appear
to be approximately normal in all cases?
If not, explain how the given shape came
about.
d. For which sample sizes would it be
reasonable to use the rule stating that
about 95% of all sample means lie within
approximately two standard errors of the
population mean?
Display 7.33 Three sampling distributions (for

n = 2, 4, and 25, not necessarily
in that order).
441
22-03-2009 21:11
Lesson
17 de 20
E17. Explain how to use a list of random digits

to get a random sample of size 5 from the
population in E15. (You will have to estimate
the percentages in the population from the
histogram.)
E18. Explain how to use a list of random digits
to get a random sample of size 5 from the
population in E16. (You will have to estimate
the percentages in the population from the
histogram.)
E19. Suppose police records in a small city show
that the number of automobile accidents
per day for 1,045 days has the frequency
distribution shown in Display 7.34. The
relative frequencies, in order, are 0.36, 0.37,
0.17, 0.09, and 0.01.
0
Display 7.34 Frequency distribution of the number
of automobile accidents per day.
a. Would it be a rare event to see two

accidents per day in this city?
The two histograms in Display 7.35 show
mean number of accidents per day for
random samples of 4 days and 8 days. Each
distribution has 200 sample means.
b. Which histogram is for samples of 4 days?
Which is for samples of 8 days?
c. Would it be a rare event to see an average
of 1.75 accidents per day over a period of
4 consecutive days in this city?
d. Would it be a rare event to see an average
of 1.75 accidents per day over an 8-day
period in this city?
e. When you use the distributions in
Display 7.35 to answer parts c and d,
what assumptions are you making? Do

they seem reasonable for this situation?
Display 7.35 Simulated sampling distributions

(for n = 4 and n = 8, not necessarily
in that order).
E20. The distribution of grades on the very first

AP Statistics Exam are given in Display 7.36.
Display 7.36 Distribution of exam grades.
a. Construct a plot of the distribution of the

population of scores.
b. The plots in Display 7.37 show simulated
sampling distributions of the mean for
random samples of size 1, 5, and 25.
Match each histogram to its sample size.
22-03-2009 21:11
Lesson
18 de 20
Display 7.37 Simulated sampling distributions

(for n = 1, 5, and 25, not necessarily
in that order).
c. A teacher reports that her class averaged

3.6 on this exam. Do you think she has
a class of 5 students or a class of
25 students? Explain your reasoning.
E21. A particular college entrance examination
has scores that are approximately normally
distributed with mean 500 and standard
deviation 100.
a. If you select one score at random, what is
the probability that it is 510 or greater?
b. If you select four scores at random, what
is the probability that their average is 510
or greater?
c. If you select 25 scores at random, what is
the probability that their average is 510 or
greater?
d. Explain how to use simulation to find the
probability that each of four randomly
selected scores is 510 or greater.
E22. The process for manufacturing a ball bearing
results in weights that have an approximately
normal distribution with mean 0.15 g and
standard deviation 0.003 g.
a. If you select one ball bearing at random,

what is the probability that it weighs less
than 0.148 g?
b. If you select four ball bearings at random,
what is the probability that their mean
weight is less than 0.148 g?
c. If you select ten ball bearings at random,
what is the probability that their mean
weight is less than 0.148 g?
E23. A particular college entrance examination
has scores that are approximately normally
distributed with mean 500 and standard
deviation 100.
a. If you pick 40 scores at random, what is
the probability that their mean will be
within 10 points of the population mean
of 500?
b. How large a random sample of scores
would you need in order to be 95% sure
that the sample mean would be within
10 points of the population mean of 500?
E24. Last January 1, Jenny thought about
buying individual stocks. Over the next
year, the mean of the percentage increases
in individual stock prices is 6.5% and the
standard deviation of these percentage
increases is 12.8%. The distribution of
price increases is approximately normal.
a. If Jenny picked ten stocks at random,
what is the probability that her mean
percentage increase in price is more
than 7%?
b. What is the smallest number of stocks
Jenny should pick at random so that the
probability is 95% or more that her mean
percentage increase in price will be more
than 5%?
E25. An emergency room in Oxford, England,
sees a mean of 67.4 children ages 715
per summer weekend, with standard
deviation 10.4. The numbers of children
are approximately normally distributed.
[Source: Stephen Gwilym et al., Harry Potter Casts a Spell
on Accident Prone Children, British Medical Journal 331
(December 2005): 15056.]
a. How many children in all would you

expect to be seen in the emergency room
443
22-03-2009 21:11
Lesson
19 de 20
over two randomly selected summer

weekends?
b. Find the probability that the emergency
room will see a total of 73 or fewer
children over two randomly selected
summer weekends.
c. Over the two summer weekends when
Harry Potter books were released, a total
of 73 children were seen in the emergency
room. What can you conclude?
E26. In a recent year, 16,597,552 people were
admitted to riverboat casinos in Illinois.
The mean casino win (customer loss) per
admission was $103.02. Assume that the
standard deviation might be about $100.
mound-shaped, with a mean of 4.3 and a

standard deviation of 1.4.
a. What is the smallest number of addresses
that should be drawn in order to have
a 98% chance that there will be at least
100 people total in the households? (You
can use your calculator to estimate the
solution to the resulting equation to the
nearest whole number of names.)
b. The cost of one free admission for a year
is $250. If the director draws the number
of addresses you recommended in part a,
how much will he have to pay because of
his poor planning?
E28. Display 7.38 gives the distribution of the
[Source: Illinois Statistical Abstract 2004, Table 27-19;
number of households in the United States
www.igpa.uillinois.edu.]
that own various numbers of color televisions.
a. What were the total winnings for the
Suppose a government agency wants to check
riverboat casinos that year?
1000 color televisions to assess whether
televisions in homes have a tendency to
b. If you randomly select 100 people who
overheat. How large a random sample of
were admitted, what is the probability
households should the agency take to have a
that the total winnings for the casinos
95% chance of getting 1000 color televisions
from these people was over $11,000?
or more among the households in its sample?
c. If you randomly select 100 people who
were admitted, what is the probability
that the total winnings for the casinos
from these people was negative (i.e., the
group as a whole won money)?
E27. The advertising agency for a new familyoriented theme park has run a promotion
that promises 100 local people free
admission for a year. In preparation for
a public drawing, the address of each
household in the community is placed in a
Display 7.38 Number of color televisions per
bin. Winning addresses will be drawn, and
household. [Source: U.S. Census Bureau,
each member of the household at those
Statistical Abstract of the United States, 2006,
addresses will get free admission for a year.
Table 963.]
The director of the advertising agency
E29. As the sample size n increases, describe what
suddenly realizes a flaw in his scheme: He
happens to the mean and standard error of
doesnt know how many addresses to draw
the sampling distribution of
in order to end up with 100 people. (The
number of people in each household wont
a. the sample mean
b. the sample total
be known until the winners are notified.)
E30. If you have a small population and you
His budget allows for 100 free admissions
sample without replacement, the formula
and he doesnt want to go over this, as it
for the standard error of the sampling
will cut into his profit. He finds from census
distribution of the mean must be adjusted:
data that the distribution of the number of
people per household in the community is
22-03-2009 21:11
Lesson
20 de 20
a. Does multiplying by
increase or
decrease the SE? Why is that a sensible
thing to do?
In the rest of this exercise, you will see how
to use this formula.
You are allowed to drop your two lowest
grades on the six tests in your statistics
class. There will be no make-up tests. If you
showed up for all six tests, your scores would
have been 55, 65, 75, 80, 90, 95.
b. What are the mean, , and standard
deviation, , of this population of test
scores?
c. Suppose you pick two test days at random
to miss class. You get a 0 on these two
tests, and these two 0s will be the grades
you drop. Construct the exact sampling
distribution of your mean test score for
the remaining four tests.
d. Verify that the mean of the sampling
distribution is equal to the mean of the
population of test scores, that is,
e. The basic formula for the standard
deviation of a population is
Use this formula to compute the standard

error of the sampling distribution from
part c. Note that there are 15 values in
this sampling distribution, so n 15.
f. Compute the standard error (SE) of the
sampling distribution using the two
short-cut formulas:
g. Which formula in part f gives the correct

SE?
Use this information for E31E34: In Section 6.1,
you learned rules for finding the mean and
variance when you add or subtract two random
variables in a probability distribution (see the box
on page 372). The same rules apply when working
with sampling distributions. Here the rules are
stated in terms of sampling distributions.
Properties of the Sampling

Distribution of the Sum and Difference
Suppose two values are taken randomly from
two populations with means 1 and 2 and
variances
and
, respectively. Then the
sampling distribution of the sum of the two
values has mean
If the two values were selected independently,
the variance of the sum is
The sampling distribution of the difference

of the two values has mean
If the two values were selected independently,
the variance of the difference is
The shapes of the sampling distributions

of the sum and the difference depend on the
shapes of the two original populations. If both
populations are normally distributed, so are
the sampling distributions of the sum and the
difference.
E31. In 2006, SAT critical reading scores had a
mean of 503 and a standard deviation of
113. The SAT math scores had a mean of
518 and a standard deviation of 115. Each
distribution was approximately normal.
[Source: 2006 College Bound Seniors: Total Group Profile
Report, www.collegeboard.com.]
a. Suppose one SAT critical reading score

is selected at random and one SAT math
score is independently selected at random
and the scores are added. Find the mean
and standard error of the sampling
distribution of this sum.
b. Find the probability that the sum of the
scores from part a is less than 800.
c. What total SAT scores are reasonably
likely?
445
22-03-2009 21:11
Lesson
1 de 12
d. What is the probability that the critical

reading score is at least 100 points higher
than the math score?
e. Suppose one student is selected at
random and his or her SAT critical
reading score and math score are
added. If you can, describe the mean
E32. The height of males ages 2029 in the United
States is approximately normally distributed
with mean 69.3 inches and standard deviation
2.92. The height of females ages 2029 in
the United States is approximately normally
distributed with mean 64.1 inches and
standard deviation 2.75. [Source: U.S. Census Bureau,
a. Suppose one male from this age group

is selected at random and one female
is independently selected at random
and their heights added. Find the mean
b. Find the probability that the sum of the
heights is less than 125 inches.
c. What total heights are reasonably likely?
d. What is the probability that the male is at
least 2 inches taller than the female?
e. Suppose that one male is selected at
random and his height is added to that
of his closest female relative aged 20
to 29. If you can, describe the mean
7.3
E33. In the case where n values are taken at

random and independently from the same
population with mean and standard
deviation , the sampling distribution of
their sum has mean + + ... + , or n
and variance 2 + 2 + ... + 2, or n 2.
a. If you roll a die three times, what is the
mean of the sampling distribution of
the sum? The variance?
b. If you roll a die seven times, what is the
the sum? The variance?
c. In 2006, SAT critical reading scores had
a mean 503 and standard deviation 113.
If 20 scores are selected at random, what
are the mean and variance of their sum?
What is the probability that the sum is
less than 10,000?
E34. F. Paca weighed rams in Lima, Peru,
and found that the distribution was
approximately normal with mean 75.4 kg
and standard deviation 7.38 kg. [Source: F. Paca
(New Mexico State University, 1977).]
a. Suppose two rams are selected at random.

Describe the sampling distribution of the
sum of their weights.
b. What is the probability that the sum of
their weights is less than 145 kg?
c. What is the probability that the sum of the
weights of ten randomly selected rams is
more than 750 kg? (Use the rules in E33.)
d. What is the probability that the first ram
selected is more than 5 kg heavier than
the second ram selected?
Sampling Distribution of the Sample Proportion

You now will move from studying the behavior of sample means to studying the
behavior of sample proportions (the proportion of successes in the sample). You
will see that the properties of sample proportions parallel those of sample means.
Sampling Distribution of the Number of Successes

Before moving on to sample proportions, you will review what you have
learned so far about the number of successes you can expect in a binomial
22-03-2009 21:11
Lesson
2 de 12
(success/failure) situation. The box and example summarize relevant information

from Section 6.2.
Properties of the Sampling Distribution of the Number
of Successes
If a random sample of size n is selected from a population with proportion of
successes p, then the sampling distribution of the number of successes X
has mean
has standard error
will be approximately normal as long as n is large enough
As a guideline, if both np and n(1 p) are at least 10, then using the
normal distribution as an approximation to the shape of the sampling
distribution will give reasonably accurate results.
Example: Probability of 30 or More Mississippians Wearing Seat Belts

The use of seat belts continues to rise in the United States, with overall seat belt
usage of 82%. Mississippi lags behind the rest of the nationonly about 60%
wear seat belts. [Source: National Highway Traffic Safety Administration, November 2005,
www.nhtsa.dot.gov.] Suppose you take a random sample of 40 Mississippians. How
many do you expect will wear seat belts? What is the probability that 30 or more
of the people in your sample wear seat belts?
Solution
You expect that 60% of the 40, or 24, Mississippians will be wearing seat belts.
To estimate the probability that 30 or more wear seat belts, you will use the
normal approximation to the binomial distribution. However, first you must
check the guideline that both np and n(1 p) are at least 10. This ensures that the
binomial distribution is sufficiently mound-shaped for the normal distribution to
be a reasonable model.
Because 60% of the population use seat belts, both np and n(1 p) are 10 or
greater:
Thus, the sampling distribution of the number of successes is approximately

normal, with mean and standard error
This distribution is shown in Display 7.39.
447
22-03-2009 21:11
Lesson
3 de 12
Display 7.39 Exact sampling distribution and normal

approximation of the number of successes when
p = 0.6 and n = 40.
The z-score for 30 successes is
The proportion below this value of z is 0.9736, so the probability of getting

30 or more successes is 1 0.9736, or 0.0264.

If you take a random sample of 40 Mississippians, what is the probability
that 75% or more wear seat belts? This is the same question answered in the
previous example, because 30 out of 40 is 75%. However, the notation (and
the computations) will reflect that you now are working with the proportion of
successes in the sample rather than the number of successes.
Some notation will make your work more efficient. Suppose you count the
number of successes in a random sample of size n from a population in which the
true proportion of successes is given by p. The symbol for the sample proportion
is (read p-hat). That is,
For example, suppose your sample of automobile drivers contains 26 who use seat
belts. Then n = 40, p = 0.60, and = = 0.65.
In Activity 7.3a, you will construct a simulated sampling distribution of the
proportion of drivers wearing seat belts in samples of size 10, 20, and 40.
Buckle Up!
What youll need: a table of random digits
1. Describe how to use a table of random digits to select a random sample of
size 10 from a population with 60% successes.
2. Use your method to select a random sample of size 10. Count the number
of successes, and compute the sample proportion, .
(continued)
22-03-2009 21:11
Lesson
4 de 12
3. Repeat step 2 until your class has 100 samples of size 10.
4. Plot your 100 sample proportions on a dot plot.
5. Calculate the mean and standard error of your simulated sampling
distribution of the sample proportion.
6. Repeat steps 2 through 5 for samples of size 20 and size 40.
[See Calculator Note 7B to learn how to use a calculator to sample for this activity.]
Display 7.40 shows the exact sampling distributions for samples of size 10, 20,
and 40 drawn from a population with 60% successes. They should be similar in
shape, center, and spread to your dot plots from Activity 7.3a.
Display 7.40 Exact sampling distributions of

p = 0.60.
for samples of size 10, 20, and 40 when

D11. Use Display 7.40 to answer these questions.
a. Which distribution is most normal in shape? Least normal? Does
the guideline that both np and n(1 p) are at least 10 work well in
predicting approximate normality? (Approximately normal means a
histogram that is symmetric and mound-shaped and has enough bars
that the shape is fairly smooth rather than lumpy.)
b. About 60% of Mississippians wear seat belts. What proportion of
Mississippians would you expect to be wearing seat belts in a sample of
size 10? Size 20? Size 40?
c. Which distribution in Display 7.40 has the largest spread? The smallest?
d. Would you be most likely to find that 75% or more of Mississippians in
the sample wear seat belts if the sample size is 10, 20, or 40?
Shape, Center, and Spread for Sample Proportions

Dividing by the sample size, n, to change from the number of successes to the
proportion of successes doesnt change the shape of the sampling distribution.
Thus, you still can use the guideline that if both np and n(1 p) are at least 10,
then the sampling distribution is approximately normal.
449
22-03-2009 21:11
Lesson
5 de 12
Finding the formulas for the mean, , and the standard error, , of the
sampling distribution of the sample proportion is simple: Just as you divide by n
to convert the number of successes to the proportion of successes, you divide by
n to convert the formulas for the mean, , and standard error, , of the number
of successes to the equivalent formulas for the proportion of successes:
These formulas tell you two facts about the sampling distribution of :
The mean does not change depending on the sample size. No matter how
large the sample size, the mean stays at p.
The spread decreases as the sample size increases.
The properties of the sampling distribution of sample proportions are
summarized in this box, followed by an example showing how to use them.
Properties of the Sampling Distribution of the Sample Proportion
If a random sample of size n is selected from a population with proportion of
successes p, then the sampling distribution of p has these properties:
The mean of the sampling distribution is equal to the mean of the
population, or
The standard error of the sampling distribution is equal to the standard
deviation of the population divided by the square root of the sample size:
As the sample size gets larger, the shape of the sampling distribution
becomes more normal and will be approximately normal if n is large
enough.
As a guideline, if both np and n(1 p) are at least 10, then using the normal
distribution as an approximation of the shape of the sampling distribution will
give reasonably accurate results.
Example: Describing the Sampling Distribution of a Sample Proportion

Drivers in the Northeast and Mid-Atlantic states had the highest failure rate, 20%,
on the GMAC Insurance National Drivers Test. (They also were the drivers most
likely to speed.) [Source: Insurance Journal, www.insurancejournal.com.] Describe the shape,
center, and spread of the sampling distribution of the proportion of drivers who
would fail the test in a random sample of 60 drivers from these states. What are
the reasonably likely proportions of drivers who would fail the test?
22-03-2009 21:11
Lesson
6 de 12
Solution
Both np = 60(0.2) = 12 and n(1 p) = 60(0.8) = 48 are at least 10, so the
sampling distribution is approximately normal.
The mean, p, is 0.2. The SE is
Thus, the reasonably likely proportions are 0.2 1.96(0.05), or between

about 0.1 and 0.3. The histogram in Display 7.41 shows the exact distribution and
normal approximation.
Display 7.41 Exact sampling distribution and normal

approximation for p = 0.2 and n = 60.
Shape, Center, and Spread for Sample Proportions

D12. Look back at Display 7.40 on page 449.
a. Compute the mean and standard error for each distribution (n = 10, 20,
and 40) using the formulas from this section.
b. Estimate the means and standard errors simply from looking at the
histograms. Do the computed means and standard errors in part a match
your estimates?
c. The table in Display 7.42 gives the exact sampling distribution for
samples of size 10, with p = 0.6. Compute the mean and SE using the
formulas from Section 6.1 given on the next page. Do these formulas give
the same mean and SE as the formulas you used in part a?
Display 7.42 Exact sampling distribution for samples of size 10,

with p = 0.6.
451
22-03-2009 21:11
Lesson
7 de 12
The expected value (mean) X and variance

distribution listed in a table are
of a probability
where pi is the probability that the random variable takes on the specific
value xi. The standard deviation is the square root of the variance.
Finding Probabilities Involving Proportions

You can now find answers to specific questions by using your knowledge of the
properties of the sampling distribution of a proportion.
Example: Using the Properties of a Sampling Distribution of a Proportion

to Find Probabilities
About 60% of Mississippians use seat belts. Suppose your class conducts a survey
of 40 randomly selected Mississippians.
a. What is the chance that 75% or more of those selected wear seat belts?
b. Would it be quite unusual to fi nd that fewer than 25% of the Mississippians
selected wear seat belts?
Solution
From the developments in this section, you know that
distribution with mean and standard error
has a sampling
Furthermore, you can consider the sampling distribution to be approximately

normal because np and n(1 p) are both at least 10:
np = 40(0.6) = 24 and n(1 p) = 40(1 0.6) = 16
This distribution is sketched in Display 7.43.
Display 7.43 Exact and approximate sampling distribution of

for p = 0.6 and n = 40.
22-03-2009 21:11
Lesson
8 de 12
Using the normal

distribution as an
approximation
a. The z-score for the value
= 0.75 in this distribution is
The proportion below this value of z is 0.9735. It follows that the probability
of getting 75% or more Mississippians who wear seat belts in a sample of size
40 is only 1 0.9735, or 0.0265.
b. The sample proportion, 0.25, is about 4.5 standard deviations below the mean
of the sampling distribution, so such a result would be unusual indeed!
Summary 7.3: Sampling Distribution

of the Sample Proportion
Sample proportions, like sample means, are among the most common summary
statistics in practical use, and knowledge of their behavior is fundamental to the
study of statistics.
The formulas in this section about the proportion of successes are derived
from those in Section 6.2 about the number of successes. Either set of formulas
can be used in ways parallel to those in Section 7.2 for the sample mean. You
compute a z-score and use the normal distribution to approximate the probability
of getting specified results from a random sample. To do this, you use these facts:
The samples must be selected randomly.
The mean of the sampling distribution of , is equal to the proportion p of
successes in the population, or = p.
The standard deviation of the sampling distribution of , also called the
standard error of a sample proportion, is given by the formula
Alternatively, the sampling distribution of the number of successes has mean

and standard error , where
The shape of the sampling distribution becomes more normal as n increases.
You can assume it is normal for computing probabilities as long as np and
n(1 p) are both at least 10.
Practice
Sampling Distribution of the Number
of Successes
P16. A survey of hundreds of thousands of college
freshmen found that 63% believe dissent is
a critical component of the political process.
Suppose you take a random sample of

100 of the freshmen surveyed. What is the
probability that you will nd that between
56 and 70 of the freshmen in your sample
believe this?
[Source: Higher Education Research Institute, UCLA,

The American Freshman, National Norms for Fall 2005.]
453
22-03-2009 21:11
Lesson
9 de 12
Sampling Distribution of the Sample

Shape, Center, and Spread for Sample
Proportion
Proportions
P17. The GMAC Insurance National Drivers Test
P18. Refer to the sampling distributions in P17.
found that about 10% of drivers ages 16 to 65
a. Compute the mean and standard error
would fail a written drivers test if they had to
for each distribution (n = 10, 20, 40, and
take one today. [Source: Insurance Journal, 2005, www
100) using the formulas in this section.
.insurancejournal.com.] The sampling distributions
b. Estimate the means and standard errors
in Display 7.44 give the proportion of drivers
simply from looking at the histograms.
who would fail for samples of size 10, 20, 40,
Do the computed means and standard
and 100 taken from a population like this
errors in part a match your estimates?
one, where p = 0.10.
c. The table in Display 7.45 gives the exact
sampling distribution for samples of size
10, with p = 0.1. Compute the mean
and SE using the formulas given in D12
on page 452. Do these formulas give the
same mean and SE as the formulas you
used in part a?
for
Display 7.44 Exact sampling distributions of
samples of size 10, 20, 40, and 100
when p = 0.10.
a. Which distribution is most normal in

shape? Least normal? Does the guideline
that both np and n(1 p) should be
at least 10 work well in predicting
approximate normality?
b. What proportion of drivers would you
expect to fail a driving test in a sample of
size 10? Size 20? Size 40? Size 100?
c. Which distribution has the largest
spread? The smallest?
d. Would you be most likely to find that
20% or more of a sample of drivers fail a
driving test if the sample size is 10, 20, 40,
or 100?
Display 7.45 Exact sampling distribution for n = 10

and p = 0.1.
P19. In the 2000 U.S. Census, 53% of the

population over age 30 were women.
[Source: U.S. Census Bureau, www.census.gov.]
a. Describe the shape, mean, and standard

error of the sampling distribution of p for
random samples of size 100 taken from
this population. Make an accurate sketch,
with a scale on the horizontal axis of this
distribution.
22-03-2009 21:11
Lesson
10 de 12
b. To be a member of the U.S. Senate, you

must be at least 30 years old. In 2000, 9 of
the 100 members of the U.S. Senate were
women. Is this a reasonably likely event if
gender plays no role in whether a person
becomes a U.S. Senator?
Finding Probabilities Involving Proportions

P20. In 2005, about 31% of the citizens of the
United States were Republicans. [Source: Harris
poll, March 9, 2005, www.harrisinteractive.com.]
a. If 435 citizens are selected at random,

what is the probability of getting 53% or
more who are Republicans?
b. In 2006, the U.S. House of Representatives
had 435 members, of whom 231 were
Republicans. Is this percentage higher
than could reasonably occur by chance?
If so, what are some possible reasons for
this unusually large sample percentage?
[Source: clerk.house.gov.]
P21. About 60% of Mississippians wear seat belts.

What proportion of seat belt users would
be reasonably likely to occur in a random
sample
a. of 40 drivers?
b. of 100 drivers?
c. of 400 drivers?
Exercises
E35. The ethnicity of about 92% of the population
of China is Han Chinese. Suppose you take
a random sample of 1000 Chinese. [Source: CIA
World Factbook.]
a. Make an accurate sketch, with a scale

on the horizontal axis, of the sampling
distribution of the proportion of Han
Chinese in your sample.
b. Make an accurate sketch, with a scale
distribution of the number of Han
Chinese in your sample.
c. What is the probability of getting 90% or
fewer Han Chinese in your sample?
d. What is the probability of getting 925 or
more Han Chinese?
e. What numbers of Han Chinese would be
rare events? What proportions?
E36. According to the U.S. Census Bureau,
22.3 percent of the Spanish-surnamed
population in the United States have one
of these surnames: Garcia, Martinez,
Rodriguez, Lopez, Hernandez, Gonzalez,
Perez, Sanchez, Rivera, Ramirez, Torres,

Gonzales. Suppose you take a random
sample of 500 Spanish-surnamed people in
the United States. [Source: David L. Word and R. Colby
Perkins, Jr., Building a Spanish Surname List for the 1990s
A New Approach to an Old Problem, Technical Working Paper
no. 13, March 1996.]
a. Make an accurate sketch of the sampling

distribution of the proportion of people
in your sample who have one of these
surnames.
b. Make an accurate sketch of the sampling
distribution of the number of people
in your sample who have one of these
surnames.
c. What is the probability of getting 20% or
fewer with one of these surnames in your
sample?
d. What is the probability of getting 105 or
more people with one of these surnames?
e. What numbers of people with one of
these surnames would be rare events?
What proportions?
455
22-03-2009 21:11
Lesson
11 de 12
E37. Refer to the situation in E35. This time,

c. Of the 50 Westvaco employees listed in
suppose you take a random sample of
Display 1.1 on page 5, 10 were under
100 Chinese rather than 1000.
the median age. Is this about what you
would expect from a random sample
a. Describe how the shape, center, and
of 50 residents of the United States, or
spread of the sampling distribution of
should you conclude that this group is
the proportion of Han Chinese in your
special in some way? If you think the
sample will be different from your sketch
group is special, what is special about it?
in E35, part a.
E40. In fall 2004, 37% of the 38,859 first-year
b. Describe how the shape, center, and
students attending the California State
spread of the sampling distribution of the
University system needed remedial work
number of Han Chinese in your sample
in mathematics. [Source: California State University,
will be different from your sketch in E35,
www.asd.calstate.edu.]
part b.
a. Suppose you select 136 students at
c. With a sample of size 100, will the
random from this population of students.
probability of getting 90% or fewer
Make an accurate sketch, with a scale
Han Chinese in your sample be larger
or smaller than the probability you
distribution of the number of students
computed in E35, part c? Explain.
who need remedial work.
E38. Refer to the situation in E36. This time,
b.
What is the probability that 68 or fewer
suppose you take a random sample of
in a random sample of 136 students need
100 Spanish-surnamed Americans rather
remedial work?
than 500.
c. Suppose you select 2850 students at
a. Describe how the shape, center, and
random. Make an accurate sketch, with
spread of the sampling distribution of
a scale on the horizontal axis, of the
the proportion of people in your sample
sampling distribution of the proportion
with one of the given surnames will be
who need remedial work.
different from your sketch in E36, part a.
d.
What is the probability of getting 54%
b. Describe how the shape, center, and
or
more who need remedial work in a
spread of the sampling distribution of the
random
sample of 2850 students?
number of people with one of the given
e. Of the 2850 students entering California
surnames will be different from your
State University, Northridge, 54% needed
sketch in E36, part b.
remedial work. Is this result about what
c. With a sample of size 100, will the
you would expect from a random sample,
probability of getting 20% or fewer with
or should you conclude that this group is
one of the given surnames be larger
special in some way?
or smaller than the probability you
computed in E36, part c? Explain.
E39. In 1991, the median age of residents of the
United States was 33.1 years. [Source: U.S. Census
Bureau, www.census.gov.]
a. What is the probability that one person,

selected at random, will be under the
median age?
b. In a random sample of 50 people from
the United States, what is the probability
of getting 10 or fewer under the median
age?
22-03-2009 21:11
Lesson
12 de 12
E41. About 60% of married women are employed.

If you select 75 married women, what is the
probability that between 30 and 40 of them
are employed? What assumptions underlie
your computation? [Source: U.S. Census Bureau,
Statistical Abstract of the United States, 2006.]
E42. Suppose 80% of a certain brand of computer

disk contain no bad sectors. If 100 such disks
are inspected, what is the approximate chance
that 15 or fewer contain bad sectors? What
assumptions underlie this approximation?
E43. The histograms in Display 7.46 are sampling
distributions of for samples of size 5, 25,
and 100, first for a population with p = 0.2
and then for a population with p = 0.4.
a. Do the means of the sampling
distributions depend on p? On n?
b. How do the spreads of the sampling
distributions depend on p and n?
c. How do the shapes of the sampling
distributions depend on p and n?
d. For which combination(s) of p and n
would you be willing to use the rule that
roughly 95% of the values lie within two
standard errors of the mean?
Display 7.46 Sampling distributions of

and 100.
E44. The guideline states that it is appropriate

to use the normal distribution as an
approximation for the sampling distribution
of a sample proportion if both np and
n(1 p) are greater than or equal to 10. To
check this out, generate simulated sampling
distributions for values of p equal to 0.90
and 0.98. Use sample sizes of 50, 100, and
500 with each value of p. Does the guideline
appear to be reasonable?
E45. In this exercise you will learn why the
introduction to this section said that the
properties of sample proportions parallel
the properties of sample means in the
previous section. Recall that 60% of
Mississippians use seat belts. Imagine the
population of Mississippians as consisting
of a barrel containing one piece of paper per
Mississippian. Those Mississippians who
wear seat belts are represented by a piece
of paper with the number 1 on it. Those
Mississippians who dont wear seat belts
are represented by a piece of paper with the
number 0 on it. Display 7.47 (on the next
page) shows this population.
for p = 0.2 and p = 0.4 for samples of size n = 5, 25,
457
22-03-2009 21:11
Lesson
1 de 8
Display 7.47 Relative frequency histogram for a

population with p = 0.6.
a. Instead of taking a random sample of

40 actual Mississippians, you take a
random sample of 40 slips of paper
out of this barrel. Suppose you get
26 slips of paper with a 1 on them and
14 slips of paper with 0 on them.
Compute the sample mean using the
formula
How does this mean
compare to the proportion of successes,
, in the sample? Why is that the case?
b. Find the mean of this population (Display
7.47) using the formula
Compute the mean using the formula in
the box on page 450. How do the means
compare?
c. Describe what you have learned in this
exercise about the similarities of a sample
proportion and a sample mean.
E46. Refer to E45, where a population with 60%
successes is thought of as a population with
60% 1s and 40% 0s.
a. Compute the standard error of the
population in Display 7.47 using the
formula
b. Compute the standard error of the
population in Display 7.47 using the
formula in the box on page 450. How
does this compare with the standard error
you computed in part a?
Chapter Summary
In this chapter, you have learned to create a sampling distribution of a summary
statistic. To create an approximate sampling distribution using simulation, you
first define a process for taking a random sample from the given population.
You then take this random sample and compute the summary statistic you
are interested in. Finally, you generate a distribution of values of the summary
statistic by repeating the process. For some simple situations, you can list all
possible samples and get the exact distribution of the summary statistic.
Some summary statistics have easily predictable sampling distributions.
If a random sample of size n is taken from a population with mean and
variance 2 , the sampling distribution of the sample mean, , has mean and
standard error
The sampling distribution of the sum of the values in the sample has mean
and standard error
These formulas are summarized in Display 7.48.
22-03-2009 21:12
Lesson
2 de 8
If a random sample of size n is taken from a population with percentage of

successes p, the sampling distribution of the sample proportion, , has mean
and standard error
The sampling distribution of the number of successes has mean np and

standard error
These formulas are summarized in Display 7.48.
In the case of means and sums, the sampling distribution becomes more and
more normal in shape as the sample size increases. This is called the Central Limit
Theorem. If the population is normally distributed to begin with, the sampling
distributions of means, sums, and differences are normal for all sample sizes.
For proportions, you can use the normal distribution as an approximation to the
sampling distribution as long as np and n(1 p) are both at least 10.
It might still be unclear to you why sampling distributions are so important.
If so, it might help you to review the discussion of the Westvaco case at the
beginning of Section 7.1. That sampling distribution helped you decide whether
the ages of the workers laid off could reasonably be attributed to the variation that
occurs from random sample to random sample, or whether you should investigate
if some other mechanism is at work.
Theres another reason why sampling distributions are so important. Suppose
a pollster takes a random sample of 1500 voters to find out what percentage p
of voters approve of some issue. The proportion in the sample who approve
almost certainly wont be exactly equal to the proportion p in the population
who approvethis is called sampling error. How large is the sampling error
likely to be? Thats what sampling distributions can tell us. That might seem a bit
backward. In this chapter, you always knew p and then found the reasonably likely
values of . But the pollster has and wants to find plausible values of p. This isnt
hard to do now that you know what values of tend to come from which values of
p, but the details will have to wait until Chapter 8.
Display 7.48 Formulas for the mean and standard error of

sampling distributions.
Chapter Summary 459
22-03-2009 21:12
Lesson
3 de 8
Review Exercises
E47. China Daily, an English-language newspaper
published in the Peoples Republic of China,
reported the Air Pollution Index (API) for
32 major Chinese cities (Display 7.49).
b. Find the mean and standard error of

the sampling distribution of the sample
mean, , for samples of size 3.
c. Can the sampling distribution be
considered approximately normal?
E48. Refer to the situation in E47 of selecting
three Chinese cities at random from the list
of 32 cities. Air quality is considered good
if the API is less than 100.
a. Describe how to use simulation to
construct the sampling distribution of the
number of cities with good air quality for
randomly selected samples of three cities.
b. Find the mean and standard error of the
sampling distribution of the number of
cities with good air quality for samples
of size 3.
c. Can the sampling distribution be
considered approximately normal?
E49. About 68% of the people in China live in
rural areas. Suppose a random sample of
200 Chinese people is taken.
Display 7.49 API for Chinese cities. [Source: China Daily,

March 13, 1999.]
The mean of this distribution, shown in

the dot plot in Display 7.50, is 140.9, with
standard deviation 119.4.
Display 7.50 Dot plot of the Air Pollution Index (API)

for 32 major Chinese cities.
Suppose China had the resources to monitor

air quality in only three of these cities.
a. Describe how to use simulation to
construct the sampling distribution
of the sample mean, , for randomly
selected samples of three cities, without
replacement. Take one such random
sample and compute .
a. Describe the shape, mean, and standard

error of the sampling distribution of ,
the proportion of people in the sample
who live in rural areas.
b. What is the probability that 75% or more
in the sample live in rural areas?
c. What values of would be rare events?
d. What is the probability that 130 or more
in the sample live in rural areas?
22-03-2009 21:12
Lesson
4 de 8
E50. The spine widths of the books in a particular

library have a distribution that is skewed
slightly to the right, with mean 4.7 cm and
SD 2.1 cm.
a. If you select 50 books at random from
this library, what is the probability that
they will fit on a shelf that is 240 cm in
length?
b. Does your answer to part a imply that
the 50 books in the philosophy section
probably will fit on one 240-cm shelf?
E51. The magazine Forbes found that the mean
age of the chief executive officers (CEOs)
of Americas 500 largest companies is about
55.7 years, with standard deviation 6.8 years.
The ages are roughly mound-shaped, as
Display 7.51 Frequency histogram for ages of CEOs.

[Source: Forbes, 2005, www.forbes.com.]
a. What do you find interesting about this

histogram?
b. If you select ten of these CEOs at random,
what is the probability that their average
age is between 55 and 60?
E52. W. J. Youden (190071) weighed many new
pennies and found that the distribution is
approximately normal with mean 3.11 g
and standard deviation 0.43 g. What are
the reasonably likely weights of a roll of
50 randomly selected new pennies? [Source:
W. J. Youden, Experimentation and Measurement (National
Bureau of Standards, 1984), pp. 1079.]
E53. Young men have an average height of about

68 in. with standard deviation 2.7 in. Young
women have an average height of about
64 in. with standard deviation 2.5 in. Both
distributions are approximately normal. A
young man and a young woman are selected

at random.
a. Describe the sampling distribution of the
difference of their heights.
b. Find the probability that he is 2 in. or
more taller than she is.
c. What is the probability that she is taller
than he is?
E54. Two buses always arrive at a bus stop between
11:13 and 11:23 each day. Over a sample of
5 days, the average time between the times
the two buses arrived was 7 minutes. Are the
buses arriving at random in the 10-minute
interval? Follow these steps to arrive at an
answer you think is reasonable.
a. The population of interest is the set of
possible times between the two arrivals
(interarrival times). Describe a way to
generate five of these interarrival times,
assuming the two buses arrive at random
and independent times during the
10-minute interval. Often take such a
sample.
b. What was the mean interarrival time for
your sample of size 5?
c. The dot plot in Display 7.52 shows the
results of 100 repetitions of the process
in parts a and b. What is the largest
mean interarrival time recorded? Give an
example of five pairs of arrival times that
would give that mean.
Display 7.52 Dot plot of the mean difference

between arrival times for two buses
for samples of 5 days.
d. Based on the sampling distribution, do

you think its reasonable to conclude
that the buses in question were arriving
at random?
Chapter Summary 461
22-03-2009 21:12
Lesson
5 de 8
E55. You select two digits at random from

0 through 9 (with replacement) and take
their average. Your opponent also selects
two digits at random and takes their average.
You win an amount equal to the difference
between your average and your opponents
average. (If the difference is negative, you
lose that amount.)
a. Describe how to simulate a distribution
of the amount you win per play of the
game. Do this simulation three times and
record the differences.
b. Describe the distribution of all possible
outcomes.
E56. In Chapter 4, you read about Kelly
Acamporas hamster experiment. The four
hamsters raised in short days had enzyme
levels
12.500 11.625 18.275 13.225
The hamsters raised in long days had enzyme
levels
6.625 10.375 9.900 8.800
In this exercise, youll use a sampling
distribution to decide whether this is
persuasive evidence that hamsters raised in
short days have higher enzyme levels than
hamsters raised in long days. Your summary
statistic for Kellys experiment will be d,
the difference of the mean enzyme level of
hamsters raised in short days and the mean
enzyme level of hamsters raised in long days.
a. What is the value of the summary statistic
d for Kellys hamsters?
b. Suppose the length of a day makes no
difference in enzyme levels, that is,
suppose the eight numbers would have
been the same if the hamsters had all
received the opposite treatment. Use
simulation to construct an approximate
sampling distribution of all possible values
of d. In other words, assign the hamsters
at random so that four get each treatment
but the enzyme level for the hamster is the
same no matter what treatment it gets.
c. Is Kellys value of d a reasonably likely

outcome if the length of a day makes no
difference? What can you conclude?
E57. Suppose you take a sample of size n with
replacement from a binomial population.
As n increases, describe what happens to
the shape, mean, and standard error of the
sampling distribution of
a. the sample proportion
b. the number of successes in the sample
E58. The histogram in Display 7.53 shows the
number of airline passengers (in thousands)
departing from or arriving at the 30 largest
world airports during a recent year. The
only outlier is Atlanta, with 83,606,583
passengers. The mean of this population
is 42,520, and the standard deviation
is 14,758.
Display 7.53 Histogram of passengers for the

30 largest world airports. [Source: Airports
Council International, www.airports.org.]
The histograms and summary statistics in

Display 7.54 show simulated distributions of
5000 sample means for samples of size 5, 10,
and 20, selected without replacement from
the numbers of passengers.
Check how well the shapes, means, and
standard deviations of the simulated
sampling distributions agree with what the
theory says they should be. Do you see any
reason why the theory you have learned
should not work well in any of these cases?
22-03-2009 21:12
Lesson
6 de 8
E59. Bottle caps are manufactured so that their

inside diameters have a distribution that
is approximately normal with mean
36 mm and standard deviation 1 mm. The
distribution of the outside diameters of the
bottles is approximately normal, with mean
35 mm and standard deviation 1.2 mm. If
a bottle cap and a bottle are selected at
random (and independently), what is the
probability that the cap is not too small to
fit on the bottle?
E60. Refer to E59. Suppose a cap is too loose if it
is at least 1.1 mm larger than the bottle. If a
cap and a bottle are selected at random, what
is the probability that the cap is too loose?
E61. A student drives to school in the morning
and drives home in the afternoon. She
finds that her commute time depends on
the day of the week and whether it is
morning or afternoon. Her data are shown
in Display 7.55.
Display 7.55 Commute times.
Display 7.54 Summary statistics and histograms of

the distribution of sample means (in
thousands) for samples of size 5, 10,
and 20.
a. What are the mean and variance 2

of the morning commute times? Of the
afternoon commute times?
b. If the student selects a day at random
and finds the total commute time for that
day, what are the mean and variance of
the sampling distribution of this total
commute time?
c. Are your answers in part b equal to the
sum of those in part a? Explain why they
should or shouldnt be equal.
Chapter Summary 463
22-03-2009 21:12
Lesson
7 de 8
AP1. Five math teachers are asked how many

pens they are currently carrying, and the
results are 1, 1, 1, 2, 2. Random samples
of size two are taken from this population
(without replacement). What is the median
of the sampling distribution of the median?
AP4. The distribution of the population of

the millions of household incomes in
California is skewed to the right. Which of
the following best describes what happens
to the sampling distribution of the sample
mean when the size of a random sample
increases from 10 to 100?
1
1.4
Its mean gets closer to the population
1.5
2
mean, its standard deviation gets
none of the above
closer to the population standard
AP2. With random sampling, which of the
deviation, and its shape gets closer to
following is the best reason not to use the
the populations shape.
sample maximum as an estimator for the
Its mean gets closer to the population
population maximum?
mean, its standard deviation gets smaller,
The sample maximum has too much
and its shape gets closer to normal.
variability.
Its mean stays constant, its standard
The sample maximum is biased.
deviation gets closer to the population
The sample maximum is difficult to
standard deviation, and its shape gets
compute.
closer to the populations shape.
The sample range is an unbiased
Its mean stays constant, its standard
estimator.
deviation gets smaller, and its shape gets
closer to normal.
The sample maximum does not have
None of the above
a normally distributed sampling
distribution.
AP5. The scores on a standardized test are
AP3. In computing the sample standard
normally distributed with mean 500 and
deviation, the formula calls for a division
standard deviation 110. In a randomly
by n 1. Which of the following is the best
selected group of 100 test-takers, what is
reason for dividing by n 1 instead of n?
the probability that the mean test score is
above 510?
For averages, always divide by n, and for
less than 0.0001
standard deviations, always divide by
n 1.
0.1817
You use only n 1 data values when
0.4638
computing standard deviations.
0.5362
Dividing by n 1 gives less variation
0.8183
in the sampling distribution of the
AP6.
A
statistics
teacher claims to be able to
population standard deviation.
guess, with better than 25% accuracy, which
Dividing by n 1 makes the sample
of four symbols (circle, wavy lines, square,
variance an unbiased estimator of the
or star) is printed on a card. To test this
population variance in random sampling.
claim, she guesses the symbol on 40 cards.
The sampling distribution of the sample
If you use the normal approximation to the
standard deviation is closer to normal
binomial to compute the probability that
when dividing by n 1.
22-03-2009 21:12
Lesson
8 de 8
she would get 15 or more (37.5%) correct

just by random guessing, which of the
following is closest to your z-score?
0.183
0.38
1.25
1.83
1067
AP7. In a large city, 20% of adults favor a new,
stricter dress code for the local schools.
Which of the following is closest to the
probability that more than 24% of 1200
adults randomly selected for a survey will
be in favor of the new dress code?
Investigative Tasks
AP9. On an average day in 2004, about 246,000
vehicles traveled east on the Santa Monica
freeway in Los Angeles to the interchange
with the San Diego freeway. Assume that a
randomly selected vehicle is equally likely
to go straight through the interchange, go
south on the San Diego freeway, or go north
on the San Diego freeway. [Source: Caltrans,
www.dot.ca.gov.]
a. What is the best estimate of the number

of vehicles that will go straight through
the interchange on an average day?
b. What numbers of vehicles are reasonably
likely to go straight through?
c. On an average day, Caltrans found that
less than 0.0001
138,300 vehicles went straight through
0.0003
the interchange, 46,800 went north on
0.0029
the San Diego Freeway, and 60,900 went
0.38
south. What can you conclude?
0.9997
AP10. You estimate that the people using an
elevator in an office building have an
AP8. A study found that infants get a mean of
average weight of roughly 150 lb with a
290 minutes of sleep between midnight
standard deviation of approximately 20 lb.
and 6 a.m. with a standard deviation of
The elevator is designed for a 2000-lb
30 minutes. A parent wants to know the
weight maximum. This maximum can be
probability that two randomly selected
exceeded on occasion but should not be
infants will sleep a total of more than
exceeded on a regular basis. Your job is
600 minutes between midnight and 6 a.m.,
to post a sign in the elevator stating the
but a statistics student tells the parent that
maximum number of people to ensure safe
theres not enough information here to
use. Keep in mind that it is inefficient to
perform that calculation. Which of the
make this number too small but dangerous
following is the best reason for the students
to make it too large.
conclusion?
a. What number would you use for
The mean of the total cannot be
maximum occupancy? Explain your
determined.
reasoning and assumptions.
Statistical techniques can be used for the
b. Otis Elevator Company sells the Holed
mean but not for the total.
Hydraulic Elevator, which lists a capacity
The standard deviation of the total
of 2000 lb, or 13 people in the United
cannot be determined.
States and 12 people in Canada. Are
The shape of the sampling distribution of
these numbers close to the number you
the total is unknown.
decided on in part a? Explain why you
would or would not expect that to be the
The value of z cannot be determined.
case. [Source: www.otis.com, 2006.]
AP Sample Test
465
22-03-2009 21:12
Lesson
1 de 2
Inference for Proportions
63% of U.S. households

answering a recent
survey reported that
they own a pet, up
from 56% in 1994,
the year the survey
started tracking this
information. Are you
confi dent that there
has been an increase
in the percentage of
U.S. households that
own a pet? Your answer
will depend on how
many households were
included in the surveys.
22-03-2009 21:12
Lesson
2 de 2
Open the daily newspaper, a newsmagazine, or your own favorite magazine and
you are likely to see the results of a public opinion poll. For example, a recent
national survey found that 55% of singles ages 1829 say that they arent in a
committed relationship and are not actively looking for a romantic partner. This
survey had a margin of error of 3%. [Source: Pew Research Center, February 13, 2006,
Not Looking for Love: Romance in America, www.pewresearch.org.]
In this chapter, you will learn the two basic techniques of statistical inference:
confidence intervals and significance testing. If you are familiar with polls, you
have seen the idea of confidence interval. The surveys result, 55% 3%, is
equivalent to a confidence interval of 52% to 58%, an interval of plausible values
for the true percentage. In your work in Chapter 1 on the Westvaco case, you used
the idea behind significance testing when you saw that if you randomly selected
three of the ten hourly workers to be laid off, you would be unlikely to get a
group with an average age as large as the average age of those actually laid off
by Westvaco.
The ideas on confidence intervals and significance testing developed here
are fundamental to work in statistical inference. They will form the basis of the
inferential procedures in the chapters to follow.

to construct and interpret a confidence interval to estimate the proportion of
successes in a binomial population
to use a significance test (hypothesis test) to decide if it is reasonable
to conclude that your sample might have been drawn from a binomial
population with a specified proportion of successes
to construct and interpret a confidence interval for the dierence between
the proportion of successes in one population and the proportion of
successes in another population
to use a test of significance to decide if it is reasonable to conclude that two
samples might have been drawn from two binomial populations that have
the same proportion of successes
to construct and interpret confidence intervals and tests of significance for
experiments
22-03-2009 21:12
Lesson
1 de 22
8.1
You can quantify how
confident you are in a
result from a sample.
Estimating a Proportion with Confidence

The Pew Research Center survey reported on the previous page found that 55%
of singles ages 1829 say that they arent in a committed relationship and are not
actively looking for a romantic partner. This percentage is based on interviews
with 1068 singles. The survey reported a margin of error of 3%.
The Pew Research Center didnt ask all young singles in the United States,
only 1068. However, unless there were some special dificulties in taking a
random sample or with the wording of the question, the researchers are 95%
confident that the error in the percentage is less than 3% either way. That is,
they are 95% confident that if they were to ask all young singles in the United
States, 55% 3%, or between 52% and 58%, would report that they arent in a
committed relationship and are not actively looking for a romantic partner. What,
exactly, can they mean by this? The answer involves the concept of reasonably
likely events.
Reasonably Likely Events

Random sampling is
the key.
In Section 7.3, you learned how much variability to expect in the proportions
of successes from repeated random samples taken from a given binomial
population. About 95% of all sample proportions will fall within about two
standard errors of the population proportion p, that is, within the interval
where n is the sample size. The sample proportions in this interval are called
reasonably likely. (Recall that this rule works well only under the condition that
both np and n(1 p) are at least 10.)
Reasonably Likely Events and Rare Events
Reasonably likely events are those in the middle 95% of the distribution of
all possible outcomes. The outcomes in the upper 2.5% and lower 2.5% of the
distribution are rare eventsthey happen, but rarely.
Example: Reasonably Likely Results from Coin Flips

Suppose you will flip a fair coin 100 times. What are the reasonably likely values
of the sample proportion ? What numbers of heads are reasonably likely?
468 Chapter 8 Inference for Proportions
22-03-2009 21:13
Lesson
2 de 22
In Section 7.3, you

learned that p = p and
Solution
Here the probability of a success p on each trial is 0.5, and you have 100 trials
a sample size of 100. From the previous formula, 95% of all sample proportions
should fall in the interval
or about 0.5 0.1. So the reasonably likely values of are in the interval from
0.4 to 0.6. Values of outside this interval can happen, but they happen rarely.
Equivalently, in about 95% of the samples, the number of successes x in the
sample will be in the interval
or about 50
10 successes.
Even though the Pew Research Center doesnt know the value of p, the actual
proportion of young singles who arent in a committed relationship and are
not actively looking for a romantic partner, it can use the idea in the preceding
example. For each possible value of p, Pew can compute how close to p most
sample proportions will be. By knowing the variability expected in random
samples taken from populations with different values of p, Pew can estimate how
close should be to the truth.

D1. Suppose 35% of a population think they pay too much for car insurance.
A polling organization takes a random sample of 500 people from this
population and computes the sample proportion, , of people who think
they pay too much for car insurance.
a. There is a 95% chance that will be between what two values?
b. Is the organization reasonably likely to get 145 people in the sample who
think they pay too much for car insurance?
D2. In parts ad, you will review the method of using simulation to construct a
sampling distribution.
a. Using a calculator, computer, or random-digit table, take a random
sample of size 40 from a population with 60% successes. Count the
number of successes in your sample. [See Calculator Note 8A.]
469
22-03-2009 21:13
Lesson
3 de 22
b. Continue taking samples of size 40 until your class has the number of
successes from 100 samples.
c. Plot your 100 values on a dot plot or a histogram.
d. From your plot in part c, what numbers of successes are reasonably likely?
e. Is your answer in part d close to what the theory predicts?
In Activity 8.1a, you will collect data about the proportion of students who can
make the Vulcan salute. Later in this section, you will learn to find a confidence
interval for the percentage of all students who can make the Vulcan salute.
The Vulcan Salute

The Vulcan salute from Star Trek, which means
live long and prosper, originated as part of a
priestly blessing in Jewish ceremony. Your goal is
to estimate what proportion of students can make
this salute, clearly and easily, with both hands.
1. Without giving anyone the chance to practice,
ask exactly 40 students to make the Vulcan
salute with both hands at once. Count the
number who are successful.
2. Suppose you can reasonably consider your
sample of 40 students a random sample of
all students. From the data you gathered, is
it plausible that 10% of all students could
make this salute? Is it plausible that the true
percentage is 60%? What percentages do
you think are plausible?
The Meaning of a Confidence Interval

In Activity 8.1b, you will make a chart that allows you to see the reasonably likely
events for all population proportions, p, when the sample size is 40. From there, it
is a short step to confidence intervals.
Constructing a Chart of Reasonably Likely Events

What youll need: a copy of Display 8.1
1. Suppose you take repeated random samples of size 40 from a population
with 60% successes. What proportions of successes would be reasonably
likely in your sample?
2. Compare the proportion of students you found able to make the Vulcan
salute in Activity 8.1a to your answer in step 1. Is it plausible to assume that
60% of all students can make the salute? Explain your reasoning.
(continued)
22-03-2009 21:13
Lesson
4 de 22
3. On a copy of the chart in Display 8.1, draw a horizontal line segment

aligned with 0.6 on the vertical axis, labeled Proportion of Successes in the
Population. The line segment should stretch in length over your interval
from step 1.
Your instructor will give you one of the population proportions whose line
segments are missing in Display 8.1.
4. Compute the reasonably likely sample proportions for your population
proportion, p.
5. On your copy of the chart in Display 8.1, draw a horizontal line segment
showing the reasonably likely sample proportions for your groups
population proportion, p.
6. Get the reasonably likely sample proportions from the other groups in your
class and complete the chart with line segments for those values of p.
7. Refer to your results from Activity 8.1a. For which population proportions
is your sample proportion reasonably likely?
Number of Successes in the Sample (n = 40)
Display 8.1 Reasonably likely sample proportions for samples of

size 40.
The line segments you drew on your copy of Display 8.1 in Activity 8.1b show
reasonably likely sample proportions when taking random samples of size 40
from a population with a given proportion of successes p.
Example: Plausible Percentages of Vulcan Saluters

In a group of 40 adults doing Activity 8.1a, exactly 30 were able to make the
Vulcan salute. Assuming this can be considered a random sample of all adults,
is it plausible that if you tested all adults, you would fi nd that 50% can make
the salute? Is it plausible that 80% can make the salute? What percentages are
plausible?
471
22-03-2009 21:13
Lesson
5 de 22
Solution
No, its not plausible that 50% of all adults can make the Vulcan salute, because
30 out of 40 (or = 0.75) isnt a reasonably likely sample proportion for a
random sample taken from a population with 50% successes. You can see this
from your chart or from Display 8.2 by going to the horizontal line segment next
to the population proportion 0.5. This line segment goes from about 0.35 to 0.65,
and that interval does not include a sample proportion, , of 0.75. Of course, it
is possible that the true percentage is 50% and that the adults were an unusual
sample, but its not very plausible.
Is it plausible that 80% is the actual percentage of all adults who can make
the salute? The horizontal line segment for a population percentage of 80% goes
from 0.65 to 0.95, and that interval does include the sample proportion, , of 0.75.
If the true percentage is 80%, you are reasonably likely to get 30 adults who can
salute Vulcan-style from a random sample of 40 adults.
To find the plausible percentages from your chart, draw a vertical line upward
from = 0.75, as in Display 8.2. The line segments it crosses represent the
plausible population percentages: 60%, 65%, 70%, 75%, 80%, and 85%. In all
these populations, a sample proportion, , of 0.75 would be a reasonably likely
result.
Display 8.2 A complete chart of reasonably likely outcomes for

samples of size 40. (These intervals were calculated
directly from the binomial distribution so you may
not get exactly the same interval if you use the formula
based on the normal approximation on page 468.)
The plausible percentages for the population proportion listed in the

preceding example (60% to 85%) are called the 95% confidence interval for the
percentage of all adults who can make the Vulcan salute. That is, you are 95%
confident that if you were to test all adults, the percentage who can make the
Vulcan salute would be somewhere in the interval from about 60% to 85%.
22-03-2009 21:13
Lesson
6 de 22
The population
proportion p is
the unknown
parameter.
95% Confidence Interval for a Population Proportion p

A 95% confidence interval consists of those population proportions p for
which the sample proportion is reasonably likely.

Use your completed chart or the one in Display 8.2 to answer these questions.
D3. According to the 2000 U.S. Census, about 60% of Hispanics in the United
States are of Mexican origin. Would it be reasonably likely that in a
survey of 40 randomly chosen Hispanics, 27 are of Mexican origin?
[Source: www.census.gov.]
D4. According to the 2000 U.S. Census, about 30% of people over age 85 are
men. In a random sample of 40 people over age 85, would you be reasonably
likely to get 60% who are men? [Source: www.census.gov.]
D5. Suppose that in a random sample of
40 toddlers, 34 know what color Elmo
is. What is the 95% confidence interval
for the percentage of toddlers who know
what color Elmo is?
D6. Polls usually report a margin of error.
Suppose a poll of 40 randomly selected
statistics majors finds that 20 are female.
The poll reports that 50% of statistics
majors are female, with a margin of
error of 15%. Use your completed
chart to explain where the figure 15%
came from.
D7. Why do we say we want a confidence interval for p rather than saying we
want a confidence interval for ?
From the Chart to a Formula

If you have noticed that Display 8.2 can be used to find a confidence interval
only when the sample size is 40, you might be thinking, Constructing a new
chart each time they give me a different sample size is going to be a lot of work
perhaps its not too late to transfer to that pottery class. But dont leave yet. There
is a quick and easy formula you can use.
To develop this formula, think again about the geometry behind a confidence
interval. Suppose that in a random sample of size 40 the sample proportion,
, turns out to be 0.6. As shown in Display 8.3, you can approximate the 95%
confidence interval by drawing a vertical line from = 0.6 and seeing which
horizontal line segments it intersects. These horizontal line segments give the
population proportions, p, for which the sample proportion is a reasonably
likely result. These populations are p = 0.45 up through p = 0.70 almost to
p = 0.75. This is the confidence interval, the set of plausible values for p.
473
22-03-2009 21:13
Lesson
7 de 22
Display 8.3 Finding a 95% confidence interval for p when the

sample proportion, , is 0.6 and the sample size, n,
is 40.
How can you find this confidence interval without the chart? The key is
to notice that the two bold line segments in Display 8.4 intersect at the point
(0.6, 0.6) and are approximately the same length. (They are approximately the
same length because the two sets of endpoints of the horizontal line segments
almost lie on two parallel lines with slope 1.)
Display 8.4 The length of the 95% confidence interval (the

vertical line segment) is the same as the length of the
horizontal line segment of reasonably likely sample
proportions for p = 0.6.
22-03-2009 21:13
Lesson
8 de 22
The bold horizontal line segment represents the reasonably likely sample
proportions, or the middle 95% of all sample proportions, from a population with
p = 0.6. You know how to find its endpoints:
The bold vertical line segment is the confidence interval, and it has the same
endpointsnamely, 0.6 0.15, or 0.45 and 0.75. So, to get the (vertical) confidence
interval, all you have to do is find the endpoints of the horizontal line segment by
substituting the known value of for the unknown value of p in the formula.

D8. Suppose you observe a sample proportion, , of 0.5 in a random sample of
size 40.
a. Use Display 8.2 to find an approximate 95% confidence interval for the
population proportion.
b. On a copy of Display 8.2, sketch in the two bold line segments for the
proportion 0.5, as in Display 8.4.
c. Explain the meaning of the bold horizontal line segment. Use a formula
to find its endpoints.
d. Explain the meaning of the bold vertical line segment. Find the
endpoints of the bold vertical line segment using your result from part c.
e. What are the endpoints of the 95% confidence interval?
D9. Explain the difference between the terms reasonably likely sample
proportions and plausible population proportions in explaining the
relationship between sample proportions and population proportions.
Using the Formula

You can use the reasoning described above to find a confidence interval for the
percentage of successes p in a population, no matter what the sample size or the
level of confidence desired. The general formula and the conditions for using it are
given in this box.
A Confidence Interval for a Population Proportion
A confidence interval for the proportion of successes p in the population is
given by the formula
z* is called the
critical value.
Here n is the sample size and is the proportion of successes in the sample.
The value of z* depends on how confident you want to be that p will be in the
(continued)
475
22-03-2009 21:13
Lesson
9 de 22
(continued)
confidence interval. If you want a 95% confidence interval, use z* = 1.96; for
a 90% confidence interval, use z* = 1.645; for a 99% confidence interval, use
z* = 2.576. Table A on page 824 or your calculator will give you the value of
z* for other confidence levels. [See Calculator Note 8B.]
This confidence interval is reasonably accurate when three conditions
are met:
The sample was a simple random sample from a binomial population.
Both n and n(1 ) are at least 10.
The size of the population is at least 10 times the size of the sample.
The first two conditions listed in the box are necessary for you to be able to
use the normal distribution (and z-scores) as an approximation to the binomial
distribution. If the third condition isnt met, your confidence interval will be
longer than it needs to be.
Margin of Error
The quantity
is called the margin of error. It is half the width of the confidence interval.
Example: Safety Violations

Suppose you have a random sample of 40 buses from a large city and find that
24 buses have a safety violation. Find the 90% confidence interval for the
proportion of all buses that have a safety violation.
Check conditions.
Do computations.
Solution
You will check the three conditions in D10.
For a 90% confidence interval, use z* = 1.645. The confidence interval for
the proportion p of all buses that have a safety violation based on a sample
proportion, , of or 0.6 can be written
or about (0.47, 0.73).
22-03-2009 21:13
Lesson
10 de 22
Conclusion in context.
You are 90% confident that the percentage of all of this citys buses that have
a safety violation is between 47% and 73%. The margin of error for this survey is
0.13, or 13%.
You can also use a calculator to calculate confidence intervals. [See Calculator
Note 8C.] Shown here are the values for the previous example.
Using the Formula

D10. In the bus example, are the three conditions for constructing a confidence
interval met? Explain. (You should always verify that the three conditions
are met before reporting a confidence interval.)
D11. What, in words, does n represent in the bus example? What, in words,
does n(1 ) represent in the bus example? What do these two quantities
represent in general?
The Capture Rate

Sometimes a confidence interval captures the true population proportion and
sometimes it doesnt. The capture rate of a method of constructing confidence
intervals is the proportion of confidence intervals that contain the population
parameter in repeated usage of the method. With a good method, the capture
rate is equal to the level of confidence advertised. For example, if a polling
company uses 95% confidence intervals in a large number of different surveys,
the population proportion p should be in 95% of them. You will test whether
that is indeed the case in Activity 8.1c (and explore this further in Investigative
Tasks AP9 and AP10 on pages 558559).
The Capture Rate

What youll need: a calculator or one table of random digits per student, a
copy of Display 8.5
1. Your instructor will assign you a group of 40 random digits, or you will
generate 40 random digits on your calculator.
2. Count the number of even digits in your sample of 40 random digits.
3. Use the formula to construct a 95% confidence interval for the proportion
of random digits that are even.
(continued)
477
22-03-2009 21:13
Lesson
11 de 22
(continued)
4. Each member of your class should draw his or her confidence interval as a
vertical line segment on a copy of Display 8.5.
5. What is the true proportion of all random digits that are even?
6. What percentage of the confidence intervals captured the true proportion?
Is this what you expected? Explain.
Display 8.5 Chart for recording a sample of 95% confidence

intervals.
Student: In the activity, about 95% of the 95% confidence intervals captured
the true population proportion of 0.5. That was no surprise.
But I dont see why that happened. Just calling something a
95% confidence interval doesnt make it one.
Statistician: Youre right. This isnt obvious. For me to explain the logic to you,
youll have to answer some questions as we go along.
Student: Okay.
Statistician: Go back and ask those of your classmates who had confidence
intervals that captured the true proportion of even digits
(p = 0.5) what values of they got. Well talk again tomorrow.
Student: (The next day) They had values of
between 0.35 and 0.65.
Statistician: Right again. Now look at Display 8.6. What do you notice about
these values of ?
Student: They all lie on the horizontal line segment for p = 0.5.
Statistician: Yes, the values of that give a confidence interval that captures
p = 0.5 are the reasonably likely outcomes for p = 0.5. What is
the chance of getting one of these good values of ?
Student: 95%!
22-03-2009 21:13
Lesson
12 de 22
Display 8.6 The population proportion p is in the confidence interval

if and only if
is a reasonably likely outcome for p.
A note about correct
language
It is correct to say that you expect the true value of p to be in 95 out of every
100 of the 95% confidence intervals you construct. However, it is not correct to
say, after you have found a confidence interval, that there is a 95% probability that
p is in that confidence interval. Here is an example that shows why. If you pick
a date in the next millennium at random, it is reasonable to say that there is a
chance you will pick a Tuesday. However, suppose the date you pick turns out
to be March 3, 3875. It sounds a bit silly to say that there is a chance that
March 3, 3875, is a Tuesday. Once the date is selected, there is no randomness
left. Either March 3, 3875, is a Tuesday or it isnt. All you can say is that the
process you have used to select a date gives you a Tuesday of the time. (This
example is credited to Wes White, an AP Statistics teacher in Los Angeles.)
The Capture Rate

D12. Was every students confidence interval in Activity 8.1c the same? Is this
how it should be? Explain.
D13. After working through parts a and b, explain the relationship, if any,
between the length of the confidence interval and the capture rate.
a. Construct an example of two confidence intervals with the same width
but different capture rates (one higher than the other).
b. Construct examples of two confidence intervals with the same capture
rate but different width.
Margin of Error and Sample Size

You might have noticed that the 95% confidence intervals for large sample sizes
are narrower than those for small sample sizes. This makes sensethe larger the
sample size, the closer should be to p. Lets see how this works for a specific
example.
479
22-03-2009 21:13
Lesson
13 de 22
Example: The Effect of Sample Size on the Margin of Error

Suppose you take a random sample and get = 0.7.
a. If your sample size is 100, find the 95% confidence interval for p, and state the
margin of error.
b. What happens to the confidence interval and margin of error if you
quadruple the sample size, to 400?
Solution
a. The 95% confidence interval for a sample size of 100 is
The 95% confidence interval is (0.6102, 0.7898), and the margin of error is
approximately 0.0898.
b. If you quadruple your sample size to 400, your 95% confidence interval
would be
Quadrupling the sample

size cuts the margin of
error in half.
The 95% confidence interval is (0.6551, 0.7449), and the margin of error is
approximately 0.0449, or half what it was before.

D14. All else being equal, will the margins of error for samples of size 80 be larger
or smaller than those for samples of size 40? Explain your reasoning.
D15. You can use Display 8.2 only for samples of size 40 and for 95% confidence
intervals.
a. How would a chart for 99% confidence intervals be different from
Display 8.2?
b. How would the sizes of the margins of error be different?
What Sample Size Should You Use?

How large the sample
should be depends on
how small a margin of
error you want.
People who conduct surveys often ask statisticians, What sample size should
I use? The simple answer is that the larger the sample size, n, the more precise
the results will be. When everything else is held constant, the margin of error is
smaller with a larger sample size than with a smaller one. However, researchers
22-03-2009 21:13
Lesson
14 de 22
always have limited time and money, so for practical reasons they have to limit
the size of their samples. For a con dence interval, the margin of error, E, is
approximately
You can solve this formula for n:
To use this formula, you have to know the margin of error, E, that is
acceptable. You have to decide on a level of con dence so you know what value
of z* to use, and you have to have an estimate of p. If you have a good estimate
of p, use it in this formula. If you do not have a good estimate of p, use p = 0.5.
Using this value for p might give you a sample size that is a bit too large, but it
will never give one that is too small. (This is because the largest possible value of
p(1 p) occurs when p = 0.5. See E21.)
Example: Estimating the Needed Sample Size

What sample size should you use for a survey if you want the margin of error to
be at most 3% with 95% confi dence but you have no estimate of p?
Solution
Because you do not have an estimate of p, use p = 0.5. Then
Because you must have a sample size of at least 1067.111, round up to 1068. To get
a margin of error around 3% is one reason why national polling organizations use
sample sizes of around 1000.

D16. Examine the formula for a con dence interval and explain why it is true that
when everything else is held constant, the margin of error is smaller with a
larger sample size than with a smaller one.
D17. Suppose it costs $5 to survey each person in your sample. You judge that p is
about 0.5. What will your survey cost if you want a 95% confidence interval
with a margin of error of about 10%? About 1%? About 0.1%?
481
22-03-2009 21:13
Lesson
15 de 22
Back to the Survey of Young Singles

The beginning of this section told about a recent Pew Research Center survey
reporting that 55% of singles ages 1829 arent in a committed relationship and
arent actively looking for a romantic partner. The sample size was 1068. This survey
had a margin of error of 3%, so the 95% confidence interval was 52% to 58%.
You now should be able to answer questions like these about this confidence
interval:
What is it that you are 95% sure is in the confidence interval?
The proportion of all singles ages 18 29 in the United States who arent in a
committed relationship and arent actively looking for a romantic partner.
What is the interpretation of the confidence interval of 52% to 58%?
You are 95% confident that if you could ask all singles ages 1829 if they arent
in a committed relationship and arent actively looking for a romantic partner,
between 52% and 58% would say yes.
What is the meaning of 95% confidence?
If you were to take 100 random samples of young singles and compute the
95% confidence interval from each sample, then you can expect 95 of the
confidence intervals to contain the proportion of all young singles ages 1829
who arent in a committed relationship and arent actively looking for a
romantic partner.
What populations plausibly could have resulted in a sample proportion of 0.55?
A 95% confidence interval consists of those population proportions for which
the observed sample proportion is a reasonably likely event. In this case, if
the population proportion were anything from 0.52 to 0.58, you would be
reasonably likely to get a sample proportion of 0.55 in a sample of size 1068.

D18. Verify the margin of error for the Pew Research Center survey.
D19. Sometimes polls report the error attributable to sampling instead of the
margin of error. Explain what this means.
D20. Most legitimate polling organizations publish details on how their polls
are conducted. Many include a statement similar to this: In addition to
sampling error, the practical difficulties of conducting a survey of public
opinion may introduce other sources of error. List at least three other
possible sources of error and explain why they are not included in the
sampling error measured by statistical formulas.
Summary 8.1: Estimating a Proportion

with Confidence
A 95% confidence interval consists of those population percentages p for which
the observed sample proportion is reasonably likely.
The formula for a confidence interval has the form
statistic (critical value) (standard deviation of statistic)
22-03-2009 21:13
Lesson
16 de 22
In the case of estimating a proportion, the confidence interval is given by
where is the observed sample proportion and n is the sample size.

With z* = 1.645, the method will capture the population proportion 90% of
the time; with z* = 1.96, it will do so 95% of the time; and with z* = 2.576, it will
do so 99% of the time.
This confidence interval is reasonably accurate when three conditions are met:
The sample was a simple random sample from a binomial population.
Both n and n(1 ) are at least 10.
The size of the population is at least 10 times the size of the sample.
There are two parts to giving an interpretation of a confidence interval: describing
what is in the confidence interval and describing what is meant by confidence. For a
95% confidence interval, for example, this interpretation would include:
You are 95% confident that, if you could examine each unit in the population,
the proportion of successes, p, in this population would be in this confidence
interval.
If you were able to repeat this process 100 times and construct the 100
resulting confidence intervals, youd expect 95 of them to contain the
population proportion of successes, p. In other words, 95% of the time the
process results in a confidence interval that captures the true value of p.
Remember that there can be sources of error other than sampling error. For
example, if the samples arent randomly selected or if there is a problem such as a
poorly worded questionnaire, the capture rates dont apply.
To estimate the sample size, n, needed for a given margin of error E, use the
formula
with a rough estimate of p if you have it; if not, use p = 0.5.
Practice
P1. Suppose 40% of students in your graduating
class plan to go on to higher education.
You survey a random sample of 50 of
your classmates and compute the sample
proportion of students who plan to go on
to higher education.
a. There is a 95% chance that will be
between what two numbers?
b. Is it reasonably likely to find that
25 students in your sample plan to go
on to higher education?
P2. Describe how to use simulation to find the

reasonably likely sample proportions for
a random sample of size 40 taken from a
population with p = 0.3.
P3. According to the U.S. Census Bureau, about
16% of the residents of the country do not
have health insurance. Suppose a polling
agency randomly selects 200 residents.
What numbers of residents without health
insurance are reasonably likely?
483
22-03-2009 21:13
Lesson
17 de 22

Use Display 8.2 on page 472 to answer P4P8.
P4. Suppose you flip a fair coin 40 times. How
many heads is it reasonably likely for you
to get?
P5. About 65% of 18- and 19-year-olds are
enrolled in school. If you take a random
sample of 40 randomly chosen 18- and
19-year-olds, would you be reasonably likely
to find that 33 were in school? [Source: U.S.
65% responded that their school was helping

them discover what type of work they would
love to do as a career. [Source: Gallup, Teens: Schools
Help Students Find Career Path, poll.gallup.com.]
a. Check to see if the three conditions for

computing a confidence interval are met
in this case.
b. Find a 95% confidence interval for the
percentage of all teens in the United
States who would respond that their
school is helping them discover a career
Census Bureau, Statistical Abstract of the United States,
path. What is the margin of error?
2006, Table 209.]
c. Find a 90% confidence interval for the
P6. About 85% of people in the United States
percentage of all teens in the United
age 25 or over have graduated from high
States who would respond that their
school. In a random sample of 40 people age
school is helping them discover a career
25 or older, how many high school graduates
path. What is the margin of error?
are you reasonably likely to get? What
proportions of high school graduates are you
d. Which confidence interval, 90% or 95%,
reasonably likely to get? [Source: U.S. Census Bureau,
is wider? Why should that be the case?
Statistical Abstract of the United States, 2006, Table 214.] P11. In the same survey as in P10, 4% of the
P7. In a random sample of 40 adults, 25% know
600 students responding gave their school
what color Elmo is. What is the 95%
a D rating (on a scale of A, B, C, D, F).
confidence interval for the percentage of
a. Check to see if the three conditions for
all adults who know what color Elmo is?
computing a confidence interval are met
P8. Suppose that in a random sample of
in this case.
40 retired women, 45% of the women
travel more than they did while they were
percentage of all teenagers in the United
working. Find the 95% confidence interval
States who would give their school a D
for the proportion of all retired women who
rating.
travel more.
c. How does the width of the confidence
interval in part b compare to that of the
confidence interval in part b of P10?
P9. Suppose that in a random sample of 40
What is the reason for this?
students from your school, 25 are wearing
sneakers. Find the 95% confidence interval
The Capture Rate
for the percentage of all students in your
P12. Suppose you know that a population
school who wear sneakers,
proportion, p, is 0.60. Now suppose 80
a. using the chart in Display 8.2
different students are going to select
b. without using the chart
independent random samples of size 40 from
this population. Each student constructs his
Using the Formula
or her own 90% confidence interval. How
P10. In a survey consisting of a randomly selected
many of the resulting confidence intervals
national sample of 600 teens ages 1317,
would you expect to include the population
conducted from July 6 to September 4, 2005,
proportion, p, of 0.60?
22-03-2009 21:13
Lesson
18 de 22
P13. A quality-control plan in a plant that

manufactures computer mice calls for taking
20 different random samples of mice during
a week and estimating the proportion of
defective mice in a 95% confidence interval
for each sample. If the percentage of defectives
is maintained at 4% throughout the week,
how many of the confidence intervals would
you expect not to capture this value?
P14. All else being equal, how would a chart
for samples of size 100 be different from
Display 8.2? How would the widths of the
confidence intervals be different?
P15. What value of z* should you use for an 80%
confidence interval? Will the margin of error
for 80% confidence be longer or shorter than
for 95% confidence? Explain your answer.
P16. Suppose you take a survey with sample size n
and get a margin of error that is three times
larger than you would like. What sample size
should you have used to obtain the desired
margin of error? Justify your answer.
P17. The accuracy of opinion polls is examined
closely in presidential election years.
Suppose your large polling organization is
planning a final poll before the presidential
election.
a. What sample size should you use to
have a margin of error of 2% with a
95% confidence level?
b. What sample size should you use to
have a margin of error of 1% with a
99% confidence level?
c. What sample size should you use if you
want a margin of error of 0.5% with
90% confidence?
P18. A Gallup poll found that 29% of adult
Americans report that drinking has been a
source of trouble in their families. Gallup
asks this question every year. What sample
size should Gallup use next year to get a

margin of error of 3% with 95% confidence
and survey as few people as possible? [Source:
Gallup News Service, July 31, 2006.]

P19. A recent survey about the use of the Internet
among teens gave this information: 76%
of middle school and high school students
who have Internet access go online to get
news or information about current events.
Of students ages 1217, the study found
that 87%, or about 20 million, have Internet
access. From that population, the study
obtained 971 responses to the question about
using the Internet to get news. [Source: Pew
Internet, 2005, www.pewinternet.org.]
a. Check that the three conditions are met

for computing a confidence interval for
the percentage of students with Internet
access who go online to get news or
information about current events.
b. Regardless of your answer to part a,
compute the 95% confidence interval.
c. What is it that you are 95% sure is in the
confidence interval?
d. Interpret this confidence interval.
e. Explain the meaning of 95% confidence.
485
22-03-2009 21:13
Lesson
19 de 22
Exercises
E1. Review the use of the chart in Display 8.2 on
page 472 by answering these questions.
a. The proportion of successes in a random
sample of size 40 is 0.90. Is this sample
proportion reasonably likely if the
population has 75% successes?
b. In the multicandidate 2000 presidential
election, Al Gore received 48% of the
votes cast, winning the popular vote
but losing in the electoral college. In
a random sample of 40 voters in this
election, what is the largest proportion of
people who voted for Vice President Gore
that is reasonably likely? The smallest
proportion?
c. Suppose a poll of a random sample of
40 shoppers finds that 16 prefer open-air
malls to enclosed malls. What is the 95%
confidence interval for the percentage of
all shoppers who prefer open-air malls?
E2. Review the use of the chart in Display 8.2 by
answering these questions.
a. The proportion of successes in a random
sample is 0.30. Is this sample proportion
reasonably likely in a sample of size 40 if
the population proportion is 0.40?
b. According to the U.S. Census Bureau,
about 16% of the residents of the country
do not have health insurance. In a
random sample of 40 residents, what
is the largest proportion of uninsured
people that is reasonably likely?
c. Suppose a random sample of 40 trees, in
a forest of thousands of trees, shows that
28 are infected by a certain beetle. What
is the 95% confidence interval for the
percentage of infected trees in the forest?
E3. A bilge alarm warns a boat captain when
water begins rising in the bottommost part
of the boat (the bilge). A working bilge
alarm is considered crucial in preventing
the flooding of fishing boats. Only 15% of
40 boats sampled from the ocean-fishing
fleet in the United Kingdom had bilge alarms
that were placed conveniently for testing and

maintenance. [Source: Banff & Buchan College and
the Universities of Glasgow and Strathclyde, Flooding of UK
Fishing Boats, January 2003.]
a. Find an approximate 95% confidence

interval, using Display 8.2, for the
proportion of all ocean-fishing boats in
the United Kingdom with appropriately
placed bilge alarms.
b. Find a 95% confidence interval using the
formula on page 475.
c. Compare the answers in parts a and b.
Give reasons for any discrepancy.
E4. Suppose a random sample of size 40
produces a sample proportion, of 0.80.
a. Find an approximate 95% confidence
interval from Display 8.2.
b. Find a 95% confidence interval using the
formula on page 475.
c. Compare the answers in parts a and b.
Give reasons for any discrepancy.
E5. A nationwide survey of 19,441 teens about
their attitudes and behaviors toward epilepsy
found that only 51% knew that epilepsy is
not contagious.
The executive summary from the Epilepsy
Foundation for this survey gave this
technical information:
The two-page survey was distributed to teens
nationally by 20 affiliates of the Epilepsy
Foundation from March 2001 through July
2001 in schools selected by each affiliate. A
total of 19,441 valid surveys were collected.
Mathew Greenwald & Associates, Inc., edited
the surveys and performed the data entry.
Greenwald & Associates was also responsible
for the tabulation, analysis, and reporting of
the data. The data were weighted by age and
region to reflect national percentages.
The margin of error for this study (at the
95% confidence level) is plus or minus
approximately 1%.
The survey was funded by the Centers for
Disease Control and Prevention. [Source:
www.efa.org, April 16, 2002.]
22-03-2009 21:13
Lesson
20 de 22
a. Do you have any reservations about the

design of this study?
b. Are the conditions for constructing a
confidence interval met?
c. Does a computation of the margin of
error using the formula in this section
agree with that given in the press release?
d. Assuming the conditions to construct
a confidence interval have been met,
what is it that you are 95% sure is in the
E6. Read this article.
Study: Teens Optimistic About Innovation
Brian Bergstein
Associated Press
BOSTONTeenagers have some seemingly
high expectations about what technology
might bring over the next decade, according
to a new Massachusetts Institute of
Technology study.
For example, 33 percent of teens predicted
that gasoline-powered cars will go the way of
the horse and buggy by 2015. Just 16 percent
of adults agreed. Meanwhile, 22 percent of
teenagers predicted desktop computers will
become obsolete a decade from now, while
only 10 percent of adults agreed. Adults, on
the other hand, were far more certain about
the demise of the landline telephone by
2015 (45 percent made that prediction) than
teenagers (17 percent). . . .
But he [the director of the survey] also
wonders whether enough of todays teens are
in a position to invent such solutions, noting
that engineering was teens third-most
attractive career choice, picked by 14 percent
as the field that most interested themand
just 4 percent of girls. Only 9 percent of
all teens said they were leaning toward
science. The top two career choices: Arts and
medicine, each picked by 17 percent of the
kids surveyed.
The LemelsonMIT program, which focuses
on encouraging young people to pursue
innovation, commissioned its invention

index in November, interviewing 500 teens
and 1,030 adults nationwide. The margin
of sampling error was plus or minus
4 percentage points for teens and 3 for
adults. [Source: www.siliconvalley.com, June 11, 2006.]
a. What would you like to know about the
survey design that is not described here?
b. Are the conditions for constructing
confidence intervals met for teens
responses? For adults responses?
c. Do the computations of the margins of
error for both teens and adults, using the
formula in this section, agree with that
given in the press release?
d. Assuming the conditions have been met
to construct a 95% confidence interval
for the proportion of teenagers who think
engineering is an attractive career choice,
interpret this confidence interval.
E7. A sample of 549 randomly selected teenagers
ages 1317 were asked whether it is
appropriate for parents to install a computer
program limiting what teens can access on
the Internet. Fifty-two percent responded
that, yes, this was an appropriate parental
measure. Check the conditions and then
construct a 90% confidence interval for the
proportion of teenagers nationwide who
would think this an appropriate measure.
[Source: Gallup, Teens Reject Restrictions on Movies,
Games, Music, 2005, www.poll.gallup.com.]
E8. In a CBS News/New York Times poll, 885

adults nationwide were asked this question:
How worried are you that popular culture
that is television, movies, and musicis
lowering the moral standards in this
country: very worried, somewhat worried,
not too worried, or not at all worried?
Forty percent of those interviewed said that
they were very worried. Check the conditions
and then construct a 90% confidence interval
for the proportion of adult Americans who
are very worried that popular culture is
lowering the moral standards of the country.
[Source: www.pollingreport.com.]
487
22-03-2009 21:13
Lesson
21 de 22
E9. Suppose a random sample of size 100

produces a sample proportion of 0.60. Can
Display 8.2 be used in this case to find a
95% confidence interval estimate of the
population proportion? Why or why not?
E10. A random sample of size 40 produces a
sample proportion of 0.05. Can Display
8.2 be used in this case to produce a
population proportion? Why or why not?
E11. A U.S. News World Report survey of
1000 adults selected from the general
public (published April 15, 1996) reported
that 81% thought TV contributes to a
decline in family values. If the sample was
randomly selected, what can you say about
the proportion of all adults who think TV
contributes to a decline in family values? (As
always, discuss whether the three conditions
are met, give the confidence interval itself,
and give an interpretation of this interval,
stating clearly what it is that you hope will be
in the confidence interval.)
E12. Another part of the U.S. News & World
Report survey went to Hollywood leaders
because the magazine also wanted to see
what Hollywood leaders thought. Of 6059
mailed surveys, only 570 were returned.
Among the returned surveys, 46% thought
TV contributed to a decline in family values.
The magazine does not report a margin of
error for this part of the survey. Should a
margin of error be reported? Explain.
E13. If everything else in your sample is left
unchanged,
a. what happens to the width of a confidence
interval as the sample size, n, increases?
b. what happens to the width of a
confidence interval as the degree of
confidence you have in the answer
increases?
E14. Explain the difference between p and .
Which is the parameter? Which is always in
the confidence interval?
E15. The poll described in E7 reports that a

random sample of 549 teenagers results
in a margin of error of approximately 5%.
a. Is this correct?
b. What sample size should be used if
the margin of error in estimating a
proportion is to be only 3%?
E16. If you want a margin of error that is
one-quarter of what you estimate it will be
with your current sample size, by what factor
should you increase the sample size?
E17. You want to determine the percentage of
seniors who drive to school. You take a
random sample of 125 seniors and ask them
if they drive to school. Your 95% confidence
interval turns out to be from 0.69 to 0.85.
Select each correct interpretation of this
situation. There might be no, one, or more
than one correct statement.
A. 77% of the seniors in your sample drive
to school.
B. 95% of all seniors drive to school from
69% to 85% of the time, and the rest drive
more frequently or less frequently.
C. If the sampling were repeated many
times, you would expect 95% of the
resulting samples to have a sample
proportion that falls in the interval from
0.69 to 0.85.
D. If the sampling were repeated many
times, you would expect 95% of the
resulting confidence intervals to contain
the proportion of all seniors who drive
to school.
E. You are 95% confident that the
proportion of seniors in the sample who
drive to school is between 0.69 and 0.85.
F. You are 95% confident that the
proportion of all seniors who drive to
school is in the interval from 0.69 to 0.85.
G. All seniors drive to school an average of
77% of the time.
H. A 90% confidence interval would be
narrower than the interval given.
22-03-2009 21:13
Lesson
22 de 22
E18. A survey was conducted to determine what

adults prefer in cell phone services. The
results of the survey showed that 73% of the
people wanted email service, with a margin
of error of plus or minus 4%. Which of these
sentences explains most accurately what is
meant by plus or minus 4%?
A. They estimate that 4% of the population
surveyed might change their minds
between the time the poll is conducted
and the time the survey is published.
B. There is a 4% chance that the true
percentage of adults who want email
service is not in the confidence interval
from 69% to 77%.
C. Only 4% of the population was surveyed.
D. To get the observed sample proportion of
73% would be unlikely unless the actual
percentage of all adults who want email
service is between 69% and 77%.
E. The probability that the sample proportion
is in the confidence interval is 0.04.
E19. In Section 5.2, you studied how to use tables
of random digits to simulate probabilities.
In the example on page 305, a simulation
of 5000 runs estimated that the probability
that both a randomly selected man and a
randomly selected woman would wash their
hands before leaving a public restroom was
about 0.67.
a. What is the margin of error for this
simulation?
b. How many runs of the simulation would
you need to estimate this probability with
a margin of error of 0.002?
E20. In a small factory, one assembly line
produces about 10% defectives (products
that need to be sent back for some
reworking), and the neighboring line
produces about 20% defectives. Ten items
are randomly selected from each line, and
the total number of defective items for the
two lines together is counted. If this number
is 5 or more, the lines are shut down for
inspection. (It is impractical to shut down
one line at a time in this small factory.)
a. Explain how to simulate the probability

that the lines will be shut down.
b. Display 8.7 shows the number of
defectives in each of 100 runs. (No values
larger than 7 were observed.) Use these
data to decide how many additional
runs to make so that you can estimate
the probability of seeing five or more
defectives in the pooled output of the two
lines with a margin of error of 0.01.
Display 8.7 Results of 100 runs of a simulation.
E21. If you dont have an estimate for p when

computing the needed sample size, you
use p = 0.5. In this exercise, you will show
that this is the safest value to use because
p = 0.5 requires the largest sample size, all
other things being equal. Answer parts ac
to prove that the largest possible value of
p(1 p) is achieved when p = 0.5.
a. Let y = x(1 x), where 0 x 1.
Why let y = x(1 x)? Why restrict
x to 0 x 1?
b. Sketch the graph of y = x(1 x). What is
the name of this type of function?
c. Where does the largest possible value of
y occur for this function? At what value
of x does this occur? What is the largest
possible value of y?
E22. Explain why it is not appropriate to use the
method of constructing a confidence interval
for a population proportion described in this
section when the condition that both n and
n(1 ) are at least 10 is violated. You can
use examples or a simulation to support your
arguments.
489
22-03-2009 21:13
Lesson
1 de 28
E23. Supply a reason for each step in parts ad

in this alternative explanation of why the
formula for a 95% confidence interval for a
population proportion is reasonable.
a. Ninety-five percent of all sample
proportions are within 1.96 standard
errors of p.
b. There is a 95% chance that the proportion
from a specific random sample will be
within 1.96 standard errors of p.
8.2
Tests of significance look

for departures from a
standard.
c. There is a 95% chance that p will be

within 1.96 standard errors of .
d. Thus, the formula for a 95% confidence
interval is approximately
E24. Observe 40 students on your campus

and estimate, using 98% confidence, the
percentage of students who carry backpacks.
Testing a Proportion
People often make decisions from data by comparing the results from a
sample to some predetermined standard. These kinds of decisions are called
tests of significance because the goal is to test the significance of the difference
between the sample and the standard. If the difference is small, there is no reason
to conclude that the standard doesnt hold. When the difference is large enough
that it cant reasonably be attributed to chance, you can conclude that the standard
no longer holds. In other words, the difference between the sample data and the
standard is stastically significant.
Here is an example of this type of reasoning. About 2% of barn swallows have
white feathers in places where the plumage is normally blue or red. The white
feathers are caused by genetic mutations. In 1986, the Russian nuclear reactor
at Chernobyl leaked radioactivity. Researchers continue to be concerned that
the radiation might have caused mutations in the genes of humans and animals
that have been passed on to offspring. In a sample of 266 barn swallows captured
around Chernobyl between 1991 and 2000, about 16% had white feathers in
places where the plumage is normally blue or red. Researchers compared the
proportion, , of 0.16 in the sample of captured barn swallows to the standard,
0.02. If the overall percentage was still only 2%, it is not reasonably likely that they
would get 16% in the sample. So they came to the conclusion that there was an
increased probability of genetic mutations in the Chernobyl area. [Source: A. P. Moller
and T. A. Mousseau. Albinism and Phenotype of Barn Swallows (Hirundo rustica) from Chernobyl,
Evolution 55 (2001): 20972104.]
In Activity 8.2a, you will record the results of 40 flips of a penny and 40 spins
of a penny. You will use the data later to perform tests of significance.
Spinning and Flipping Pennies

What youll need: pennies (at least one per student)
People tend to believe that pennies are balanced. They generally have no
qualms about flipping a penny to make a fair decision. Is it really the case that
penny flipping is fair? What about spinning pennies?
(continued)
22-03-2009 21:14
Lesson
2 de 28
1. Begin spinning the penny that your

instructor provides. To spin, hold the
penny upright on a table or the floor
with the forefinger of one hand and
flick the side edge with a finger of the
other hand. The penny should spin
freely, without bumping into things
before it falls. Spin pennies until your
class has a total of 40 spins. Count the number of heads and compute .
2. Do you believe heads and tails are equally likely when spinning pennies, or
can you reject this standard? Explain.
3. Begin flipping your penny. Let the penny fall, and record whether it lands
heads or tails. Flip pennies until your class has a total of 40 flips. Count the
number of heads and compute .
4. Do you believe heads and tails are equally likely when flipping pennies, or
can you reject this standard? Explain.
Informal Significance Testing

The logic involved in deciding whether to reject the standard that spinning a
penny results in heads 50% of the time is the same as that used to construct a
confidence interval.
Example: Jenny and Mayas Spins

Jenny and Maya wonder if heads and tails are equally likely when a penny is spun.
They spin pennies 40 times and get 17 heads. Should they reject the standard that
pennies land heads 50% of the time?
Variation in sampling
means that different
samples usually give
different results.
Solution
Of course, Jenny and Maya shouldnt expect to get exactly 20 heads (50% of 40)
even if heads and tails are equally likely. There is variation in samplingyou dont
get exactly the same result each time you take a random sample from a population.
Jenny and Maya construct the sampling distribution in Display 8.8, which shows
the probabilities of getting different values of when n = 40 and p = 0.5.
Display 8.8 Sampling distribution of
for n = 40 and p = 0.5.
491
22-03-2009 21:14
Lesson
3 de 28
Jenny and Mayas sample proportion, , of or 0.425 lies near the middle of
this distribution. This means that if the percentage of heads actually is 50% when
a penny is spun, getting 17 heads out of 40 spins is reasonably likely. So Jenny and
Maya have no evidence to argue against the standard that a penny lands heads
50% of the time when spun.
A sample proportion is said to be statistically significant if it isnt a
reasonably likely outcome when the proposed standard is true. (The statement
that the proposed standard is true is called the null hypothesis, indicated by
the symbol by H0 .) Jenny and Mayas result of 17 heads out of 40 spins wasnt
statistically significant. Their value of , 0.425, is a reasonably likely outcome if
spinning a penny is fair.
Example: Miguel and Kevins Spins

Miguel and Kevin spin pennies and get 10 heads out of 40 spins for a sample
proportion, , , of 0.25. Is this a statistically significant result?
Reject a value of the

standard that is not likely
to have produced the
sample proportion.
Solution
As shown in Display 8.8, if 0.5 is the true proportion of heads when a penny is
spun, it would be a rare event to get only 10 heads in a sample of 40 spins. Miguel
and Kevin have a statistically significant result. This leads them to conclude that
0.5 probably is not the true proportion of heads when a penny is spun.
Another way of seeing this is to note that p = 0.5 isnt included in the 95%
confidence interval for the true proportion of heads when a penny is spun. Look
at the vertical line at = 0.25 in Display 8.9. It doesnt intersect the horizontal
line that represents the population with = 0.5. That is, the sample proportion
= 0.25 is far enough from the standard of 0.5 to reject 0.5 as a plausible value
for the proportion of times heads appear when a penny is spun.
Display 8.9 A sample proportion of

likely when p = 0.5.
= 0.25 isnt reasonably
22-03-2009 21:14
Lesson
4 de 28
p: the population
proportion
: the sample proportion
p0: the hypothesized

value of the population
proportion
When you spin a penny, there are three different proportions to keep straight:
The true proportion of heads, p, when a penny is spun. No one knows this
value for sure. You might suspect from your own results that p does not
equal 0.5.
The proportion of heads, , in your sample of 40 spins. Jenny and Mayas
sample proportion was = or 0.425. Miguel and Kevins sample
proportion was = , or 0.25.
The proportion of heads that you hypothesize is the true proportion of heads
when a coin is spun. The symbol used for this standard value is p0. You have
been testing the standard that spinning a penny is fair, so you have been using
p0 = 0.5.

D21. Consider the question of whether spinning a penny is fair. You will need
your penny-spinning results from Activity 8.2a.
a. When investigating whether spinning a penny is fair, what is the
standard?
b. What is your sample proportion, , of heads from Activity 8.2a?
c. Locate your sample proportion, , on the chart in Display 8.9. If spinning
a penny is fair, is your sample proportion, , reasonably likely?
d. What is your conclusion? Do your data provide evidence against the
standard?
D22. Use your penny-flipping results from Activity 8.2a to answer the questions
in D21.
The Test Statistic

Miguel and Kevin could see from Display 8.8 that it is not reasonably likely that
they would get = 0.25 when p = 0.5, so they rejected the null hypothesis that a
spun penny comes up heads half the time. How could they have determined this
without constructing the entire distribution?
They need to find out if their value of lies in the middle 95% of all possible
values of . The distribution in Display 8.8 is approximately normal, so Miguel and
Kevin can compute a z-score for their sample proportion, , of 0.25:
493
22-03-2009 21:14
Lesson
5 de 28
Having a sample proportion that is 3.16 standard errors below the hypothesized
mean p0 , of 0.5, would definitely be a rare event if it is true that half of spun pennies
land heads. The value of is outside the reasonably likely outcomes for p = 0.5,
so Miguel and Kevin reject the null hypothesis that spinning a penny is fair. The
z-score that Miguel and Kevin computed is an example of a test statistic.
The Test Statistic for Testing a Proportion
To determine if is reasonably likely or a rare event for a given standard p0 , you
need to check the value of the test statistic
A z-score in
disguise!
This tells you how many standard errors the sample proportion
standard, p0 .
lies from the
The Test Statistic

D23. What does it mean if the test statistic has the value 0?
D24. If everything else remains constant, what will happen to the test statistic z when
a. the sample size n increases?
b. gets farther from p0?
D25. Could Miguel and Kevin be wrong when they decide that spinning a penny
isnt fair? Explain.
P-Values
Instead of simply reporting whether a result is statistically significant, it is
common practice also to report a P-value.
The P-value for a test is the probability of seeing a result from a random
sample that is as extreme as or more extreme than the result you got from your
random sample if the null hypothesis is true.
In Chapter 1 you found the equivalent of a P-value by simulation when
examining the chance that the mean age of three randomly selected laid-off
workers would exceed a given value. Now, you will learn a way to find such values
by theory rather than by simulation.
Example: Interpreting a P-Value

Chimpanzees were given the choice of pulling one of two heavy rakes toward
them. Pulling one of the rakes would bring food with it. Pulling the other rake
would not bring in food. Display 8.10 shows a typical setup for this study. The
round object represents the food. The researchers found that the correct rake was
22-03-2009 21:14
Lesson
6 de 28
chosen significantly more often than would be expected by chance (P = 0.01).

Interpret this P-value in the context of the situation. [Source: Victoria Horner and Andrew
Whiten, Causal Knowledge and Imitation/Emulation Switching in Chimpanzees (Pan troglodytes) and
Children (Homo sapiens), Animal Cognition 8 (2005): 16481.]
Display 8.10 Diagram of chimpanzee rake setup.
Solution
The P-value 0.01 means that if the chimps were selecting a rake at random,
which gives them a 5050 chance of selecting the one that rakes in the food, the
probability is only 0.01 that the chimps would rake in the food as often as or more
often than they did. The researchers therefore can conclude that the chimps were
able to deliberately choose the rake that got them the food.
Example: Computing a P-Value for Jenny and Maya

Find the P-value for Jenny and Mayas test of whether spinning a penny is fair.
Solution
Jenny and Maya got 17 heads out of 40 spins, so the value of the test statistic is
Assume for now that the null hypothesis is true: Spinning a penny gives you
heads 50% of the time. The probability of getting a value of z less than 0.95 can
be found in Table A on page 824 or from a calculator. This value is approximately
0.171, as shown in Display 8.11.
Display 8.11 The probability of 17 or fewer heads with 40 spins

of a coin, if spinning a coin is fair.
However, Jenny and Maya notice that getting 23 or more heads is just as
extreme as getting 17 heads or fewer. Now the sketch looks like Display 8.12.
495
22-03-2009 21:14
Lesson
7 de 28
The phrase in italics is

crucial in describing the
meaning of a P-value.
The probability of getting a result as far out in the tails of the distribution as they
did is 2(0.171), or 0.342. This is their P-value. If the probability of getting heads
is 0.5, there is a 34.2% chance of getting 17 heads or fewer or getting 23 heads
or more in 40 spins. Because this P-value is relatively large, Jenny and Maya
dont have statistically significant evidence to conclude that spinning a penny is
anything other than fair.
Display 8.12 The probability of at most 17 or at least 23 heads

with 40 spins of a coin, if spinning a coin is fair.
[See Calculator Note 8D to learn how to find a value of z using your calculator.]
A large value of z
indicates that is far
from .
In summary, the test statistic, z, is computed using the hypothesized value, p0,
as the mean. A small P-value tells you that the sample proportion you observed is
quite far away from p0 . Your data arent behaving in a manner consistent with the
null hypothesis. A large P-value tells you that the sample proportion you observed
is near p0 , so your result isnt statistically significant. So, the P-value weighs the
evidence found from the data: A small P-value places the weight against the null
hypothesis, and a large P-value weighs in as consistent with the null hypothesis.
P-Values
D26. Why is the phrase if the null hypothesis is true necessary in the
interpretation of a P-value?
D27. Does the P-value give the probability that the null hypothesis is true?
Critical Values and Level of Significance

You have been rejecting the hypothesized standard when would be a rare
event if that standard were true. This is equivalent to rejecting the standard
when the value of the test statistic, z, is in the outer 5% of the standard normal
distribution, that is, when z is less than 1.96 or greater than 1.96. Dividing
points such as 1.96 are called critical values (denoted z*). The corresponding
proportion, 0.05 in this case, is called the level of significance (denoted ).
Other critical values can be used depending on the level of significance desired.
22-03-2009 21:14
Lesson
8 de 28
Example: Finding Critical Values

Suppose you want to reject the null hypothesis when the test statistic, z, is in
the outer 10% of the standard normal distributionthat is, with a level of
significance equal to 0.10. What should you use as critical values? Should you
reject the null hypothesis if z turns out to be 1.87?
Solution
Find the values of z in Table A on page 824 that cut off the top 5% and the
bottom 5% of the standard normal distribution. The value that cuts off the
bottom 5% lies about halfway between 1.64 and 1.65. Thus, the critical values
for = 0.10 are 1.645, as shown in Display 8.13. (For most critical values, you
can simply use the nearest value of z, to two decimal places.) You can also use
your calculator to find critical values. [See Calculator Note 8B.]
Display 8.13 The test statistic z = 1.87 is more extreme than

z* = 1.645.
Because z = 1.87 is more extreme than z* = 1.645, reject the null hypothesis.
The sample proportion, , is farther from p0 than would be reasonably likely if p0
were the true population proportion.
The box summarizes the use of critical values.
Using Critical Values and the Level of Significance
If the value of the test statistic z is more extreme than the critical values, z*,
you have chosen (or, equivalently, the P-value is less than ), you have evidence
against the null hypothesis. Reject the null hypothesis and say that the result is
statistically significant.
On the other hand, if z is less extreme than the critical values (or,
equivalently, the P-value is greater than ), you do not have evidence against
the null hypothesis and so you cannot reject it.
If a level of significance isnt specified, it is usually safe to assume that
= 0.05 and z* = 1.96.
497
22-03-2009 21:14
Lesson
9 de 28

D28. Suppose your null hypothesis is that spinning (or flipping) a coin is fair. Find
the test statistic, z, for your results from Activity 8.2a. Using = 0.05, give
your conclusion
a. for spinning 40 pennies
b. for flipping 40 pennies
D29. If everything else remains constant, is it easier to reject the null hypothesis
when the level of significance, , is larger or smaller?
D30. Select the correct words in each sentence.
a. A larger critical value makes the level of significance, , smaller/larger
and so it will be harder/easier to reject the null hypothesis.
b. A smaller critical value makes the level of significance, , smaller/larger
and so it will be harder/easier to reject the null hypothesis.
The Formal Language of Tests of Significance

You have been using informal language to describe tests of significance. More
formal terminology appears in the box. You will examine each component of a
significance test thoroughly in the rest of this section.
Components of a Significance Test for a Proportion

1. Give the name of the test and check the conditions for its use. For a
significance test for a proportion, three conditions must be met:
The sample is a simple random sample from a binomial population.
Both np0 and n(1 p0) are at least 10.
The population size is at least 10 times the sample size.
2. State the hypotheses, defining any symbols. When testing a proportion,
the null hypothesis, H0, is
H0: The percentage of successes, p, in the population from which the
sample came is equal to p0.
The alternative hypothesis, Ha, can be of three forms:
For now, we will use
just the first Ha.
Ha: The percentage of successes, p, in the population from which the

sample came is not equal to p0.
sample came is greater than p0.
sample came is less than p0.
(continued)
22-03-2009 21:14
Lesson
10 de 28
3. Compute the test statistic, z, and find the critical values, z*, and the
P-value. Include a sketch that illustrates the situation.
The test statistic is
4. Write a conclusion. There are two parts to stating a conclusion:

Compare the value of z to the predetermined critical values, or compare
the P-value to . Then say whether you reject the null hypothesis or dont
reject the null hypothesis, linking your reason to the P-value or to the
critical values.
Tell what your conclusion means in the context of the situation.
Example: Jenny and Maya

Illustrate the components of a significance test using Jenny and Mayas results
from spinning a penny 40 times and getting 17 heads. Use = 0.05.
Always check conditions
first.
State your hypotheses

next.
Compute the test

statistic, draw a sketch,
and find the P-value.
Solution
1. Apparently, Jenny and Maya spun the pennies carefully and independently,
so we can consider the number of heads they got to be a binomial random
variable. Both np0 and n(1 p0) are at least 10, because np0 = 40(0.5), or
20, and n(1 p0) = 40(0.5), or 20. The population of all possible spins is
infinitely large, so it definitely is at least 10 times their sample size of 40. The
conditions for doing a significance test for a proportion are met.
2. H0: The proportion of heads when a penny is spun is equal to 0.5.
Ha: The proportion of heads when a penny is spun is not equal to 0.5.
Or you could write
H0: p = 0.5, where p is the proportion of heads when a penny is spun
Ha: p 0.5
3. As you saw in the example on page 495, the test statistic is
499
22-03-2009 21:14
Lesson
11 de 28
The critical values are 1.96 and the P-value is 0.342. A sketch illustrating the
situation is shown in Display 8.14.
Display 8.14 A z-score of 0.95.

Write a conclusion that
is linked to the value of
z or to the P-value and is
stated in the context of
the situation.
4. Because z = 0.95 is less extreme than the critical value, 1.96, this isnt a
statistically significant result. Equivalently, because the P-value, 0.342, is larger
than = 0.05, this isnt a statistically significant result.
If spinning a penny results in heads 50% of the time, you are reasonably
likely to get only 17 heads out of 40 flips. Thus, Jenny and Maya cannot
reject the null hypothesis that spinning a penny is a fair process. (You should
never say that you accept the null hypothesis, because if you constructed
a confidence interval for p, all the other values in itnot just 0.5are also
plausible values of p.)
You will more easily remember the structure of a test of significance if you
keep in mind what you should do before you look at the data from the sample.
Check the conditions for the test. Checking the conditions to be met for a
significance test for a proportion requires only knowledge of how the sample
was collected, the sample size, and the value of the hypothesized standard,
p0. (For some tests, as you will learn later, you will have to peek at the data in
order to check conditions.)
Write your hypotheses. The hypotheses should be based on the research
question to be investigated, not on the data. Ideally, the investigator sets the
hypotheses before the data are collected. This means that the value of should
not appear in your hypotheses.
Decide on the level of significance. The level of significance, , and the
corresponding critical values, z*, are set by the investigator before the data
are collected. Some fields have set standards for the level of significance,
typically = 0.05.
Sketch the standard normal distribution. Mark the critical values, z*. You
can place the value of the test statistic, z, on this distribution after you look
at the data.
You will use the data (number of successes and value of ) only to compute
the test statistic and to write your conclusion.
22-03-2009 21:14
Lesson
12 de 28

D31. What is the difference in the meaning of the symbols p, p0, and ? Of the
three, which varies depending on the sample you select? Which of the three
is unknown?
D32. A large study of births in the United States found that about 51.2% of all
babies born in the United States are boys and about 48.8% are girls. This
is a statistically significant difference, meaning that the difference cant
reasonably be attributed to chance. [Source: Centers for Disease Control, 2005, www.cdc
.gov; Trend Analysis of the Sex Ratio at Birth in the United States, National Vital Statistics ,
Reports 53 no. 20 (June 14, 2005): 1.]
Just because a
result is statistically
significant
doesnt mean
the difference is
of any practical
significance.
a. Is this result of any practical significance to a pair of prospective parents

who were hoping for a girl?
b. Give an example where this difference would have some practical
significance.
D33. Suppose the null hypothesis is true and you have a random sample of
size n.
a. What is the approximate probability that the test statistic will be greater
than 1.96? What is the probability that it will either be larger than 1.96
or be less than 1.96?
b. What is the probability that the test statistic will exceed 1.645? What
is the probability that it will either be larger than 1.645 or be less
than 1.645?
c. Why do you need to know that H0 is true to answer parts a and b?
Types of Errors
The reasoning of significance tests often is compared to that of a jury trial. The
possibilities in such a trial are given in this diagram.
In the same way, there are two types of errors in significance testing.
Miguel and Kevin got a test statistic with a large absolute value and so
concluded that spinning a penny is not fair. However, Jenny and Maya got a test
statistic with a value that was close to 0, so the result from their sample was quite
consistent with the idea that spinning a penny is fair. Who is right? Thousands of
501
22-03-2009 21:14
Lesson
13 de 28
statistics students have contributed to the solution of this problem by spinning

pennies. Their instructors have reported that the true proportion of heads seems
to be around 0.4. So Miguel and Kevin made the correct decision. Jenny and Maya
have made the error of not rejecting a null hypothesis that is falsethey have
made a Type II error.
If, like Miguel and Kevin, your test statistic is large in absolute value, then
there are several possibilities to consider:
The null hypothesis is true and a rare event occurred. That is, it was just bad
luck that resulted in being so far from p0.
The null hypothesis isnt true, and thats why the sample proportion, , was so
far from p0.
The sampling process was biased in some way, so the value of is itself
suspicious.
If you can rule out the last possibility, then the usual decision is to reject the null
hypothesis. However, you might be making a Type I error rejecting H0 even
though H0 is actually true.
If, like Jenny and Maya, your test statistic is small in absolute value, then there
are also several possibilities to consider:
The null hypothesis is true, and you got just about what you would expect in
the sample.
The null hypothesis isnt true, and it was just by chance that turned out to be
close to p0.
The sampling process was biased in some way, so the value of is itself
suspicious.
If you can rule out the last possibility, then the usual decision is to say that you
cannot reject the null hypothesis. However, you might be making a Type II error,
as Jenny and Maya did. This doesnt mean they did anything wrong. The error
occurred because the results from their sample just happened to be consistent
with the null hypothesis.
Example: An ESP Test and a Type I Error

Several Internet sites allow you to take a test to see if you have extrasensory
perception (ESP). Ann took one of these tests. She was presented with the back
of one of the five different cards described in E22 in Chapter 5 (page 314).
She guessed which of the five cards it was. In 100 tries with different cards, she
guessed right 29 times. The Web site said that her result was significant and the
evidence for ESP was fair. What is the P-value for this test? Discuss what type of
error might have been made.
Solution
Conditions have been met to use a normal approximation. The test statistic is
22-03-2009 21:14
Lesson
14 de 28
z-score of 2.25.
The corresponding P-value is 0.02, as illustrated in Display 8.15.
Display 8.15 The P-value for a
Ann knows that she does not have ESP and, in fact, didnt even try to discern
what the card was, instead selecting her choices rapidly and at random. In other
words, the null hypothesis is true that, in the long run, she will guess the correct
card 20% of the time. A Type I error has been made. Out of every 100 people who
take such a test, we expect that two of them will get 29 or more cards right (or 11
or fewer cards right). Ann was one of the lucky two. When she tried the ESP test
again, she correctly identified only 17 cards.
If H0 is true, the
probability of a Type I
error is a.
Suppose the null hypothesis is true. What is the chance that you will reject
it, making a Type I error? The only way you can make a Type I error is to get a
rare event from your sample. For example, if you are using a significance level of
0.05, you would make a Type I error if you get a value of the test statistic larger
than 1.96 or smaller than 1.96. This happens only 5% of the time no matter
what the sample size. Thus, if the null hypothesis is true, the probability of a
Type I error is 0.05. If you used z* = 2.576, the critical values for a significance
level of 0.01, then the probability of making a Type I error would be only 0.01. If
the null hypothesis is true, the probability of a Type I error is equal to the level of
significance. To lower the chance of a Type I error, then, your best strategy is to
have a low level of significance or, equivalently, large critical values.
Types of Errors
D34. Suppose you use critical values of 2 and the null hypothesis is actually
true. What is the probability that you will get a sample that results in
rejecting the null hypothesis? Explain. What type of error is this?
D35. To avoid Type I errors, why not always use a very large critical value?
Power and Type II Errors

If Jenny and Maya had spun more pennies, they soon would have realized that
the true proportion of heads is less than 0.5. However, their sample size was too
small for their test to detect a significant difference between the proportion of
heads in their sample and the hypothesized value, 0.5. So their analysis resulted
in a Type II error, that is, a failure to reject a false null hypothesis. Another
way to state this is to say that their test did not have enough power to be able
to detect that the true proportion of heads, p, is different from the proportion
hypothesized, p0.
503
22-03-2009 21:14
Lesson
15 de 28
Power of a Test
The probability of
a Type II error is
denoted .
The power of a test is the probability of rejecting the null hypothesis. When the
null hypothesis is false,
power = 1 probability of a Type II error
It follows that if the probability of a Type II error is small, then the power of the
test is large.
Example: Jenny and Maya Get More Power

Jenny and Maya start again and this time spin 200 pennies instead of only 40.
They happen to get the same proportion of heads, = 0.425, that they got with
only 40 spins. If they use = 0.05 as the level of signifi cance, what conclusion
would they come to now? Explain why this is the case.
Solution
Their new test statistic is
z=
Because z = 2.12 is more extreme than the critical value, 1.96, Jenny and
Maya now can come to the correct conclusion that the proportion of heads when
a penny is spun is not 0.5.
The reason why the larger sample size gave a statistically significant result and
the smaller sample size did not is illustrated in the diagrams that follow. Display
8.16 shows that a sample proportion, , of 0.425 is reasonably likely for a fair
coin spun 40 times. With 40 spins, Jenny and Maya have no evidence to reject the
hypothesis that spinning is fair.
Display 8.16 Sampling distribution of when n = 40 and

p = 0.5. The value = 0.425 is indicated by the
vertical line.
22-03-2009 21:14
Lesson
16 de 28
Display 8.17 shows that a sample proportion, , of 0.425 is not reasonably

likely for a fair coin spun 200 times. The difference between the two distributions
is their spread. The spread is controlled by the sample size: bigger sample, less
spread. If spinning a coin is fair, Jenny and Maya should get quite close to 50%
heads with 200 spins. Because they were far away from 0.50 in this distribution,
they can correctly conclude that spinning a coin doesnt result in 50% heads. In
other words, the larger sample size produces greater power to make the correct
decision.
Display 8.17 Sampling distribution of when n = 200 and

p = 0.5. The value of = 0.425 is indicated by the
vertical line.
Bigger n, more power.
Smaller n, less power.
As the previous example shows, if the null hypothesis is false and should be
rejected, a larger sample size increases the probability that you will be able to
reject it. However, if the null hypothesis is true, increasing the sample size has no
effect on the probability of making a Type I error. The only way you can decrease
the probability of making a Type I error is to make the level of significance,
, smaller. As youll see in the next example, however, if the null hypothesis is
actually false, this strategy results in a higher probability of a Type II error and
lower power!
Example: Miguel and Kevin Lose Power and Make an Error

Suppose Miguel and Kevin were worried about making a Type I error.
Consequently, they decided to use a 0.001 level of significance. That is, if it is true
that spinning a penny is fair, they wanted to have only a 0.001 chance of deciding
wrongly that spinning isnt fair. What is their conclusion now?
Solution
With = 0.001, the critical values found from a calculator are = 3.29. (If you
are using Table A on page 824, you will find that 3.27 cuts off an area of
approximately 0.0005.) With this level of significance, the value of their test
statistic, z = 3.16, is inside the critical values, so Miguel and Kevin would not
have rejected the false null hypothesis that spinning a penny is fair. They would
have made a Type II error because their test lost power.
505
22-03-2009 21:14
Lesson
17 de 28
In the previous examples, you observed three important properties of power:

Power increases as the sample size increases, all else being held constant.
Power decreases as the value of decreases, all else being held constant.
Power increases when the true population proportion, p, is farther from the
hypothesized value, p0. When p is farther from p0, tends to be farther from
p0, so the test statistic tends to be larger.
The box summarizes what you should understand about types of error and
power.
Types of Error and Power
Type I Error
When the null hypothesis is true and you reject it, you have made a Type I
error. The probability of making a Type I error is equal to the significance level,
, of the test. To decrease the probability of a Type I error, make smaller.
Changing the sample size has no effect on the probability of a Type I error.
If the null hypothesis is false, you cant make a Type I error.
Type II Error
When the null hypothesis is false and you fail to reject it, you have made a
Type II error. To decrease the probability of making a Type II error, take a
larger sample or make the significance level, , larger.
If the null hypothesis is true, you cant make a Type II error.
Power
Power is the probability of rejecting the null hypothesis.
When the null hypothesis is false, you want to reject it and therefore you want
the power to be large. To increase power, you can either take a larger sample or
make larger.

D36. Explain why an increase in sample size increases the power of a test, all else
remaining unchanged.
D37. What happens to the power of a test as the population proportion, p,
moves farther away from the hypothesized value, p0, all else remaining
unchanged?
D38. Recall that the power of a test is the probability of rejecting the null
hypothesis. Can a statistical test of the type discussed in this chapter ever
have a power of 1? Can a statistical test of the type discussed in this chapter
ever have a power of 0? If so, would either be desirable from a practical point
of view?
22-03-2009 21:14
Lesson
18 de 28
One-Sided Tests of Significance

When testing the effectiveness of a new drug, the investigator must establish
that the new drug has a better cure rate than the older treatment (or that there
are fewer side effects). He or she isnt interested in simply rejecting the null
hypothesis that the new drug has the same cure rate as the older treatment but
instead needs to know if it is better. In such situations, the alternative hypothesis
should state that the new drug cures a larger proportion of people than does the
older treatment. This is called a one-sided test of significance (or one-tailed test
of significance). Tests of significance can be one-sided if the investigator has an
indication of which way any change from the standard should go. This must be
decided before looking at the data.
The tests of significance discussed previously were two-sidedthe
investigator was interested in detecting a change from the standard in either
direction. For example, Jenny and Maya were testing to see if the percentage of
heads when a penny is spun is di erent from 0.5.
In testing a proportion, the alternative hypothesis can take one of three forms:
Ha: The percentage of successes p in the
population from which the sample
came is not p0. In symbols, p p0.
came is greater than p0. In symbols,
p > p0.
came is less than p0. In symbols,
p < p0.
The first alternative hypothesis is for a two-sided test; the latter two are for
one-sided tests. For a one-sided test, the P-value is found using one side only.
Example: One-Sided Test of Signifcance

The editors of a magazine have noticed that people seem to believe that a
successful life depends on having good friends. They would like to have a story
about this and use a headline such as Most Adults Believe Friends Are Important
for Success. So they have commissioned a survey to ask a random sample of
adults whether a successful life depends on having good friends. In a random
sample of 1027 adults, 53% said yes. Should the editors go ahead and use their
headline?
Solution
As always, the answer must include statistical evidence, in this case a test of
significance. You will use p0 = 0.5 because most means greater than 0.5.
507
22-03-2009 21:14
Lesson
19 de 28
The conditions for a

one-sided test are the
same as they are for a
two-sided test.
The null hypothesis for
a one-sided test also is
the same as it is for a
two-sided test.
1. The sample was a simple random sample of adults in the United States.
Both np0 = 1027(0.5) = 513.5 and n(1 p0) = 1027(0.5) = 513.5 are at
least 10. The population of all adults in the United States is much larger than
10(1027) = 10,270.
2. The claim (the alternative hypothesis) is that most adults believe this. This is
the same as more than 50%. Let p represent the proportion of adults in the
United States who believe that a successful life depends on having good friends.
The null hypothesis is that the percentage who believe this is 50% (or less):
H0: p = 0.5
The alternative
hypothesis for a
one-sided test is
different from that
for a two-sided test.
The test statistic for
a one-sided test is
the same as it is for
a two-sided test.
The P-value for a

one-sided test will be
half as large as it is for
a two-sided test.
The border-line equality, p = 0.5, can be used for the null hypothesis because
rejection of this value in favor of some value larger than 0.5 would certainly
imply that any smaller value of p would be rejected as well.
The alternative hypothesis is that this percentage is more than 50%:
Ha: p > 0.5
3. The test statistic is
As shown in Display 8.18, the P-value of 0.0274 is the probability of getting a

test statistic greater than 1.92 if the null hypothesis is true.
Display 8.18 The P-value for the successful life/friends problem.

State conclusion in
context with link to
computations.
4. The P-value (found in the upper tail) is fairly small. If the percentage of all
adults who believe a successful life depends on having good friends is 50%,
then the probability of getting a sample proportion of 53% or more is only
0.0274. Because getting a sample proportion, , of 0.53 or more is so unlikely
if the null hypothesis is true, reject the null hypothesis. This is quite strong
evidence that the true percentage must be greater than 50%. The editors
should feel free to use their headline.

D39. What would have been the conclusion if 40% of the sample in the preceding
example believed that friends are important for success?
D40. If everything else remains the same, is it easier to reject a false null
hypothesis with a one-sided test or with a two-sided test?
22-03-2009 21:14
Lesson
20 de 28
Summary 8.2: Tests of Significance

Suppose you want to decide whether it is reasonable to assume that your random
sample comes from a population with proportion of successes p0. A test of
significance for a proportion tells you whether the results from your sample are
so different from what you would expect that you should reject that value of p0
as a plausible value. If the sample proportion, , is relatively close to p0, you can
reasonably attribute the difference to chance variation and should not reject p0 as
the possible proportion of successes in the population. If the sample proportion,
, is relatively far from p0, you cannot reasonably attribute the difference to chance
variation and should reject p0 as the possible proportion of successes in the
population.
The four components of a test of significance are listed here, with some
suggestions about how to interpret each one.
Checking conditions. In real surveys, it is almost always the case that the
sample was not a simple random sample from the population. If the sample
comes from a more complicated design that uses randomization, inference is
possible. In other cases, convenience samples are usedpeople are chosen for
a sample because they were easy to find. In such cases, confidence intervals
and significance tests generally should not be used.
Writing hypotheses. Remember that, when you write the hypotheses, you are
not supposed to have seen the sample yet. Thus, the value from the sample, ,
does not appear in the null hypothesis.
If you use the symbol p in the null hypothesis, always say in words what it
represents in the context of the problem.
Using a one-sided alternative hypothesis makes it easier to reject a false
null hypothesisif you use the correct side. But you can use a one-sided
alternative hypothesis only if, before looking at the sample, you have an
indication of which direction from the standard the true proportion falls.
When in doubt about which type of test to use, use a two-sided test.
Doing computations. The test statistic measures how compatible the result
from the sample is with the null hypothesis. Test statistics typically come in
this form:
statistic parameter
________________________
standard deviation of statistic
In the case of testing a proportion, this becomes
z=
Writing a conclusion. If | z | is large and thus the P-value is small (smaller

than some significance level that you have decided on in advance), this is
evidence against the null hypothesis and so you should reject it. If the P-value
is larger than the predetermined value , do not reject the null hypothesis.
509
22-03-2009 21:14
Lesson
21 de 28
If you reject the null hypothesis, you are saying that the difference between
and p0 is so large that you cant reasonably attribute it to chance variation. If
you dont reject the null hypothesis, you are saying that the difference between
and p0 is small enough so that it looks like the kind of variation you would
expect from a binomial situation in which the proportion of successes is p0.
Always write your conclusion in the context of the situation. Do not simply
write Reject the null hypothesis or Do not reject the null hypothesis.
Explain what your conclusion means in terms of the context of the problem.
You should never write Accept the null hypothesis. The reason is that the
sample can tell you only whether the null hypothesis is reasonable. If you get a
very small P-value, then it isnt reasonable to assume that the null hypothesis
is true and you should reject it. However, if you get a large P-value, then the
null hypothesis is plausible but still might not give the exact value of p.
Finally, just because a result is statistically significant doesnt mean the
difference has any practical significance in the real world. In other words, the
difference might exist, but it might be so small as to have no impact.
Practice
P20. A 1997 article reported that two-thirds of
teens in grades 712 want to study more
about medical research. You wonder if this
proportion still holds today and decide to
test it. You take a random sample of 40 teens
and find that only 23 want to study more
about medical research. [Source: CNN Interactive
Story Page, April 22, 1997, www.cnn.com.]
a. What is the standard (the hypothesized

value, p0, of the population proportion)?
b. What is the sample proportion, ?
c. Use Display 8.9 to determine whether the
result is statistically significant. That is, is
there evidence leading you to believe that
the proportion today is different from the
proportion in 1997?
P21. A student took a 40-question truefalse test
and got 30 answers correct. The student
says, That proves I was not guessing at the
answers.
a. What is the standard, p0?
c. Use Display 8.9 to determine whether the
result is statistically significant.
d. Does the answer to part c prove that the
student was not guessing?
P22. This year, 75% of the seniors wanted extra

tickets for their graduation ceremony. To
anticipate whether there might be a change
in that percentage next year, the junior class
took a random sample of 40 juniors and
found that 32 would want extra tickets.
a. What is the standard, p0?
c. Use Display 8.9 to determine whether this
result is statistically significant.
d. Is this statistical evidence of a change?
The Test Statistic
P23. For the situation in P20, what value of the
test statistic should the junior class use to
test whether there is statistical evidence of a
change?
P24 Forty-five dogs and their owners, chosen at
random, were photographed separately. A
judge was shown a picture of each owner and
pictures of two dogs and asked to pick the
dog that went with the owner. The judge was
right 23 times. What value of the test statistic
should be used to test whether the judge did
better than could reasonably be expected just
by guessing? [Source: Based on a study published in
Psychological Science, 2004.]
22-03-2009 21:14
Lesson
22 de 28
P-Values
P25 Find the P-value for Miguel and Kevins test
statistic, 3.16, from pages 493494. Write a
sentence explaining what this P-value means
in the context of their situation.
P26. Suppose you re-create the chimpanzee
experiment in the example on page 494
with an apparatus that tests whether cats can
select the rake that results in the food. You
test 50 randomly chosen cats and find that
28 select the rake that pulls in the food. Find
the P-value for this test and interpret it.
P27. According to the U.S. Census Bureau,
about 69% of houses across the country
are occupied by their owners. Your class
randomly samples 50 houses in your
community and finds that 30 houses are
occupied by the owners. Follow these steps
to determine whether your community
differs from the nation as a whole as to
the percentage of houses that are owneroccupied.
a. State the null hypothesis to be tested.
What is the alternative hypothesis?
b. Find the value of the test statistic.
c. Find the P-value for this test and explain
what it means.
d. Write a conclusion, based on your
analysis, in the context of the problem.
P28. Which of statements AE is the best
explanation of what is meant by the P-value
of a test of significance?
A. Assuming that you had a random
sample and the other conditions for a
significance test are met, the P-value is
the probability that H0 is true.
B. Assuming that H0 is true, the P-value is
the probability of observing a value of a
test statistic at least as far out in the tails
of the sampling distribution as is the
value of z from your sample.
C. The P-value is the probability that H0
is false.
D. Assuming that the sampling distribution
is normal, the P-value is the probability
that H0 is true.
E. Assuming that H0 is true, the P-value is

the probability of observing the same
value of z that you got in your sample.
P29. If your null hypothesis is that spinning or
flipping a coin is fair, find the test statistic,
z, for each of these situations and give your
conclusion. Use = 0.05.
a. A class spun quarters 500 times and got
194 heads.
b. A class flipped quarters 500 times and got
265 heads.
P30. Use Table A on page 824 or your calculator
to answer these questions.
a. What critical value is associated with a
significance level of 0.12?
b. What significance level is associated with
critical values of z* of 1.73?
P31. Use Table A on page 824 or your calculator
to answer these questions.
a. What level of significance is associated
with critical values of z* of 2.576?
b. What critical values are associated with
a significance level of 2%?
P32. Using the data in P29 that there were
194 heads out of 500 spins, carry out the
four steps in a test of the null hypothesis
that spinning a quarter is fair.
P33. Suppose that, in a random sample of
500 bookstores across the United States,
265 also sell DVDs. Carry out the four steps
in a test of the null hypothesis that half the
bookstores in the United States sell DVDs.
P34. State an appropriate null hypothesis for each
of these situations.
a. You wonder if a student is guessing on a
multiple-choice test of 60 questions where
each question has five possible answers.
b. You wonder if there is an equal proportion
of male and female newscasters on local
television stations.
511
22-03-2009 21:14
Lesson
23 de 28
c. You wonder if people who wash their cars

once a week are most likely to wash them
on Saturday.
P35. The United States 2000 Census found that
4% of all households were multigenerational
(consisting of three or more generations of
parents and their children). You want to test
the null hypothesis that this percentage is the
same this year as it was in 2000. You write
the null hypothesis as p = 0.04. Which of
these statements best describes what p stands
for? [Source: www.census.gov.]
A. the proportion of all households that

were multigenerational in 2000
B. the proportion of all households that are
multigenerational this year
C. the proportion of multigenerational
households in the sample in 2000
D. the proportion of multigenerational
households in the sample this year
E. the proportion of all multigenerational
households in both 2000 and this year
Types of Errors
P36. Hila is rolling a pair of dice to test whether
they land doubles of the time. She doesnt
know it, but the dice are actually fair. She will
use a significance level, , of 0.05. Hila rolls
the dice 100 times and gets doubles 22 times.
a. What conclusion should Hila come to?

b. Did Hila make an error? If so, which type?
P37. Jeffrey and Taline each want to test if the
proportion of adults in their neighborhood
who have graduated from high school is
0.94, as claimed in the newspaper. Jeffrey
takes a random sample of 200 adults and
uses = 0.05. Taline takes a random sample
of 500 adults and uses = 0.05. Suppose the
newspapers percentage is actually right.
a. Is it possible for Jeffrey or Taline to make
a Type I error? If so, who is more likely to
do so?
b. Is it possible for Jeffrey or Taline to make
a Type II error? If so, who is more likely
to do so?
P38. Jeffrey and Taline each want to test if the
proportion of adults in their neighborhood
who took chemistry in high school is 0.25,
as claimed in the newspaper. Jeffrey takes
a random sample of 200 adults and uses
= 0.05. Taline takes a random sample of
500 adults and uses = 0.05. Suppose the
newspapers percentage is actually wrong.
do so?
to do so?
P39. A statistical test is designed with a significance
level of 0.05 and a sample size of 100. A
similar test of the same null hypothesis is
designed with a significance level of 0.10 and
a sample size of 100. If the null hypothesis is
false, which test has the greater power? Sketch
a diagram to support your answer.
P40. A statistical test is designed with a significance
level of 0.05 and a sample size of 100. A
similar test of the same null hypothesis is
designed with a significance level of 0.05 and
a sample size of 200. If the null hypothesis is
false, which test has the greater power? Sketch
a diagram to support your answer.
22-03-2009 21:14
Lesson
24 de 28
P41. Suppose a medical researcher wants to

test the hypothesis that the proportion of
patients who develop undesirable side effects
from a certain medication is the reported
6%. The researcher is pretty sure that the
true proportion is close to either 8% or 10%.
She has a random sample of patients using
the medication and chooses a 5% level of
significance. Is the power of the test larger if
the true proportion is 8% or 10%? Sketch a
diagram to support your answer.
P42. In a poll of 1000 randomly sampled adults
in the United States, 46% said they were
satisfied with the quality of K12 education
in the nation. Does this imply that less than
a majority of adult residents are satisfied
with the quality of education? Answer this
question by working through the steps in
parts ad. [Source: Gallup, Slim Majority Dissatisfied with
Education in the U.S., 2005, poll.gallup.com.]
a. Verify that the conditions for the test are

satisfied.
b. What is the null hypothesis? What is the
most appropriate alternative hypothesis?
c. Calculate a test statistic and the
corresponding P-value.
d. Write a conclusion in the context of the

original question.
P43. In the same poll of adults as described in
P42, 51% of the 1000 respondents said that
they were dissatisfied with the quality of
K12 education in schools today. Does this
imply that a majority of adult residents are
dissatisfied with the quality of education?
Work through the same steps as in P42.
P44. You claim that the percentage of teens in
your community who know that epilepsy
is not contagious is larger than the national
percentage, 51%. (See E5 on page 486.) You
take a random sample of 169 teens in your
community and find that 55% know that
epilepsy is not contagious. Carry out the
steps of a test of your claim.
P45. Determine whether each statement is true
or false.
a. Always look at the data before writing the
null and alternative hypotheses.
b. All else being equal, using a one-sided
test will result in a larger P-value than
using a two-sided test.
c. The P-value is the probability that the
null hypothesis is true.
d. A statistically significant result means the
P-value is small.
Exercises
E25. A random sample of dogs was checked to see
how many wore a collar. The 95% confidence
interval for the percentage of all dogs that
wear a collar turned out to be from 0.82 to
0.96. Which of these is not a true statement?
A. You can reject the hypothesis that 75% of
all dogs wear a collar.
B. You cannot reject the hypothesis that
90% of all dogs wear a collar.
C. If 90% of all dogs wear a collar, then you
are reasonably likely to get a result like
the one from this sample.
D. If 75% of all dogs wear a collar, then you
are reasonably likely to get a result like
the one from this sample.
E. The true proportion of dogs that wear a

collar may or may not be between 0.82
and 0.96.
513
22-03-2009 21:14
Lesson
25 de 28
E26. A Gallup poll of a recent year estimated

that 27% of adult Americans approved and
67% disapproved of the way Congress was
handling its job. The margin of error was
given as 3%. Which of these is not a true
statement? [Source: Gallup, www.gallup.com, 2006.]
A. You can reject the hypothesis that 35%
of adult Americans approved of the way
Congress was handling its job on that date.
B. You cannot reject the hypothesis that
70% of all Americans disapproved of the
way Congress was handling its job on
that date.
C. The true proportion of all Americans
who approved of the way Congress was
handling its job on that date must lie
D. If, in fact, 25% of all Americans approved
of the way Congress was handling its job
on that date, then the observed sample
result is reasonably likely.
E. If, in fact, 75% of all Americans
disapproved of the way Congress was
handling its job on that date, then the
observed sample result is not reasonably
likely.
E27. Suppose 22 students out of a random sample
of 40 students carry a backpack to school.
Follow steps ac to test the claim that 60% of
the students in the school carry backpacks
to class.
a. State the null hypothesis in words and in
symbols.
b. Calculate the value of the test statistic.
Calculate the P-value for the test. Use this
P -value in a sentence that explains what it
represents.
c. What is your conclusion? Explain it in the
context of this problem.
E28. Follow the three steps in E27 for a random
sample of 40 students, of which 31 carry
backpacks to class.
E29. The Gallup Organization asked 1003 adults
to answer this question: If you could pick
only one specific problem for science to
solve in the next 25 years, what would it be?

Thirty percent responded a cure for cancer.
Use a test of significance to determine
whether it is plausible that, if you could ask
all adults in the United States this question,
only 25% would pick a cure for cancer. [Source:
Gallup, June 2001, www.publicagenda.org.]
E30. Americans increasingly favor raising the

driving age, a USA Today/CNN/Gallup poll
has found. Nearly two-thirds61%say
they think a 16-year-old is too young to have
a drivers license. Only 37% of those polled
thought it was okay to license 16-year-olds,
compared with 50% who thought so in
1995. A slight majority, 53%, thinks teens
should be at least 18 to get a license. The poll
of 1002 adults was conducted Dec. 1719,
2004. [Source: USA Today polls, Is 16 Too Young to
Drive a Car? March 2, 2005.]
If you asked all adult residents of the United

States if they thought that a 16-year-old is
too young to have a drivers license, is it
plausible that only 55% would say yes?
E31. A Gallup poll asked 506 adult Americans
this question: Increased efforts by business
and industry to reduce air pollution
might lead to higher prices for the things
consumers buy. Would you be willing to pay
$500 more each year in higher prices so that
industry could reduce air pollution, or not?
The report said, Almost two-thirds (63%)
claim they would be willing to pay $500
more per year for this purpose. Following
the four steps of a significance test,
determine if there is convincing evidence
that the true proportion of Americans who
would be willing to pay $500 more for this
purpose differs from two-thirds. [Source: Gallup,
www.gallup.com, March 16, 2001.]
E32. In August 2005, a Harris poll asked, Do

you agree or disagree with this statement?
Protecting the environment is so important
that requirements and standards cannot be
too high, and continuing environmental
improvements must be made regardless of
cost. Of 1217 adults surveyed nationwide,
74% agreed, 24% disagreed, and 1%
22-03-2009 21:14
Lesson
26 de 28
were unsure. Following the four steps

of a significance test, determine if this is
convincing evidence that the true proportion
of the nations adults who agree with the
statement is different from two-thirds. Use
= 0.01. [Source: The Harris Report, August 2005,
www.pollingreport.com.]
E33. Suppose that when flipping a penny, the

probability of getting heads is 0.5. Suppose
100 people perform a test of significance
with this penny. Each person plans to reject
the null hypothesis that p = 0.5 if their
P-value is less than 0.05. How many of these
people do you expect to make a Type I error?
E34. Fifty laboratories across the country are
conducting similar tests on the possible
side effects of a certain medication. Each
laboratory tests the null hypothesis that no
more than 15% of the subjects taking the
medication develop undesirable side effects,
and each test is done at the 10% significance
level. One of these two questions can be
answered. Answer that one, and explain why
you cannot answer the other.
a. How many of the laboratories do you
expect to reject the null hypothesis if
it is true?
b. How many of the laboratories do you
expect to reject the null hypothesis if
it is false?
E35. Jeffrey and Taline each want to test whether
the proportion of adults in their state who
have graduated from college is 0.6, as
claimed in the newspaper. Jeffrey takes
a random sample of 200 adults and uses
= 0.01. Taline takes a random sample of
200 adults and uses = 0.05. Suppose the
newspapers percentage is actually right.
do so?
to do so?
E36. Jeffrey and Taline each want to test whether
the proportion of adults in their state who
have gone to nursery school is 0.55, as
claimed in the newspaper. Each takes a

random sample of 200 adults. Jeffrey uses
= 0.01. Taline uses = 0.05. Suppose the
newspapers percentage is actually wrong.
do so?
to do so?
E37. A friend wants to do a two-sided test of
whether spinning pennies is fair. His null
hypothesis is that the percentage of heads
when a penny is spun is 50%. He plans to do
20 spins and use a 5% level of significance.
a. Use the binomial probability formula
on page 385 to construct the binomial
distribution for n = 20 and p = 0.5.
b. Use your distribution from part a to find
the numbers of heads that would result in
rejecting the null hypothesis.
c. You are pretty sure that the true percentage
of heads is about 40%. Supposing that is
the case, use exact binomial probabilities
to compute the probability that your friend
will be able to reject the null hypothesis.
d. What should you suggest to your friend?
E38. A quality improvement plan for a business
office calls for selecting a sample of ten
invoices that came in during a specified week
and counting the number that have not been
paid within 30 days. Initially, management
wants to test to see if the proportion of
invoices not paid on time exceeds 30%.
a. Suppose the plan calls for rejecting
the null hypothesis if the number of
invoices not processed on time is six or
greater out of the sampled ten. Use the
binomial probability formula on page 385
to construct the binomial distribution
for n = 10 and p = 0.3. Then use this
distribution to find the significance level
for this test.
b. Suppose that, in fact, the true rate of
invoices not processed on time is 40%.
What is the power of the test described in
515
22-03-2009 21:14
Lesson
27 de 28
part a to detect this alternative? Use exact

binomial probabilities to calculate this
power.
E39. In P24 on page 510, you computed a test
statistic for this study: Forty-five dogs
and their owners, chosen at random, were
photographed separately. A judge was shown
a picture of each owner and pictures of two
dogs and asked to pick the dog that went
with the owner. The judge was right 23
times.
a. If you use a significance level of 0.05, do
you reject the null hypothesis that, in the
long run, the judge will select the correct
dog half the time?
b. Suppose you compute the 95% confidence
interval for the proportion of times the
judge will select the correct dog. Will the
value of p0 from the null hypothesis be in
this confidence interval? Explain.
c. Compute the confidence interval to check
your answer to part b.
d. Explain how the margin of error for the
confidence interval is related to the power
of the test.
e. How would you suggest that the
investigator get more power for this test?
E40. To perform a significance test for a proportion,
you must decide on the hypotheses (null and
alternative), the level of significance, and the
sample size. Explain the effects of each of these
components on the power of the test.
E41. At the beginning of this section, you read
that 2% of barn swallows have white feathers
in places where the plumage is normally blue
or red. However, about 16% of barn swallows
captured around Chernobyl after 1991
had such genetic mutations. The number
captured around Chernobyl was relatively
large266 barn swallows.
a. If you want to determine whether the
increase in the proportion of mutations
is statistically significant, should this be a
one-sided or two-sided test?
b. What condition(s) of the significance test
for a proportion are not satisfied?
c. Display 8.19 shows the sampling

distribution of the sample proportion,
p, for n = 266 and p = 0.02. Use it to
estimate the P-value for the test you
selected.
Display 8.19 Sampling distribution of

and p = 0.02.
for n = 266
d. What should the researchers conclude?

E42. A psychologist was struck by the fact that
at square tables many pairs of students sat
on adjacent sides rather than across from
each other and wondered if people prefer
to sit that way.The psychologist collected
some data, observing 50 pairs of students
seated in the student cafeteria of a California
university. The tables were square and had
one seat available on each of the four sides.
The psychologist observed that 35 pairs sat
on adjacent sides of the table, while 15 pairs
sat across from each other. He wanted to
know if the evidence suggested a preference
by students for adjacent sides, as opposed to
simply random behavior. [Source: Joel E. Cohen,
Turning the Tables, in Statistics by Example: Exploring
Data, edited by Frederick Mosteller et al. (Reading, Mass.:
Addison-Wesley, 1973), pp. 8790.]
22-03-2009 21:14
Lesson
28 de 28
a. What is the probability that if two

students sit down randomly at a square
table, they sit on adjacent sides?
b. What is an appropriate null hypothesis
to test whether students seating pattern
is random, as opposed to their having a
preference to sit on adjacent sides?
c. What is the appropriate alternative
hypothesis?
d. Does this situation meet the conditions
for a significance test?
e. What is the test statistic?
f. Find the P-value for this test.
g. Write a conclusion for the psychologist.
E43. A psychology professor flashed a pair
of head shots before subjects and asked
them to identify the face that displayed
more competence. The subjects didnt
know it, but the head shots were the two
opposing candidates in 600 recent House of
Representative elections. It turned out that
66.8% of the candidates who were perceived
as more competent by the subjects were
winners in the election. The article says that
the statistical test used tests the proportion
of correctly predicted races against the
chance level of 50%. [Source: A. Todorov, A. N.
E44. Suppose a test rejects this null hypothesis in

favor of the alternative:
H0: p = p0 versus Ha: p > p0
a. Explain why it also would have been

reasonable to reject the null hypothesis if
it had been of this form: H0: p p0.
b. Would the null hypothesis necessarily
have been rejected if the alternative had
been Ha: p p0? Explain.
E45. Suppose that 42 out of 80 randomly selected
students prefer hamburgers to hot dogs and
that 38 prefer hot dogs to hamburgers.
a. Test the null hypothesis that the
percentage of all students who prefer
hamburgers is 0.55.
b. Test the null hypothesis that the percentage
of all students who prefer hot dogs is 0.55.
c. How can you reconcile your conclusions
in parts a and b?
Now suppose you go out and get a larger
random sample of 800 students. Suppose
that 420 of these students prefer hamburgers
to hot dogs and that 380 prefer hot dogs to
hamburgers.
d. Test the null hypothesis that the
percentage of all students who prefer
Mandisodza, A. Goren, and C. C. Hall, Inferences of
hamburgers is 0.55.
Competence from Faces Predict Election Outcomes, Science
308 (2005): 162326.]
e. Test the null hypothesis that the percentage
of all students who prefer hot dogs is 0.55.
a. To determine the significance of this
result, should a one-sided or two-sided
f. How can you reconcile your conclusions
test be used?
in parts a, b, d, and e?
b. What is ? What is p0?
c. The test statistic given in the article is
8.25. Is this correct?
d. The P-value reported was P < 0.001. Is
this correct?
e. Does the sample size make an important
contribution to the power of the test in
this situation? Explain.
517
22-03-2009 21:14
Lesson
1 de 8
E46. In an election between two men, does

the taller man tend to win more often?
Display 8.20 gives the results of the U.S.
presidential elections from 1900 to 2004.
a. Display these data graphically.
b. Do these data meet the conditions of a
significance test?
c. If you write a slightly different type of
null hypothesis, you can proceed with a
test of significance and write a reasonable
conclusion. Begin by writing a null

hypothesis to test whether a sequence
of coin flips is likely to result in as many
heads as there were times when the taller
man won. hat is, you will test whether
the fact that the taller man has won more
often can reasonably be attributed to
chance. If so, there is nothing to explain.
If not, an explanation is called for.
Display 8.20 Heights of presidential candidates. [Source: Paul M. Sommers, Presidential Candidates
Who Measure Up, Chance 9, no. 3 (1996): 2931.]
8.3
Many decisions are
based on comparisons.
A Confidence Interval for the Difference of Two Proportions

In Section 8.1, you found a confidence interval estimate for the proportion of
successes in a single population. The more common and important situation
involves taking two samples independently from two different populations
with the goal of estimating the size of the difference between the proportion
of successes in one population and the proportion of successes in the other.
A recent poll of 29,700 U.S. households found that 63% owned a pet. The
percentage in 1994 was 56%. [Source: American Pet Products Manufacturers Association, www
.appma.org.] The two populations are the households in the United States in 1994
and the households now. The question you will investigate is What was the
change in the percentage of U.S. households that own a pet? The obvious answer
22-03-2009 21:15
Lesson
2 de 8
is that the percentage increased by 7 percentage points, but this is only an estimate
because 7% is the difference of two sample percentages that probably are not
exactly equal to the population percentages. In this section, you will learn how
to find the margin of error to go with the difference, 7%.
A confidence interval for the difference of two proportions, p1 p2, where
p1 is the proportion of successes in the first population and p2 is the proportion
of successes in the second population, has this familiar form:
(
z* standard error of (
Here and are the proportions of successes in the two samples. Substituting
what you have so far into the formula gives the 95% confidence interval for the
difference between the proportion of U.S. households that own pets now and
the proportion that owned pets in 1994:
0.07
1.96 standard error of (
The hard part, as always, is estimating the size of the standard error.
The Formula for the Confidence Interval

The statistic of interest is a difference of random variables, and you know from
Section 6.1 that the mean of a difference is the difference of the means. If the
variables are independent, the variance of the difference is the sum of the
variances. In symbols,
In addition, under the independence assumption, the sampling distribution of the

differences will be approximately normal if the sample sizes are suitably large.
Taking the square root of the variance, the standard error is
From Section 7.3, the standard error of the distribution of the sample proportion
is
which can be estimated by
where n1 is the sample size. Similarly, the standard error of the distribution of the
sample proportion is
which can be estimated by
where n2 is the sample size.
519
22-03-2009 21:15
Lesson
3 de 8
Substituting these into the formula for the standard error of a difference, you
can estimate the standard error of the difference of two proportions:
Now you can write the complete confidence interval for the difference
between two proportions.
Confidence Interval for the Difference of Two Proportions
The confidence interval for the difference, p1 p2, of the proportion of
successes in one population and the proportion of successes in a second
population is
(
z*
where is the proportion of successes in a random sample of size n1 taken

from the first population and is the proportion of successes in a random
sample of size n2 taken from the second population. ( The sample sizes dont
have to be equal.)
The conditions that must be met in order to use this formula are that
the two samples are taken randomly and independently from two
populations
each population is at least 10 times as large as its sample size
n1 , n1(1
), n2 , and n2(1
) are all at least 5
Example: A Difference in Pet Ownership?

As described at the beginning of this section, a recent pet ownership survey
found that 63% of the 29,700 U.S. households sampled own a pet. A 1994 survey,
taken by the same organization, found that 56% of the 6,786 U.S. households
sampled owned a pet. Find and interpret a 95% confidence interval for the
difference between the proportion of U.S. households that own a pet now and the
proportion of U.S. households that owned a pet in 1994.
Check conditions.
Solution
The two samples can be considered random samples. They were taken
independently from the population of U.S. households in two different years.
The number of U.S. households in each year is larger than 10 times 29,700.
Finally, each of the following is at least 5.
n1
= 29,700(0.63) = 18,711
n2
= 6,786(0.56)
3,800
n1(1
n2(1
) = 29,700(0.37) = 10,989
) = 6,786(0.44)
2,986
22-03-2009 21:15
Lesson
4 de 8
Do computations.
The 95% confidence interval for the difference of the two population
proportions, p1 and p2, is
(
z*
= ( 0.63 0.56 )
= 0.07
1.96
0.013
Alternatively, you can write this confidence interval as (0.057, 0.083).

You also can calculate this interval using your calculator by entering each
sample size and the number of successes in each sample. [See Calculator Note 8E.]
Write your interpretation

in context.
You are 95% confident that the difference in the two rates of pet ownership
is between 0.057 and 0.083. This means that it is plausible that the difference in
the percentage of households that own pets now and the percentage in 1994 is
5.7%. It is also plausible that the difference is 8.3%. Note that a difference of 0
does not lie within the confidence interval. This means that if the difference in
the proportion of pet owners now and in 1994 actually is 0, getting a difference
of 0.07 in the samples is not at all likely. Thus, you are convinced that there was a
change in the percentage of households that own a pet.
Which sample
proportion should be
p1 and which p2 ?
It doesnt matter which sample proportion you call and which you call , as
long as you remember which is which. Most people like to assign the larger one to
be so that the difference,
, is positive. If you like working with negative
numbers, do it the other way!

D41. A survey found that 45% of households own a dog, 34% own a cat, and 20%
own both.
a. What percentage of households in the survey own neither a cat nor a dog?
b. Are owning a dog and owning a cat independent events?
c. Should you use the techniques of this section to estimate the difference
in the percentage of households that own a dog and the percentage that
own a cat? Explain.
Activity 8.3a will help you understand how confidence intervals for
differences of proportions may vary from sample to sample.
521
22-03-2009 21:15
Lesson
5 de 8
Whos Yellow?
What youll need: a sample of 50 Skittles, a sample of 50 Milk Chocolate
M&Ms (or bags prepared by your instructor)
1. Count the number of yellow candies in your sample of Skittles and in your
sample of M&Ms. Compute the two sample proportions.
2. Find the 95% confidence interval for the difference in the two sample
proportions:
proportion of Skittles that are yellow proportion of M&Ms that are yellow
3.
4.
5.
6.
7.
(Every student should subtract in this order, even if the difference is

negative.)
Draw your confidence interval on a large copy of Display 8.21. Other
members of your class should draw their confidence intervals on the same
chart.
Compare the centers of the confidence intervals. Are they close to each
other, or do they vary a lot? Why do they vary?
Compare the widths of the confidence intervals. Are the widths fairly
equal, or do they vary a lot? Why is there any variability in the width?
How many of the confidence intervals would you expect to capture the true
difference in proportions?
Skittles have 20% yellow candies. Milk Chocolate M&Ms have 14% yellow
candies. How many of the confidence intervals actually captured the true
difference in these proportions?
Display 8.21 Chart for recording 95% confidence intervals for the
difference in the proportion of yellows in Skittles
and M&Ms.
22-03-2009 21:15
Lesson
6 de 8
Summary 8.3: A Confidence Interval for

the Difference of Two Proportions
When the sample sizes, n1 and n2, are large enough, the sampling distribution of
is approximately normal, is centered at , and has standard error
=
In practice, you estimate
using the two sample proportions. This leads
to a confidence interval for the difference of the two population proportions of

(
z*
where z* are the values that enclose an area in a normal distribution equal
to the confidence level, which typically is 0.95. Dont forget to check the three
conditions on page 520 that must be met to use this confidence interval, especially
the condition that the two samples were taken independently from two different
populations.
When interpreting, say, a 95% confidence interval, you should say that you
are 95% confident that the difference between the proportion of successes in the
first population and the proportion of successes in the second population is in
this confidence interval. (However, say this in the context of the situation.) If the
confidence interval includes 0, you cant conclude that there is any difference in
the proportions in the two populations.
Practice
P46. Suppose the surveys on pet ownership
described in the example on page 520 had
used sample sizes of only 100 people. That is,
a 1994 survey of 100 U.S. households found
that 56% owned a pet, and this year a survey
of 100 U.S. households found that 63%
owned a pet.
a. Check the conditions for constructing a
confidence interval for the difference of
two proportions.
b. Compute and interpret the confidence
interval.
c. Is 0 in the confidence interval? What does
your answer imply?
P47. A poll of 549 teenagers asked if it was
appropriate for parents to install a special
device on the car to allow parents to monitor
teenagers driving speeds. Among the

325 1315-year-olds in the sample, 50%
responded yes to this question. Among the
224 1617-year-olds, only 28% responded
yes. [Source: Gallup, Teens Slow to Support Speed Monitors,
2005, www.poll.gallup.com.]

two proportions.
b. Estimate the difference between the
proportion of all 1315-year-olds who
would answer yes to this question and
the proportion of all 1617-year-olds
who would answer yes.
c. Interpret the resulting interval in the
context of the problem.
d. Is 0 in the condence interval? What does
your answer imply?
523
22-03-2009 21:15
Lesson
7 de 8
Exercises
E47. A statistics instructor kept track of students
grades on her final examination and whether
they did the exam in pencil or in pen. Of
the 75 students who used a pencil, 63 got
grades of 70% or higher. Of the 19 students
who used a pen, 10 got grades of 70% or
higher. Her 95% confidence interval for the
difference is (0.07, 0.55). Assume that these
students can be considered random samples
from the students who could take this
instructors final examination. Which of these
choices are the best two interpretations of this
A. The teacher must observe more students,
because this confidence interval is too
wide to be of any possible use.
B. The teacher has evidence that a larger
proportion of students who use a pencil
score 70% or higher on her final than
students who use a pen.
C. If the teacher could observe all students
who could take her final examination,
she is 95% confident that the difference
in the proportion who use a pencil and
get a grade of 70% or higher and the
proportion who use a pen and get a grade
of 70% or higher is between 0.07 and 0.55.
D. If the teacher could observe all students
who could take her final examination, she
is 95% confident that between 7% and
55% of them will get 70% or higher.
E48. According to the American Automobile
Association, a study found that 85.8% of
207,638 drivers who were involved in a fatal
car accident had a valid drivers license. Of
16,226 motorcycle drivers involved in a fatal
accident, 80.1% had a valid drivers license.
A 95% confidence interval for the difference
in proportions is (0.05, 0.06). You can
consider these random samples of drivers
involved in fatal accidents. Which of these
choices are the best two interpretations of

this confidence interval? [Source: AAA Foundation
for Traffi c Safety, Unlicensed to Kill: The Sequel, 2003, www
.aaafoundation.org.]
A. You can be confident that, if you looked

at all drivers involved in fatal accidents,
a larger percentage of drivers of cars
would have a valid license than drivers of
motorcycles.
B. If all drivers involved in a fatal accident
were observed, you can be 95% confident
that the true percentage of drivers who
dont have a drivers license would be
between 5% and 6%.
C. Thjs confidence interval is so narrow
because the sample sizes are so big.
D. If all drivers who dont have a valid
drivers license were observed, you are
95% confient that between 5% and
6% more of the unlicensed drivers of
motorcycles would have a fatal accident
than unlicensed drivers of cars.
E49. Researchers at the University of California,
San Diego, found that when given a choice of
two dogs, judges correctly matched 16 out of
25 purebred dogs with their owners. When
the dogs were mutts, only 7 dogs out of 20
were correctly matched with their owners.
[Source: www.newscientist.com.]

two proportions.
difference of proportions.
d. Is 0 in the confidence interval? What does
your answer imply?
e. What would you recommend that the
researchers do?
22-03-2009 21:15
Lesson
8 de 8
E50. The USC Annenberg School Center for the

Digital Future found that 66.9% of Americans
used the Internet in 2000 and 78.6% used the
Internet in 2005. Assume that the samples
were independently and randomly selected
and that the sample size was 2000 in both
years. [Source: www.digitalcenter.org.]
two proportions.
your answer imply?
E51. A Harris poll asked this question of about
425 men and 425 women: When you get a
sales or customer service phone call from
someone you dont know, would you prefer
to be addressed by your first name or by
your last name, or dont you care one way
or the other? Twenty-three percent of the
men and 34% of the women said they would
prefer being addressed by their last name.
Although this was not quite the case, you
can assume that the samples were random
and independent. [Source: Harris, November 14, 2001,
www.harrisinteractive.com.]
in the United States responded that 16 was

the correct age. In 2004, only 35% of 1000
randomly sampled adults responded this
way. [Source: USA Today/CNN/Gallup poll, December
1719, 2004.]

two proportions.
b. Estimate the true difference between the
1995 and 2004 proportions favoring 16 as
the correct age to begin driving.
your answer imply?
E53. In a recent national survey, 15,200 high
school seniors in 128 schools completed
questionnaires about physical activity.
Male students were significantly more
likely (59%) than female students (48%) to
have played on sports teams run by their
school during the 12 months preceding the
survey. Check the accuracy of the phrase
significantly more likely, assuming that
there were equal numbers of male and
female students in this survey and that
the samples are equivalent to independent
simple random samples. [Source: Participation in
School Athletics, www.childtrendsdatabank.org.]

two proportions.
d. Is 0 in the conffdence interval? What does
your answer imply?
E52. At what age do you think people should
be permitted to have a drivers license? In
1995, 46% of 1000 randomly sampled adults
525
22-03-2009 21:15
Lesson
1 de 13
E54. In the report referenced in E53, the

participation of females in school athletics
increased from 47% in 1991 to 48% recently.
The total sample size (males and females)
for 1991 was approximately 15,400. Does
this increase represent a significant change
in the level of participation for females?
E55. What is the effect of an increase in the sample
sizes on the width of a confidence interval for
the difference of two proportions?
E56. Statements I and II are interpretations
of a 95% confidence interval for a single
proportion. Write similar statements for a
95% confidence interval for a difference of
two proportions.
I. A 95% confidence interval consists of
those population proportions p for which
the proportion from the sample, , is
reasonably likely to occur.
II. If you construct a hundred 95% confidence
intervals, you expect that the population
proportion p will be in 95 of them.
E57. The formula for the confidence interval for
the difference of two proportions involves z.
Why is it okay to use z in this case?
E58. What is suggested if the confidence interval
for the difference of two proportions
a. includes 0?
b. does not include 0?
8.4
E59. A Harris online poll asked 99 online

respondents, Would you support the
proposed bill that would eliminate the U.S.
penny? There were 22 (or 22.2%) yes votes
and 77 (or 77.8%) no votes. [Source: February 25,
2002, www.harriszone.com.] A student computed a
95% confidence interval for the difference of
the proportion of the population answering
yes and the proportion answering no as
follows:
)
z*
= (0.222 0.778)
1.96
= 0.556
0.116
Assuming that the respondents can be

considered a random sample from some
population, is the students method correct?
If so, write an interpretation of this interval.
If not, do a more appropriate analysis.
E60. In the USA Today poll on driving ages
described in E52, 35% of 1000 randomly
sampled adults chose 16 as the preferred
initial driving age while 42% chose 18.
Can the difference of the proportion in the
population who prefer 16 and the proportion
in the population who prefer 18 be estimated
by the confidence interval developed in this
chapter? Explain why or why not.
A Significance Test for the Difference of Two Proportions

In the previous section, you learned how to estimate the size of the difference of two
proportions. But sometimes you must decide between two alternatives. For example:
Are snowboarders or skiers more likely to be seriously injured?
Have opinions about the general state of the country changed over the past
5 years?
Is there a difference between the proportions of teen boys and girls who favor
dress codes?
22-03-2009 21:16
Lesson
2 de 13
Should the observed

difference be attributed
to chance alone?
The size of any difference involved is not the issue, as it was in Section 8.3. All
you care about is whether you have enough evidence to conclude that there is a
difference. In this section, you will learn to perform a test of significance in order
to decide if the observed difference can reasonably be attributed to chance alone
or if the difference is so large that something other than chance variation must be
causing it.
A Sampling Distribution of the Difference

Before you learn how to use a test of significance to answer questions like those on
the previous page, you will learn about what sizes of differences you should expect
in your samples when there is no actual difference in the proportions of successes
in the two populations. For example, as candy connoisseurs are well aware,
there is no difference in the percentage of orange candies in Skittles and in Milk
Chocolate M&Ms. Yet, when you take random samples of each kind of candy, you
usually get a difference in the two proportions of orange candies in your samples.
looks
In Activity 8.4a, you will examine what a sampling distribution of
like when equals .
Differences in Proportions of Orange

What youll need: a sample of 50 Skittles,
a sample of 50 Milk Chocolate M&Ms
(or bags prepared by your instructor)
This activity
demonstrates
chance variation.
1. Count the number of orange candies

in your sample of Skittles and in your
sample of M&Ms. Compute the two
sample proportions.
2. Compute the difference in the two
sample proportions:
proportion of Skittles that are orange
proportion of M&Ms that are orange
3.
4.
5.
6.
7.
(Every student should subtract in this

order, even if the difference is negative.)
Each member of your class should mark his or her difference in sample
proportions above a number line so that everyone can see the distribution.
Repeat with fresh samples until you have at least 100 differences.
What is the shape of the sampling distribution?
Where should the sampling distribution be centered? Where is it centered?
What, approximately, is the standard error of the sampling distribution?
527
22-03-2009 21:16
Lesson
3 de 13
Approximate sampling distributions (constructed from 5000 pairs of samples)

for samples of size 30, 50, and 100 are shown in Display 8.22. In each case, the
proportion of successes in the first population is 0.2, and the proportion of
successes in the second population is 0.2.
Display 8.22 Simulated sampling distributions of

sample sizes.
when
= 0.2 for various
Note three facts about these approximate sampling distributions:

Each sampling distribution is approximately normal in shape and becomes
more so with larger sample sizes.
The mean of each sampling distribution is at = 0.2 0.2 = 0.
The formula from Section 8.3 can be used to find the standard error:
=
Display 8.23 lists the SEs computed from the formula and from the
approximate sampling distributions in Display 8.22. They match very closely.
22-03-2009 21:16
Lesson
4 de 13
Display 8.23 SEs from the formula and from the simulation.

D42. Use your calculator or computer to take 200 pairs of random samples, both
samples of size 25 and both from populations with 35% successes.
a. Construct a plot of the differences of the two sample proportions.
b. From your plot, estimate the mean and standard error of the sampling
distribution of
.
c. Find the theoretical value of the mean and standard error of the sampling
distribution of
. Compare your answers to your estimates in part b.
The Theory of a Significance Test for the Difference

of Two Proportions
Lets see how these ideas apply to a real sample survey. Think about the left-handers
you know or know about. Queen Victoria was left-handed, and so were Julius
Caesar, Leonardo da Vinci, and Benjamin Franklin. Angelina Jolie and Oprah
Winfrey are left-handed, and so are Bruce Willis, Keanu Reeves, and Spike Lee.
You probably know more left-handed males than females. So you might ask this
question: Is there objective evidence to show that the proportion of left-handed
males in the country is larger than the proportion of left-handed females? In fact,
there are objective data that can help us answer this question.
The governments Health and Nutrition Survey (HANES) of 197680 recorded
the handedness of random samples of 1067 males and 1170 females from across
the country. The study found that 113 of the males and 92 of the females were
left-handed. From these data you can construct a significance test to answer the
question posed above. [Source: Freedman, Pisani, and Purves, Statistics, 3rd ed. (Norton, 1998), p. 537.]
First, a few words about the HANES survey are in order. The survey did
not randomly sample 1067 males and then randomly sample 1170 females to
ask about handedness. It randomly (by a somewhat complex design) selected a
large sample of adults from across the country and then sorted out the males and
females. The resulting two samples still can be considered to be independent and
random, as long as the sample sizes are large.
Example: Are Men More Likely to Be Left-Handed?

The survey just described found that 113 of 1067 males were left-handed while
92 of 1170 females were left-handed. Perform a signifi cance test to determine
whether males are more likely to be left-handed than females. (Conditions will
be checked in D43.)
8.4 A Significance Test for the Diff erence of Two Proportions
529
22-03-2009 21:16
Lesson
5 de 13
Define symbols and

state hypotheses.
Solution
Let p1 denote the proportion of left-handers among all males in the adult
population, and let p2 denote the proportion among all adult females. The
null hypothesis is almost always one of no difference or no effect, so the
appropriate null hypothesis here is
H0: There is no difference between the proportions of left-handed males
and left-handed females. In symbols, p1 = p2, or p1 p2 = 0.
The alternative hypothesis, sometimes called the research hypothesis, is a
statement of what the researcher is trying to establish. Here you are looking
for evidence that the proportion of left-handers is greater among males, so the
alternative hypothesis is
Ha: The proportion of left-handers among male adults is greater than the
proportion among female adults. In symbols, p1 > p2, or p1 p2 > 0.
Compute the test

statistic and P-value.
The test statistic builds on the best estimates of these population proportions,
namely, the sample proportions
and
The general form of a test statistic for testing hypotheses is
The difference from the sample (statistic) is , or 0.106 0.079 = 0.027.

The hypothesized difference (parameter) is 0, the value under the null hypothesis
p1 p2 = 0. The standard error of the estimate is given exactly by
is called the pooled

estimate of the common
proportion of successes.
You could do as you did in Section 8.3: Estimate p1 with and estimate p2
with . However, you can do even better. The null hypothesis states that the
proportion of males who are left-handed is equal to the proportion of females
who are left-handed, that is, = . You can estimate this common value of
and by combining the data from both samples into a pooled estimate, .
This pooled estimate is found by combining males and females into one group:
The standard error of the difference then is approximately
22-03-2009 21:16
Lesson
6 de 13
The test statistic then takes on the value
State the conclusion in

context.
This test statistic is based on the difference of two approximately normally

distributed random variables, and . You learned in Chapter 7 that such a
difference itself has a normal distribution. Thus, you can use Table A on page 824
to find the P-value. The test statistic z = 2.23 has a (one-sided) P-value of 0.0129.

normally distributed.
Display 8.24 P -value for a one-sided test with z = 2.23.
This P-value is very small, so reject the null hypothesis. The difference
between the rates of left-handedness in these two samples is too large to attribute
to chance variation alone. The evidence supports the alternative that males have a
higher rate of left-handedness.
As you have seen, the significance test for the difference of two proportions
proceeds along the same lines as the test for a single proportion. The steps in the
significance test for the difference of two proportions are given in this box.
Components of a Significance Test for the Difference of Two
Proportions for Surveys
1. Check conditions.
A random sample of size n1 is taken from a large binomial population with
proportion of successes p1. The proportion of successes in this sample is .
A second and independent random sample of size n2 is taken from a large
binomial population with proportion of successes p2. The proportion of
successes in this sample is . All these quantities must be at least 5:
Each population size must be at least 10 times the sample size.

(continued)
531
22-03-2009 21:16
Lesson
7 de 13
(continued)
Null hypothesis:
p1 = p2
Alternative
hypothesis: p1
p2,
p1 > p2, or p1< p2
Computations with
a sketch
2. Write a null and an alternative hypothesis.

You can write the null hypothesis in several different ways as long as you
define p1 and p2:
H0: The proportion of successes p1 in the first population is equal to the
proportion of successes p2 in the second population.
H0: p1 = p2, where p1 is the proportion of successes in the first population
and p2 is the proportion of successes in the second population.
H0: p1 p2 = 0, where p1 is the proportion of successes in the first
population and p2 is the proportion of successes in the second
population.
The form of the alternative hypothesis depends on whether you have a
two-sided or a one-sided test:
Ha: The proportion of successes p1 in the first population is not equal to
the proportion of successes p2 in the second population. In symbols,
p1 p2 or p1 p2 0.
Ha: The proportion of successes p1 in the first population is greater than
the proportion of successes p2 in the second population. In symbols,
p1 > p2 or p1 p2 > 0.
Ha: The proportion of successes p1 in the first population is less than the
proportion of successes p2 in the second population. In symbols,
p1 < p2 or p1 p2 < 0.
3. Compute the test statistic and a P-value.
where
The P-value is the probability of getting a value of z as extreme as or even
more extreme than that from your samples if H0 is true. See Display 8.25.
Display 8.25
Conclusion
P -values for two-sided and one-sided tests.
4. Write a conclusion.
State whether you reject or do not reject the null hypothesis. Link this
conclusion to your computations by comparing z to the critical value z*
or by appealing to the P-value. You reject the null hypothesis if z is more
extreme than z* or, equivalently, if the P-value is smaller than the level of
significance, . Then write a sentence giving your conclusion in context.
22-03-2009 21:16
Lesson
8 de 13
Example: Significance Test for a Difference

Should schools be allowed to restrict what students wear? A random sample of
548 teens, about equally split between boys and girls, was asked this question.
A report on the results of the poll stated: Gender seems to play a role in teens
opinions about dress codes. While a slight majority of girls (57%) support
restrictions on school wear, fewer boys (42%) agree. [Source: Gallup, Youth Survey: Should
Schools Police Fashion?, 2005, poll.gallup.com.] Is this evidence of a difference between the
responses of boys and girls in the population sampled in this poll? Perform a test
of significance at the 1% significance level.
Check conditions.
You should show these

computations worked
out, not just say they
are trueeven if that
is obvious.
Hypotheses
Computations with a
diagram
Solution
1. Although Gallup uses a sampling method more complex than simple random
sampling, you can consider these to be independent random samples. (As
mentioned previously, dividing one random sample into two groupsboys
and girlsgives an approximately random sample from each group when the
sample sizes are large.) The proportion of girls in the sample who support
restrictions is = 0.57, and the proportion of boys in the sample who
support restrictions is = 0.42. Using the estimates of n1 = 274 and
n2 = 274, you can see that all of
n1 = 274(0.57)
156
n1 ( 1
n2
115
n2(1
= 274(0.42)
) = 274(1 0.57)
) = 274(1 0.42)
118
159
are larger than 5. Both the number of teenage girls and the number of teenage
boys in the United States are much larger than 10 times each sample size.
2. H0: The proportion p1 of teenage girls supporting dress codes is equal to the
proportion p2 of teenage boys supporting dress codes.
Ha: p1 p2. (Note that you can use the symbols p1 and p2 because you defined
exactly what they stand for in the null hypothesis.)
3. The pooled estimate is
With a test statistic of 3.51, the P-value for a two-sided test is 0.0002(2), or
0.0004. See Display 8.26 on the next page.
533
22-03-2009 21:16
Lesson
9 de 13
Display 8.26 A two-sided test with z = 3.51.

Note that you can use your calculators two-proportion z-test, which carries
more decimal places, to get the more accurate value z = 3.503 and a P-value
of 0.00046. [See Calculator Note 8F.]
Conclusion in context
4. This difference is statistically significant, and you reject the null hypothesis.
If the proportions of girls and boys supporting dress codes were equal, then
you would expect only 4 out of 10,000 repeated samples of size 274 to have
a difference in sample proportions of 15% or larger. Because the P-value,
0.0004, is less than = 0.01, you cannot reasonably attribute the difference
to chance variation. You conclude that the proportions of girls and boys
supporting dress codes are different.
Display 8.27 shows a printout for this example.
Display 8.27 A printout for the significance test, with supporting

response coded 1.
22-03-2009 21:16
Lesson
10 de 13
The Theory of a Significance Test for the Difference

of Two Proportions
D43. In the left-handedness example on page 529, the null hypothesis was rejected.
a. Which type of error might have been made in the significance test? Is the
chance of making this error very large?
b. How can you check quickly that the condition is met that all of n1 ,
n1(1 ), n2 , and n2(1 ) are at least 5?
D44. What does it suggest about the two proportions if the significance test for a
difference of two proportions
a. fails to reject the null hypothesis?
b. rejects the null hypothesis?
D45. Suppose one population has 10% successes and a second population has 20%
successes. Generate an approximate sampling distribution of the diffrence
between sample proportions for samples of size 10 from each population.
Now repeat the procedure for samples of size 50 from each population.
Explain how the two sampling distributions illustrate the need for the
condition that n1 , n1(1 ), n2 , and n2(1 ) must each be at least 5.
Summary 8.4: A Significance Test for the Difference

of Two Proportions
To get a simulated sampling distribution of the difference
, you
take independent random samples from two different populations with
proportions of success p1 and p2
compute the proportion of successes in each sample, and
subtract the proportions of successes in the samples, that is, find
repeat this process many times
You learned how to test whether two samples were drawn from populations that have
the same proportion of successes. Follow these steps as with any significance test:
Check the conditions. You need two suffciently large random samples
selected independently from two different populations.
Write a null and an alternative hypothesis. The null hypothesis typically is
that the two populations contain the same proportion of successes.
Compute the test statistic and a P-value. The test statistic is
where is computed by combining both samples.

more extreme than that from your samples if H0 is true.
Write a conclusion. State whether you reject or do not reject the null
hypothesis, linking this decision to your value of z or to your P-value. Then
restate your conclusion in the context of the situation.
535
22-03-2009 21:16
Lesson
11 de 13
Practice
P48. According to the 2006 Statistical Abstract of
the United States, 12% of American adults
visited the zoo in the last year. Suppose you
take a random sample of 1000 American
adults and compute the proportion, , who
went to the zoo last year. Then you take
another random sample of 800 American
adults and compute the proportion, , in that
sample who went to the zoo last year. You
subtract the two proportions to get .
You repeat this process until you have a
distribution of millions of values of ,
( Thats why the problem says suppose.)
a. What is the expected value of the
difference ?
b. What is the standard error of the
distribution of ?
c. Sketch the distribution of , with a
scale on the horizontal axis.
d. Compute the probability that the
difference will be 0.05 or greater.
P49. Suppose you take two independent samples
from two different populations that have the
same proportion of successes. You repeat this
millions of times and plot the distribution
of the differences. Answer true or false to
each statement.
a. It is always the case that = .
b. It is true that = .
c. Sometimes the difference between the two
sample proportions can be relatively large.
d. Theoretically, the plot should be centered
exactly at 0.
e. To have a greater chance of having the
difference nearer 0, have larger
sample sizes.
The Theory of a Significance Test for the
Difference of Two Proportions
P50. A random sample of 600 probable voters
was taken 3 weeks before the start of a
campaign for mayor, and 321 of the 600
said they favored the new candidate over
the incumbent. However, it was revealed
the week before the election that the new
candidate had dozens of outstanding parking
tickets. Subsequently, a new random sample
of 750 probable voters showed that 382
voters favored the new candidate.
Do these data support the conclusion that
there was a decrease in voter support for
the new candidate after the parking tickets
were revealed? Give appropriate statistical
evidence to support your answer.
P51. What type of error might have been made
in the significance test in the example on
page 533 about teens supporting a dress code?
Exercises
E61. Milk Chocolate M&Ms are 24% blue, and
Almond M&Ms are 20% blue. Suppose you
take a random sample of 100 of each. You
compute the proportion that are blue in each
sample. You then subtract the proportion of
blue in the sample of Almond M&Ms from
the proportion of blue in the sample of Milk
Chocolate M&Ms to get
. You repeat
this process until you have a distribution of
millions of values of
.
a. What is the expected value of
?
distribution of
?
c. Sketch the distribution of

, with a
E62. Almond M&Ms and Peanut Butter M&Ms
both are 20% blue. Suppose you take a
random sample of 100 of each. You compute
the proportion that are blue in each sample.
You then subtract the proportion of blue in
the sample of Peanut Butter M&Ms from
the proportion of blue in the sample of
Almond M&Ms to get
. You repeat
22-03-2009 21:16
Lesson
12 de 13
d. Interpret the P-value you found in part c

in the context of this situation.
E65. A poll of 256 boys and 257 girls ages 1217
asked, Do you feel like you are personally
making a positive difference in your
community? More girls (76%) than boys
(63%) answered yes. [Source: www.ncpc.org,
this process until you have a distribution of

millions of values of
.
a. What is the expected value of
?
distribution of
?
c. Sketch the distribution of
, with a
E63. A study of newly diagnosed lung cancer
patients who had enrolled in an Early Lung
Cancer Action Program regimen of screening
found that 234 of the 258 patients with
tumors measuring 15 mm or less had no
metastases (spread to other parts of the body)
and that 98 of the 118 patients with tumors
measuring 1625 mm had no metastases.
The researchers needed to determine whether
the difference in these two proportions
is statistically significant. [Source: Computed
April 17, 2002.]
Tomographic Screening for Lung Cancer: The Relationship of

Disease Stage to Tumor Size, Archives of Internal Medicine 166
(February 13, 2006): 32125.]
a. The investigators chose to do a one-sided

test of significance. Why does that make
sense?
b. Are the conditions for inference met?
c. The P-value reported in the article was
0.02. Is that correct?
d. Interpret this P-value in the context of
this situation.
E64. A study at UCLA found that 55% of
100 cosmetic surgery patients take herbal
supplements while only 24% of a similar
group of 100 people not undergoing surgery
take herbal supplements. The researchers
needed to determine whether the difference
in these two proportions is statistically
significant. [Source: Los Angeles Times, February 20, 2006,
page F4.]
a. Would you have chosen to do a one-sided

or a two-sided test of significance?
Explain your reasons.
b. Are the conditions for inference met?
c. Using the test you chose in part a, find
the P-value for a significance test of the
difference in the two proportions.
a. Based on using a one-sided test, is this a

statistically significant difference? That is,
if all teens were asked, are you confident
that a larger proportion of girls than
boys would answer yes? Assume that the
samples were selected randomly.
b. The report says, Participants were
selected through random digit dialing.
Do you have any concerns about whether
such a procedure would give a random
sample?
proportion of all teens who would answer
yes. What additional assumption do you
need to make to do this?
E66. An annual poll of the sleeping habits of
Americans found that of a random sample
of 177 Americans ages 1829, 30% slept
8 hours or more on a weekday. Of a random
sample of 616 people ages 3049, 24% slept
8 hours or more on a weekday. Is this a
statistically significant difference? (When
answering this question, include all four
steps in your test of significance.) [Source:
National Sleep Foundation, 2005, www.sleepfoundation.org.]
E67. In 2005, as NASA prepared to launch the

New Horizons probe that will fly to Pluto
on a 9-year mission, a poll asked 1000 adult
Americans about their attitude toward the
U.S. space agency. Sixty percent gave NASA
a good to excellent rating. At the end of
1999, in a poll of about the same size, only
53% gave this high a rating. Was NASA
being looked upon more favorably by the
American public in 2005 than in 1999?
Would you say that there is strong evidence
to support your conclusion? (Always include
all four steps in your test of significance.)
[Source: Gallup, Public Favorable Toward NASA, Space
Exploration, 2006, www.poll.gallup.com.]
537
22-03-2009 21:16
Lesson
13 de 13
E68. The legal drinking age in the United

States is 21, and many people think that
underage drinking is a serious problem. This
question was asked in two recent separate
polls: Which one of the following do you
think would have the most influence on
persuading teenagers not to drink? One
poll randomly sampled 1008 adults, and the
other randomly sampled 514 teens. Among
the adults, 26% chose parents and 12%
chose friends. Among the teens, 18% chose
parents and 26% chose friends. (There
was virtual agreement on the other choices,
alcohol awareness programs and penalties.)
the same proportion of contented cows. Your

two-sample test has a P-value of 0.36. Select
the one best interpretation of this P-value.
[Source: USA Today teen drinking poll, 2001.]
a. Do adults choose parents significantly

more often than teens choose parents?
Would you say that there is strong
evidence to support your conclusion?
b. Can you use the methods of this section
to test for a significant difference
between the proportion of teens who
choose parents and the proportion of
teenagers who choose friends? Why or
why not?
E69. What is the trend in Internet use? National
polls of about 1000 adults each estimated
that 47% of adults logged on to the Internet
for an hour or more daily in 2003 and that
this increased to 51% by 2005. [Source: Gallup,
Internet Catches More of Americans Time, 2006, www.poll
.gallup.com.]
a. Is this a statistically significant increase?

b. Are you justified in saying that a majority
of adults in the United States logged on to
the Internet daily in 2005? Explain your
reasoning.
E70. In the same polls on Internet use as
referenced in E69, it was estimated that 33%
of adults log on to the Internet for more than
an hour daily but 27% never log on. Can you
test to see if this is a significant difference
using the methods of this section? Why or
why not?
E71. You feed Diet A to a random sample of cows
and Diet B to another random sample. Your
null hypothesis is that the two diets produce
A. If the null hypothesis that there is no

difference in the proportion of contented
cows produced by the two diets is true,
the probability is 0.36 that the proportion
of contented cows produced by each diet
wont be very different.
B. You cant conclude that the diets make a
difference because its fairly likely to get
as big a difference in the proportion of
contented cows produced by the two diets
as you did even if there is no difference at
all in the overall proportion of contented
cows that each diet produces.
C. The probability is 0.36 that the two diets
differ in the proportion of contented
cows that they produce. Because this
probability is so high, you can reject the
null hypothesis.
D. If the null hypothesis is true, then the
probability is 0.36 that the two diets differ
in the proportion of contented cows.
E. You cant reject the null hypothesis
because if, in fact, the null hypothesis
is true, the probability is 0.36 that the
difference in the true proportions is the
same as in your experiment.
E72. Explain how knowledge of sampling
distributions for differences in sample
proportions is used in the development of
the significance test of the difference of two
proportions.
22-03-2009 21:16
Lesson
1 de 15
For E73 and E74: In this section, you have been

testing the null hypothesis that p1 p2 equals 0.
The null hypothesis does not have to state that
the difference in the two population proportions
is 0. You can test for any difference you want.
In the z-statistic for the next two problems,
replace p1 p2 with the difference in population
proportions you are testing for rather than
replacing it with 0. Then, when estimating the SE,
, note that you should not
use a pooled estimate of a common population
proportion. The pooled estimate can be used only
under the hypothesis that the two population
proportions are equal, which is not what you are
hypothesizing here.
E73. Return to the example on page 529, where
a survey found that 113 of 1067 males were
left-handed while 92 of 1170 females were
8.5
An experiment uses the
available subjects.
left-handed. Perform a test to determine if

there is statistically significant evidence that
the percentage of males who are left-handed
is greater than the percentage of females who
are left-handed by more than 2 percentage
points.
E74. You can take either Bus A or Bus B to
school, and youre concerned about being
late to class. You check out Bus A on 100
randomly selected mornings and find that
you would be late to class 5 mornings. You
check out Bus B on 64 randomly selected
mornings and find that you would be late
5 mornings. Because you live a bit closer to
the stop for Bus B, you will use Bus A only if
you can conclude that the difference in the
proportions of late arrivals is more than 2%.
Which bus will you use?
Inference for Experiments

To use the methods for comparing two proportions discussed thus farconfidence
interval and test of significancethe underlying condition was that the data came
from random samples selected independently from two different populations. This is
the typical situation with sample surveys, such as surveys comparing pet ownership
rates for two different years. However, this is not the typical situation in experiments.
Experiments usually require investigators to use those subjects who volunteer
or those experimental units available for the study. If there are two treatments,
this group of subjects is then split into two groups by random assignment of the
treatments. In such situations, the population is the group of subjects being used
in the study.
Activity 8.5a will lead you through a simulation to help you see what happens
to the sampling distribution of the difference of two proportions under this new
method of randomization.
Random Assignment in an Experiment

What youll need: a group of students (the number in the group must be
divisible by 4; any remaining students will be in charge of passing out the cards);
a small card for each student, half marked sit and half marked stand (you
can use 3 5 cards or old playing cards); a second small card for each student,
75% marked with a C and 25% marked with a B; a set of B/C cards for every
student or small group of students, if you want to speed things up; paper or a
blackboard at the front of the room for a large dot plot
(continued)
539
22-03-2009 21:16
Lesson
2 de 15
(continued)
Suppose your entire class is afflicted with terminal boredom. In this experiment,
you will be randomly assigned one of two treatments, sit or stand, for your
terminal boredom. You will also be dealt a card that shows the result of your
treatment, B for still bored, or C for cured. The goal is to study the sampling
distribution of the difference between the proportions cured by the two
treatments.
1. Shuffle the sit/stand cards and give one to every student. Implement your
assigned treatment by either sitting or standing.
2. Shuffle the C/B cards and give one to every student. This card tells you
whether your treatment worked or not. If you get a card marked C, you are
cured! If you get a card marked B, you are still terminally bored.
3. For each treatment, find the proportion of students who were cured. Then
. Here
find the difference in the proportions who were cured,
represents the proportion of sitting students who were cured, and
represents the proportion of standing students who were cured. Mark the
difference on a dot plot.
4. Run this experiment again. Because its obvious that this particular
treatment has no effect whatsoever on terminal boredom, you might as
well keep the same C or B card that you were given before. To see the
effect of a different random assignment of treatments, shuffle the sit/stand
cards and hand them out again. For each treatment, find the proportion of
students who were cured. Then find the difference in the two proportions,
. Mark this second difference on the dot plot.
5. Repeat step 4 until you have enough differences to enable you to see the
shape, center, and spread of the sampling distribution.
Heres how to speed things up: Make duplicate sets of the B/C cards, with
75% marked C and 25% marked B. Shuffle and deal a set of the cards to
make two equal piles, the one on the left representing the sit treatment
and the one on the right representing the stand treatment. (You have just
randomly assigned treatments to your subjects.) Compute
and
record the difference as before on the dot plot.
6. Describe the shape, mean, and standard error of your approximate sampling
distribution. Is your distribution centered where you expect it to be?
7. Compare the SE to what you get by using the formula for the standard
error of the difference in sample proportions,
For this experiment, n1 and n2 are each half the number of students, and p1
and p2 are each 0.75, the probability of being cured. What can you conclude?
Display 8.28 shows the sampling distribution from Activity 8.5a for a class
of 40 students. The students participated in a sham experiment in which the
22-03-2009 21:16
Lesson
3 de 15
treatments were of no value whatsoever in curing their affliction. The histogram

shows the difference in the proportion of students cured by the two sham
treatments. This class used a computer to generate 10,000 rerandomizations for
the different treatments and thus has 10,000 differences.
Display 8.28 The difference in the proportion of students cured

by two sham treatments when the probability of a
cure is 0.75 no matter which treatment the student
received. Here n1 = n2 = 20.
Note that this approximate sampling distribution of the differences in

proportions from an experiment
looks approximately normal
is centered at 0, as expected
has standard error about 0.137
Now comes the great piece of luck. You dont have to do a simulation to get the
standard error. The formula you used for a sample survey works fine:
Of course, in the real-life situations you will be studying next, you will not
know the true success rates of the treatments and will have to estimate them
from the proportions of successes in the treatment groups. But that is nothing
newthis formula will continue to work fine just as you have used it before.
Confidence Interval for a Difference in Proportions

from an Experiment
For experiments, the form of the confidence interval and all the computations
remain the same, as do the conditions on sample size. But the interpretation is a
little different because you are estimating this difference:
proportion of successes if all units had received Treatment A
proportion of successes if all units had received Treatment B
541
22-03-2009 21:16
Lesson
4 de 15

from an Experiment
The confidence interval for the difference p1 p2 in the proportions of
successes in a population of experimental units subjected to two different
treatments is
Here is the observed proportion of successes in the group of size n1 given the
first treatment, and is the observed proportion of successes in the group of
size n2 given the second treatment. (The group sizes do not have to be equal.)
The conditions that must be met in order to use this formula are that
the two treatments are randomly assigned to the population of available
experimental units
n1
, n1(1
), n2
, and n2 (1
) are all at least 5
Example: Confidence Interval for an Experiment

From 1986 to 1988, an important study on the treatment of patients with
AIDS-related complex (ARC) was run by a collaborative group of researchers
from eight European countries and Australia. (AIDS-related complex is a
condition that generally leads to a diagnosis of AIDS.) A total of 134 patients
with ARC agreed to participate in the study, which was a double-blind,
randomized clinical trial. Two treatments were to be compared. Each patient
was given either zidovudine (commonly known as AZT) by itself or a
combination of AZT and acyclovir (ACV). One outcome measure was the
number of ARC patients who developed AIDS during the 1 year of the study
(Display 8.29).
Display 8.29
Dont we have to have a

random sample?
ARC patients progressing to AIDS during the study.

[Source: David A. Cooper et al., The Efficacy and Safety of
Zidovudine Alone or as Cotherapy with Acyclovir for the Treatment
of Patients with AIDS and AIDS-Related Complex: A Double Blind,
Randomized Trial, AIDS 7 (1993): 197207.]
The group of ARC patients werent randomly selected from the population of
all ARC patients but were volunteers who had to give informed consent. Thus,
they might be quite different from the population of all ARC patients and should
not be regarded as a random sample from any specifi c population.
22-03-2009 21:16
Lesson
5 de 15
However, the treatments were randomly assigned to the ARC patients who
were available. (This type of randomization is typical in medical research,
agricultural research, and many other kinds of experiments for which getting a
simple random sample from the entire population of interest is impossible.)
During the life of the study, 12 of 67 patients (or 17.9%) using AZT alone
progressed to AIDS, and 10 of 67 patients (or 14.9%) using AZT plus ACV
progressed to AIDS. Find a 95% Confidence interval for the difference between
p1, the proportion of patients in this study who would have developed AIDS if
they all had taken the AZT treatment, and p2, the proportion who would have
developed AIDS if they all had taken the AZT plus ACV treatment.
Check conditions.
Solution
The first condition that needs to be met is that the treatments were randomly
assigned to the subjects. The second condition is that n1 = 12, n1( 1 ) = 55,
n2 = 10, and n2 (1 ) = 57 are all 5 or more. Both conditions are met here.
The 95% Confidence interval for the difference between the two proportions
p1 and p2 is
Do computations.
Write your interpretation

in context.
You also can write this Confidence interval as (0.095, 0.155).

If all 134 patients could have been given each treatment, we estimate that
the difference between the rate of progression to AIDS if all had been given AZT
alone and the rate of progression if all had been given AZT plus ACV would be
someplace between a 9.5% difference in favor of AZT alone and a 15.5% difference
in favor of AZT plus ACV. Note that a difference of 0 lies well within the Confidence
interval. This means that if the difference in the proportions of patients who would
progress to AIDS is actually 0, getting a difference of 0.03 in our treatment groups
is reasonably likely. Thus, we arent convinced that the two therapies differ with
respect to the rate of progression to AIDS in this group of patients.
from an Experiment
D46. Discuss how the randomization used in a two-population sample survey
differs from the randomization used in a two-treatment experiment.
D47. The ARC data in the example came from a double-blind, randomized
clinical trial. What is the meaning of each part of this phrase?
D48. Suppose the difference in the proportion of patients who improve with
Treatment A and the proportion of patients who improve with Treatment B is
0.05. Does this mean that the majority of patients improve with Treatment A?
Does it mean that Treatment B isnt very good?
543
22-03-2009 21:16
Lesson
6 de 15
Significance Test for a Difference in Proportions

from an Experiment
You now know that the sampling distribution of the difference between two
proportions will still be approximately normal under the randomization scheme
used in experiments and that the formula for the estimated standard error
remains the same. Thus, for a significance test you can construct a test statistic
just as you did in the case of sample survey data. The main difference between
the two testing scenarios is in the stating of the hypotheses and conclusions. An
outline of a significance test for comparing proportions from an experiment is
given in the box.
Components of a Significance Test for the Difference
of Two Proportions from an Experiment
1. Check conditions.
Two treatments, A and B, are randomly assigned to available experimental
units, with n1 receiving Treatment A and n2 receiving Treatment B. With
representing the proportion of successes from Treatment A and
representing the proportion of successes from Treatment B, all of the
following values must be at least 5:
2. Write a null and an alternative hypothesis.

The null hypothesis is that the two treatments would have resulted in
the same proportion of successes if each treatment could have been
given to all subjects. The alternative hypothesis for a two-sided test is
that the two treatments would not have resulted in the same proportion
of successes if each treatment could have been given to all subjects. The
alternative hypothesis for a one-sided test is that a specific treatment would
have resulted in a higher proportion of successes than the other if each
treatment could have been given to all subjects.
3. Compute the test statistic and a P-value.

more extreme than that from your samples if H0 is true. See Display 8.30.
(continued)
22-03-2009 21:16
Lesson
7 de 15
Display 8.30
P -values for a two-sided test (left) and one-sided

tests (center and right).
4. Write a conclusion.
State whether you reject or do not reject the null hypothesis. Link this
conclusion to your computations by comparing z to the critical value, z*, or
by appealing to the P-value. You reject the null hypothesis if z is more extreme
than z* or, equivalently, if the P-value is smaller than the level of significance,
. Then write a sentence giving your conclusion in the context of the situation.
To see how to apply these ideas to an experiment, return to the clinical trial
experiment comparing the two treatments for AIDS-related complex (ARC).
Example: Significance Test for an Experiment

The investigators next question dealt with the survival rate among 131 patients
who had already developed AIDS before entering the study. These data are shown
in Display 8.31.
Display 8.31
Treatment results among AIDS patients.
Here
, where is the proportion who survived in the group
treated with AZT alone, and
, where is the proportion who
survived in the group treated with AZT plus ACV. In order for a new therapy (in
this case, AZT plus ACV) to become accepted practice among physicians, the
research must show that the new therapy is an improvement over the old one.
Can that claim be made based on these data?
State hypotheses.
Solution
Because the researchers are attempting to show that the new therapy is better than
the old one, a one-sided significance test is in order. The hypotheses are
H0: The new therapy is not better than the old, or p1 = p2, where p1 is the
proportion of patients who would have survived if all 131 patients could have
been given AZT alone and p2 is the proportion of patients who would have
survived if they all could have been given AZT plus ACV.
Ha: The new therapy is better than the old one, or p1 < p2.
545
22-03-2009 21:16
Lesson
8 de 15
Check conditions.
Treatments were randomly assigned to the available subjects, and each of

n1
Compute the test

statistic and the P -value.
= 41
n1( 1
) = 28
n2
= 49
n2(1
) = 13
is at least 5.
The pooled estimate, , of the probability that a patient will survive, under the
null hypothesis that the treatment a patient received made no difference in whether
he or she survived, is
or about 0.687. The test statistic then takes on the value
This test statistic is based on the difference of two approximately normally

distributed random variables, and . You learned in Chapter 7 that such a
difference is itself approximately normal. Thus, you can use Table A on page
824 to find the P-value. The test statistic z = 2.42 has a (one-sided) P-value of
0.0078, as illustrated in Display 8.32.
Display 8.32 P-value for a one-sided test with z = 2.42.

This P-value is very small, so reject the null hypothesis. If all subjects in the
experiment could have been given the AZT plus ACV treatment, you are confi dent
that there would have been a larger survival rate than if all had received only AZT.
The difference between the survival rates for these two treatments is too large to
be attributed to chance variation alone.
Significance Test for a Difference in Proportions

from an Experiment
D49. Discuss the differences in stating hypotheses for testing the equality of
proportions in a two-population sample survey versus a two-treatment
experiment.
D50. In the AZT versus AZT plus ACV example, the null hypothesis was rejected.
Which type of error might have been made in the significance test? Discuss
the practical ramifications of making this error in this context.
22-03-2009 21:16
Lesson
9 de 15
Inference for an Observational Study

Sometimes it is impossible, either ethically or practically, to assign treatments
to subjects or experimental units. This often is the situation in epidemiological
studies of the spread or progression of a disease, because it is unethical to give a
subject a harmful disease as part of an experiment. However, possible associations
between living habits and the progression of a disease may be observed by
studying many people with the disease to see if, for example, people who exercise
regularly have slower progression of the disease than people who dont exercise
regularly. As you saw in Chapter 4, studies in which the conditions of interest
are already built into the units being studied are called observational studies.
Even though no randomization is involved, inference procedures can still be
applied if you are very careful in stating the question and your conclusion. What
you are asking is, Could the result I see in the observed data have reasonably
happened by chance? In other words, if the data had been generated through a
randomized experiment, would such a result be likely? If the answer is no, then
there is evidence of an association that should be investigated further. Remember,
observational studies do not allow you to make cause-and-effect conclusions or to
estimate population parameters in a rigorous way, but they can provide evidence
of possible associations.
Example: Confidence Interval in an Observational Study

Do you know someone who is trying to quit smoking? Does the person have
a plan to stop? How successful has he or she been? Some have argued that a
spur-of-the-moment decision to stop smoking (cold turkey, if you will) is just
as successful as the best-laid plan to stop smoking. In a recent study on this
question, researchers observed 611 documented attempts to quit smoking.
Among the 297 unplanned, spur-of-the-moment attempts, 194 succeeded
6 months or longer. Among the 314 planned attempts, 133 succeeded 6 months
or longer. [Source: www.medpagetoday.com.] Use a 95% confidence interval to explore a
possible association between unplanned attempts and success rate.
Solution
The conditions for a standard confidence interval for the difference between
proportions are not met because there was no randomization in this study.
The researchers studied the 611 attempts that were available to them from data
collected in a national survey, but the treatments, planned and unplanned
attempts to quit smoking, were not randomly assigned. Nevertheless, the 95%
confidence interval can be informative. This develops from standard procedures:
The confidence interval is (0.15, 0.31).
547
22-03-2009 21:16
Lesson
10 de 15
This confidence interval provides some evidence that there is an association

between the type of attempt and the success of the attempt. Those who attempt
to quit smoking spur-of-the-moment might have a success rate that is larger than
those who use planned attempts by, perhaps, something between 15% and 31%.
However, this study cannot determine whether the method itself resulted in a
higher success rate or some other variable was responsible, such as a difference
in the willpower of people who choose the two methods. Nor can this study
determine if the difference applies to the larger population of smokers who
attempt to quit. All that can be concluded is that this difference cannot easily be
attributed to chance alone. Thus, this result might suggest to the researchers that
they should collect more data on this question, although they probably can never
do a randomized study. (They could not randomly assign a person to make an
immediate decision to stop smoking.)
Example: Significance Test in an Observational Study

Does a helmet law have much impact on whether bicycle riders actually wear
helmets? Anticipating the passing of a helmet law, the government of North
Carolina commissioned a study of helmet usage among bicycle riders at various
locations across the state in 1999. This study provided baseline data for studying
the effects of the law. In 2002, 6 months after the law was passed, a similar
study was conducted at the same types of locations. Some of the resulting data
summaries are shown in Display 8.33.
Display 8.33 Helmet usage at various locations. [Source: University of

North Carolina Highway Safety Research Center, www.hsrc.unc.edu.]
Is there sufficient evidence to say that the difference in the percentage using
helmets on local streets before the law and after the law is so large that it cannot
reasonably be attributed to chance alone?
Solution
Even though some of the sites for observing riders were randomly selected,
the riders observed were those who just happened to be there at the time of
observation. In addition, there were no randomly assigned treatments and
no controls on any of the myriad other variables that could affect helmet use.
Nevertheless, a test of signifi cance can shed some light on the possible association
between the law and helmet usage. The null hypothesis is that there is no
difference in the proportions of helmet users on local streets for 1999 and 2002.
The alternative hypothesis of interest is that the 2002 rate of helmet use is higher
than the 1999 rate (perhaps because of the law).
22-03-2009 21:16
Lesson
11 de 15
Under the null hypothesis of no difference in proportions, the pooled

estimate of helmet use on local streets is
If you let = 0.16 denote the proportion of helmet users observed on local
streets in 1999 and = 0.19 denote the proportion observed in 2002, the test
statistic becomes
If this were a formal one-sided test of significance, the P-value would be 0.04.
This indicates that, if there was no change in the percentage of helmet users,
there would be very little chance of seeing a difference this large in the percentage
wearing helmets in samples of randomly selected bicycle riders. Thus, there is some
evidence that there was a higher rate of helmet-wearing in 2002, but that could
be due to a combination of many factors, one of which might be the law. The law
could have coincided with an increased emphasis on safety education, a general
trend in the behavior of bicycle riders, the age or experience of the riders, or even a
change in the weather. None of these possible factors are measured, controlled, or
balanced by randomization.

D51. Discuss the differences in the types of conclusions that can be reached in a
randomized experiment versus those that can be reached in an observational
study.
D52. There is a large controversy in educational research about whether randomized
experiments can be used to reach conclusions about curriculum materials
and methods. For example, suppose a new method for teaching fractions is to
be compared to a standard method that has been in use for years. In order to
be accepted as better, the new method must be tested against the old in real
classroom situations. Should these tests be set up as randomized experiments
or as observational studies? What do you suppose is a key argument in favor
of randomized experiments? What do you suppose is a key argument against
trying to conduct randomized experiments?
Summary 8.5: Inference for Experiments

Both confidence interval estimation and tests of significance proceed in
inference for experiments much as they did in inference for sample surveys,
but the hypotheses and conclusions must be stated differently. This is because
an experiment generally does not result in a comparison based on independent
random samples from two populations. Rather, one population of available
experimental units is split randomly into two groups by the assignment of
treatments. As long as the basic sample-size considerations are met, however, the
same methods used in the analysis of sample surveys apply.
549
22-03-2009 21:16
Lesson
12 de 15
Practice
Confidence Interval for a Difference in
Proportions from an Experiment
P52. The landmark Physicians Health Study
study of the effect of low-dose aspirin on
the incidence of heart attacks (see E79 on
page 552) checked to see if aspirin use was
associated with increased risk of ulcers. This
was a randomized, double-blind, placebocontrolled clinical trial. Male physician
volunteers with no previous important
health problems were randomly assigned
to an experimental group (n1 = 11,037) or
a control group (n2 = 11,034). Those in the
experimental group were asked to take a pill
containing 325 mg of aspirin every second
day, and those in the control group were
asked to take a placebo pill every second day.
Of those who took the aspirin, 169 got an
ulcer, compared to 138 in the placebo group.
Find and interpret a 95% confidence interval
for the difference of two proportions. [Source:
Steering Committee of the Physicians Health Study Research
Group, The Final Report on the Aspirin Component of the
Ongoing Physicians Health Study, New England Journal of
Medicine 321, no. 3 (1989): 12935; www.content.nejm.org.]
P53. In a classic 1962 social experiment in Ypsilanti,

Michigan, 3- and 4-year-old children were
assigned at random to either a treatment
group receiving two years of preschool
instruction or a control group receiving no
preschool instruction. The participants in the
study were followed through their teenage
years. One variable recorded was the number
in each group arrested for a crime by the time
they were 19 years old. The data are shown in
Display 8.34.
Display 8.34
Arrests by the age of 19, by preschool

attendance. [Source: F. Ramsey and D. Schafer,
The Statistical Sleuth (Duxbury, 2002), p. 546;
Time, July 29, 1991.]
b. Estimate the true difference between

the proportions arrested, using a 90%
c. Carefully state a conclusion in the context
of the problem.
d. This is an experiment, but is a clear
cause-and-effect relationship established
here? Explain your reasoning.
Significance Test for a Difference in
Proportions from an Experiment
P54. The battle against high cholesterol has been
raging for years. A randomized clinical
trial compared two drugs designed to lower
cholesterol (Lipitor and Zocor) with regard
to both the cholesterol outcome and the
rate of heart attacks over a 5-year period.
It is the prevention of heart attacks that is
the ultimate goal, and Zocor is much less
expensive than Lipitor (because Zocors
patent expired in 2006). High doses of
Lipitor won out over low doses of Zocor on
lowering cholesterol, but the heart attack
issue was much closer. Of 4420 patients
on Lipitor, 411 had heart attacks. Of 4452
patients on Zocor, 463 had heart attacks.
[Source: USA Today, November 16, 2005.]
a. Is this evidence of a significant difference

in heart attack rates for the two drugs?
b. Prior to the study, many experts thought
that the high doses of Lipitor should be
much more effective against heart attacks
than the low doses of Zocor. Conduct the
appropriate one-sided test to check this
claim. What is your conclusion? How do
you reconcile it with your conclusion in
part a?
P55. In P54, which type of error could have
been made in part a? In part b? Which do
you think is the more serious error in the
practical context of this problem? Explain
your reasoning.
a. Are the conditions met for constructing

a confidence interval for the difference of
two proportions?
22-03-2009 21:16
Lesson
13 de 15

P56. Refer to the example on the attempts to
quit smoking on page 547. Set up a
statistical analysis to test the claim that the
spur-of-the-moment method is actually
better than the planned method for quitting.
State your conclusion carefully in light of the
fact that this is an observational study.
P57. Do victims of violence exhibit more
violence toward others? In a classic study
of this question, a researcher found records
of 908 people who had been abused as
children. Then, matching the demographic
characteristics of this group as closely as she
could, she found records of 667 individuals
who had not been abused as children.
Based on further searches of records, she
discovered that 103 of the abused children
later perpetrated violent crimes, while 53 of
the children not abused perpetrated violent
Exercises
E75. A famous medical experiment was
conducted by Nobel laureate Linus Pauling
(19011994), who believed that vitamin C
prevents colds. His subjects were 279
French skiers who were randomly assigned
to receive vitamin C or a placebo. Of the
139 given vitamin C, 17 got a cold. Of the
140 given the placebo, 31 got a cold. [Source:
L. Pauling, The Signifi cance of the Evidence About Ascorbic
Acid and the Common Cold, Proceedings of the National
Academy of Sciences 68 (1971): 267881.]
a. Find and interpret a 95% confidence

interval for the difference of two
proportions.
difference of two proportions. Does
your conclusion change from your
interpretation in part a?
E76. A randomized clinical trial on Linus
Paulings claim that vitamin C helps prevent
the common cold was carried out in Canada
among 818 volunteers, with results reported
in 1972. The data showed that 335 of the
411 in the placebo group got colds over the
winter in which the study was conducted,
crimes. [Source: C. S. Wisdom, The Cycle of Violence,

Science 244 (1989): 160.]
a. What kind of study was this: sample

survey, experiment, or observational
study?
b. Estimate the difference in violent crime
rates for the two groups, using a 90%
c. Write a conclusion in the context of the
problem, being careful to consider the
type of design that was used. Does this
prove that abuse of children leads to
violent crime down the road?
P58. Refer to the example on helmet use in North
Carolina on page 548. Among the other
three locations (collector streets, greenways,
and mountain biking trails), which shows
the strongest association between helmet use
and the law?
while 302 of the 407 in the vitamin C group

got colds. [Source: T. W. Anderson, D. B. Reid, and
G. H. Beaton, Vitamin C and the Common Cold, Canadian
Medical Association Journal 107 (1972): 5038.]
a. Find and interpret a 95% confidence

interval for the difference of two
proportions.
difference of two proportions. Does
your conclusion change from your
interpretation in part a?
E77. In 1954, the largest medical experiment of
all time was carried out to test whether the
newly developed Salk vaccine was effective
in preventing polio. This study incorporated
all three characteristics of an experiment: use
of a control group of children who received
a placebo injection (an injection that felt
like a regular immunization but contained
only salt water), random assignment of
children to either the placebo injection
group or the Salk vaccine injection group,
and assignment of each treatment to several
hundred thousand children. Of the 200,745
children who received the Salk vaccine,
551
22-03-2009 21:16
Lesson
14 de 15
82 were diagnosed with polio. Of the 201,229

children who received the placebo, 162 were
diagnosed with polio. Find and interpret a
95% confidence interval for the difference
of two proportions. [Source: Paul Meier, The Biggest
E80. A related part of the landmark study

discussed in E79 dealt with the question
of whether aspirin reduces the seriousness
of a heart attack should one occur. Of the
139 volunteers in the aspirin group who
Public Health Experiment Ever, in Statistics: A Guide to the
had a heart attack, 10 died. Of the 239 in
Unknown, 3rd ed., edited by Judith M. Tanur et al. (Pacific
the placebo group who had a heart attack,
Grove, CA: Brooks/Cole, 1989).]
26 died. Is the difference in death rate
E78. Two different methods of treating wool
statistically significant?
with a mothproofing agent were tested in a
E81.
Exercise Helps Delay Onset of Dementia
laboratory by randomly dividing 40 wool
read
a headline in the Gainesville Sun on
samples into two groups of 20, one group
January
17, 2006. Researchers followed a
randomly assigned to Method A and the
group of people age 65 and older from 1994
other to Method B. Moth larvae were then
to 2003. Among the 1185 free of dementia at
attached to each sample of wool and observed
the end of this time, 77% reported exercising
for a fixed period of time. For Method A, 8 of
three or more times a week. Among the 158
the 20 larvae died; for Method B, 12 of the 20
who showed signs of dementia at the end of
larvae died. Is there evidence of a difference
the period, 67% reported exercising three or
in the effectiveness of these two methods?
more times a week.
Answer the question by constructing and
a. Is this an experiment, a sample survey, or
interpreting an appropriate 95% confidence
an observational study?
interval. [Source: G. E. P. Box, W. G. Hunter, and J. S.
Hunter, Statistics for Experimenters (Wiley, 1978), pp. 13233.]
b. Test to see if the difference in these
proportions is statistically significant.
E79. The Physicians Health Study conducted a
famous experiment on the effect of lowThen state your conclusions carefully, in
dose aspirin on heart attacks. This was a
light of your answer to part a. Can this
randomized, double-blind, placebo-controlled
study demonstrate that exercise causes a
clinical trial. Male physician volunteers with
delay in the onset of dementia?
no previous important health problems
E82. In one early study (1956) of the relationship
were randomly assigned to an experimental
between smoking and mortality, Canadian
group (n1 = 11,037) or a control group
war pensioners were asked about their
(n2 = 11,034). Those in the experimental
smoking habits and then followed to see
group were asked to take a pill containing
if they were alive 6 years later. Of the
325 mg of aspirin every second day, and
1067 nonsmokers, 950 were still alive. Of
those in the control group were asked to take
the 402 pipe smokers, 348 were still alive.
a placebo pill every second day. After about
[Source: G. W. Snedecor and W. G. Cochran, Statistical
5 years, there were 139 heart attacks in the
Methods, 6th ed. (Iowa State Press, 1967), pp. 21516.]
aspirin group and 239 in the placebo group.
Is there significant evidence that aspirin use
reduces the rate of heart attacks in this group?
b. Is there statistically significant evidence
Test at the 1% level. [Source: Steering Committee of the
that pipe smoking increases the rate
Physicians Health Study Research Group, The Final Report
of mortality? State your conclusions
on the Aspirin Component of the Ongoing Physicians Health
carefully, in light of your answer to part a.
Study, New England Journal of Medicine 321, no. 3 (1989):
12935; www.content.nejm.org.]
22-03-2009 21:16
Lesson
15 de 15
E83. The great baseball player Reggie Jackson

variation, or should you look for another
batted .262 during regular season play and
explanation? Use = 0.05. [Source: Christopher
Ward, The War of the Revolution (New York: Macmillan,
.357 during World Series play. He was at
1952), p. 96.]
bat 9864 times in the regular season and 98
I. the proportion of American troops who
times in the World Series. Can this difference
were wounded (including those killed)
reasonably be attributed to chance, or did
and the proportion of the British troops
Reggie earn his nickname Mr. October?
who were wounded (including those
(The World Series is played in October.) Use
killed)
= 0.05. Note that you do not have two
random samples from different populations.
II. the proportion of the American troops
You have Reggies entire record, so all you
who were killed and the proportion of the
can decide is whether the difference is about
British troops who were killed
the size you would expect from chance
E85. A test of significance asks the question
variation or you should look for some other
Could the result I see have happened
explanation. [Source: Major League Baseball, February
by chance, or should I look for another
23, 2002, www.mlb.com.]
cause? In randomized experiments, the
answer to this question can be found simply
by studying what happens in repeated
randomizations, without recourse to
standard errors, test statistics, or the normal
distribution. Here is the construction of such
a test based on the AIDS data in Display 8.31
and the underlying null hypothesis that there
is no difference in the death rates for the two
treatments, AZT and AZT plus ACV.
1. Make 90 chips numbered 1 to represent
those alive and 41 chips numbered 0 to
represent those not alive at the end of the
study time.
2. Randomly split the 131 chips into two
groups, one with 69 chips (representing
those treated with AZT alone) and one
with 62 chips (representing those treated
E84. At Bunker Hill, near Boston, in one of the
with AZT plus ACV).
first battles of the American Revolution,
3. Calculate the difference between the two
2400 British troops were engaged, of which
proportions of 1s (subtracting AZT plus
1054 were wounded. Of the 1054 wounded,
ACV from AZT).
226 died. Out of 1500 American participants
4.
Repeat the process many times and plot
in the battle, losses were estimated at 140
the
values of the generated differences in
killed and an additional 271 wounded who
proportions. Under the null hypothesis,
didnt die of their wounds. (The number
each of the generated differences has the
of casualties varies somewhat depending
same chance of occurring.
on the source.) You have all the necessary
A plot of 200 runs of this simulation
information about the two populations. Can
procedure is shown in Display 8.35 on the
the difference in the pairs of proportions in
next page.
I and II reasonably be attributed to chance
553
22-03-2009 21:16
Lesson
1 de 6
Display 8.35 A plot of 200 runs of a randomization

process.
a. The observed difference in the

proportions for the data from the real
experiment was about 0.20. How many
times, out of 200, did the randomization
procedure generate a difference this small
or smaller? What does this suggest about

the chance of seeing a value this small
under the null hypothesis?
b. Explain what your conclusion would
be regarding the two treatments based
solely on the randomization procedure.
Does your conclusion agree with the one
reached in the example on pages 545546?
E86. Refer to the data in P53 on page 550. Use
the randomization procedure from E85 to
test the hypothesis that preschool makes no
difference in whether a person is arrested for
a violent crime.
E87. Refer to the data in the example on page 542
(Display 8.29). Use the randomization
procedure from E85 to test the hypothesis
that AZT plus ACV is better than AZT
alone for reducing the incidence of AIDS.
Does your conclusion agree with the one
reached in the example by constructing and
interpreting a confidence interval?
Chapter Summary
To use the confidence intervals and significance tests of this chapter, you need either
a random sample from a population that is made up of successes and failures or
two random samples taken independently from two distinct such populations. If the
study is an experiment, you should have two treatments randomly assigned to the
available subjects. (When there is no randomness involved, you proceed with the
test only if you state clearly the limitations of what you have done. If you reject the
null hypothesis in such a case, all you can conclude is that something happened that
cant reasonably be attributed to chance.)
You use a confidence interval if you want to find a range of plausible values for
p, the proportion of successes in a population
p1 p2 , the difference between the proportion of successes in one population
and the proportion of successes in another population
Both confidence intervals you have studied have the same form:
statistic
(critical value) (standard deviation of statistic)
Both significance tests you have studied include the same steps:
1. Justify your reasons for choosing this particular test. Discuss whether the
conditions are met, and decide whether it is okay to proceed if they are not
strictly met.
2. State the null hypothesis and the alternative hypothesis.
22-03-2009 21:17
Lesson
2 de 6
3. Compute the test statistic and find the P-value. Draw a diagram of the
situation.
4. Use the computations to decide whether to reject or not reject the null
hypothesis by comparing the test statistic to z* or by comparing the P-value
to , the significance level. Then state your conclusion in terms of the context
of the situation. (Simply saying Reject H0 is not sufficient.) Mention any
doubts you have about the validity of your conclusion.
Review Exercises
E88. A 6th-grade student, Emily Rosa, performed
an experiment to test the validity of
therapeutic touch. According to an article
written by her RN mother, a statistician,
and a physician: Therapeutic Touch (TT)
is a widely used nursing practice rooted in
mysticism but alleged to have a scientific
basis. Practitioners of TT claim to treat
many medical conditions by using their
hands to manipulate a human energy field
perceptible above the patients skin. To
investigate whether TT practitioners
actually can perceive a human energy field,
21 experienced TT practitioners were tested
to determine whether they could correctly
identify which of their hands was closest to
the investigators hand. They placed their
hands through holes in a screen so they
could not see the investigator. Placement of
the investigators hand was determined by
flipping a coin. Practitioners of TT identified
the correct hand in only 123 of 280 trials.
[Source: Journal of the American Medical Association 279
(1998): 100510.]
a. What does blinded mean in the context

of this experiment? How might it have
been done? Why wasnt double-blinding
necessary?
b. Write appropriate null and alternative
hypotheses for a significance test.
c. Is it necessary to actually carry out the
test, based on these data?
d. The article says, The statistical power of
this experiment was sufficient to conclude
that if TT practitioners could reliably
detect a human energy field, the study
would have demonstrated this. What
does this sentence mean?
E89. Most teens are not careful enough about the

information they give out about themselves
online. In a survey that randomly sampled
971 teenagers who have online access, 78%
agreed with this statement. [Source: Pew Internet
and American Life Project, 2005, www.pewinternet.org.]
a. Compute the 95% confidence interval for

the population proportion of teenagers
who would agree with the statement.
b. Interpret this confidence interval, making
it clear exactly what it is that you are 95%
sure is in the confidence interval.
c. Explain the meaning of 95% confidence.
E90. A 2001 Gallup poll found that 51% of the
American public assigned a grade of A or
B to the public schools in their community.
In 2000, the comparable figure was 47%.
Assuming a sample size of 1108 in both
2000 and 2001, find and interpret a 90%
confidence interval for the difference of two
proportions. [Source: Gallup, www.gallup.com.]
E91. A study of all injuries from the two winter
seasons 19992000 and 20002001 at the
three largest ski areas in Scotland found
that of the 531 snowboarders who were
injured, 148 had fractures. Of the 952 skiers
who were injured, 146 had fractures. (For
both groups, most of the other injuries were
sprains, lacerations, or bruising.) [Source: www
.ski-injury.com, March 29, 2002.]
a. Is this difference statistically significant at

the 0.05 level?
b. There are about twice as many skiers as
snowboarders. Can you use this fact and
the data from this study to determine
whether snowboarders are more likely
than skiers to be injured?
Chapter Summary
555
22-03-2009 21:17
Lesson
3 de 6
B. The sample proportion will fall in 95% of

the confidence intervals we construct.
C. In the long run, the population
proportion will fall in 95% of the
confidence intervals we construct.
D. We are 95% confident that the range of
reasonably likely outcomes contains the
population proportion.
E. We are 95% confident that we have made
a correct decision when we reject the null
hypothesis.
E97. How could this explanation of margin of
error be improved?
Statisticians over the years have developed
M. Morgan, and M. Abrams, Changes in Brain Function of
quite specific ways of measuring the accuracy
Depressed Subjects During Treatment with Placebo, American
of samplesso long as the fundamental
Journal of Psychiatry 159, no. 1 (January 2002): 12229.]
principle
of equal probability of selection is
E93. Suppose a null hypothesis is tested at the
adhered
to
when the sample is drawn.
0.05 level of significance in 100 different
For example, with a sample size of 1,000
studies in which the null hypothesis is true.
national adults (derived using careful random
a. How many Type I errors do you expect?
selection procedures), the results are highly
b. Find the probability that no Type I error
likely to be accurate within a margin of error
is made in any of the 100 studies.
of plus or minus three percentage points.
E94. To become a lawyer, a person must pass the
So, if we find in a given poll that President
bar exam for his or her state. Suppose you are
Clintons approval rating is 50%, the margin
searching for possible inequities that result
of error indicates that the true rating is very
in unequal proportions of males and females
likely to be between 53% and 47%. It is
passing the bar exam. Each state has its own
very unlikely to be higher or lower than that.
exam, so you check a random sample of the
To be more specific, the laws of probability
exam records in each of the 50 states. For
say that if we were to conduct the same
each state, you plan to perform a significance
survey 100 times, asking people in each
test of the difference of the proportion of
survey to rate the job Bill Clinton is doing
males and females passing the exam, using
as president, in 95 out of those 100 polls, we
=0.05. If the difference is statistically
would find his rating to be between 47% and
significant for any state, you will conclude
53%. In only five of those surveys would we
that there has been some inequity. Do you
expect his rating to be higher or lower than
see anything wrong with this plan?
that due to chance error? [Source: Frank Newport,
E95. Suppose you use the formula for a 90%
Lydia Saad, and David Moore, How Polls Are Conducted,
in Where America Stands (New York: John Wiley & Sons, Inc.,
confidence interval for many different
1997). Used by permission of John Wiley & Sons, Inc.]
random samples from the same population.
E98.
Some people complain that election polls
What fraction of times should the intervals
cannot be right, because they personally
produced actually capture the population
were not asked how they voted. Write an
proportion? Explain your reasoning.
explanation about how polls can get a good
E96. What is the best explanation of the use of the
idea of how the entire population will vote
term 95% confidence?
by asking a relatively small number of voters.
A. We can never be 100% confident in
statistics; we can only be 95% confident.
E92. To study the brains response to placebos,
researchers at UCLA gave 51 patients with
depression either an antidepressant or
a placebo. The article reports that the
51 subjects then were randomly assigned
to receive 8 weeks of double-blind
treatment with either placebo or the active
medication. . . . Overall, 52% of the subjects
(13 of 25) receiving antidepressant
medication responded to treatment, and
38% of those receiving placebo (10 of 26)
responded. Perform a test to determine
if the difference in the proportion who
responded is statistically significant. Use
= 0.02. [Source: A. F. Leuchter, I. A. Cook, E. A. Witte,
22-03-2009 21:17
Lesson
4 de 6
AP1. Only 33% of students correctly answered

a difficult multiple-choice question on an
exam given nationwide. Ms. Chang gave
the same question to her 35 students,
hypothesizing that they would do better
than students nationwide. Despite the lack of
randomization, she performed a one-sided
test of the significance of a sample
proportion and got P = 0.03. Which is
the best interpretation of this P-value?
Only 3% of her students scored better
than students nationally.
If the null hypothesis is true that her
students do the same as students
nationally, there is a 3% chance that her
students will do better than students
nationally on this question.
Between 30% and 36% of her population
of students can be expected to answer
the question correctly.
There is a 3% chance that her students
are better than students nationally.
There is a 3% chance that a random sample
of 35 students nationwide would do as well
as or better than her students did.
AP2. Researchers constructed a 95% confidence
interval for the proportion of people who
prefer apples to oranges. They computed a
margin of error of 4%. In checking their
work, they discovered that the sample size
used in their computation was of the
actual number of people surveyed. Which
is closest to the correct margin of error?
1%
2%
4%
8%
16%
AP3. A survey of 200 randomly selected alumni
of Lincoln High School found that 105 favor
abolishing the school dress code. Is this
convincing evidence that more than half of
all alumni favor abolishing the dress code?
Yes, because 105 is more than half of 200.
Yes, by constructing a 95% confidence
interval, you can see that it is plausible
that more than half of all alumni favor
abolishing the dress code.
No, by constructing a 95% confidence

interval, you can see that it is plausible
that 50% or less of all alumni favor
abolishing the dress code.
No, because the survey only used a
sample of alumni.
Because alumni probably have strong
opinions that arent normally distributed,
no conclusion can be reached.
AP4. In a study of the effectiveness of two tutors in
preparing students for an exam, 50 students
were randomly assigned either to Mr. A or to
Mr. B. A larger proportion of students tutored
by Mr. A passed the exam, resulting in a
95% confidence interval for the difference of
two proportions of (0.05, 0.45). Which is
not a correct conclusion to draw from this?
There is statistically significant evidence
that Mr. A is the better tutor.
You cannot reject a null hypothesis of
equal tutor effectiveness.
The difference in the percentage who
passed the exam and were tutored by
Mr. A and the percentage who passed
and were tutored by Mr. B is 20%.
A Type II error may be made.
The design of this study didnt have
enough power to pick up any but a very
large difference in passing rates.
AP5. With = 0.05, researchers conducted a
test of the difference of two proportions
to compare the rate of alcohol use among
teens this year and in 1990. The rates for
both years are based on large, independent
random samples of teens. Which is the best
interpretation of = 0.05 in this context?
Theres a 5% chance that the rate of
alcohol use has changed.
Theres a 5% chance that the rate of
alcohol use has not changed.
If the rate of alcohol use has not changed,
theres a 5% chance that the researchers
will mistakenly conclude that it has.
AP Sample Test
557
22-03-2009 21:17
Lesson
5 de 6
If the rate of alcohol use has changed,

theres a 5% chance that the researchers
will mistakenly believe it hasnt.
The study has enough power to detect
a difference in the rate of alcohol use of
0.05 or more.
AP6. A student obtains a random sample of
M&Ms and a random sample of Skittles.
She finds that 7 of the 40 M&Ms are yellow
and 13 of the 35 Skittles are yellow. Her
null hypothesis is that the proportion of
yellow candies is equal in both brands. Her
alternative hypothesis is that the proportion
is higher in Skittles. What is her conclusion,
if she uses = 0.05?
Reject the null hypothesis. The difference
in the two proportions can reasonably be
attributed to chance alone.
Reject the null hypothesis. The
difference in the two proportions cannot
reasonably be attributed to chance alone.
Do not reject the null hypothesis. The
difference in the two proportions can
Do not reject the null hypothesis. The
difference in the two proportions cannot
Accept the null hypothesis because the
difference in the two proportions is not
AP7. In a pre-election poll, 51% of a random
sample of voters plan to vote for the
incumbent. A 95% confidence interval was
computed for the proportion of all voters
who plan to vote for the incumbent. What is
the best meaning of 95% confidence?
If all voters were asked, there is a 95%
chance that 51% of them will say they
plan to vote for the incumbent.
You are 95% confident that the
confidence interval contains 51%.
In 100 similar polls, you expect that 95
of the confidence intervals constructed
will contain the percentage of all voters

who plan to vote for the incumbent.
In 100 similar polls, you expect that 95
of the confidence intervals constructed
will contain 51%, which is the best
estimate of the percentage of all voters
The probability is 0.95 that the
confidence interval constructed in this
poll contains the percentage of all voters
AP8. Sheldon takes a random sample of 50 U.S.
housing units and finds that 30 are owneroccupied. Using a significance test for a
proportion, he is not able to reject the null
hypothesis that exactly half of U.S. housing
units are owner-occupied. Later, Sheldon
learns that the U.S. Census for the same
year found that 66.2% of housing units are
owner-occupied. Select the best description
of the type of error in this situation.
No error was made.
A Type I error was made because a false
null hypothesis was rejected.
A Type I error was made because a false
null hypothesis wasnt rejected.
A Type II error was made because a false
null hypothesis was rejected.
A Type II error was made because a false
null hypothesis wasnt rejected.
Investigative Tasks
AP9. Fifty students want to know what percentage
of Skittles candies are orange. Each student
takes a random sample of 50 candies and
constructs a 90% confidence interval using
the formula
a. Of all Skittles candies, 20% are orange.

How many of their fifty 90% confidence
intervals would you expect to capture
the population proportion of 0.20?
22-03-2009 21:17
Lesson
6 de 6
b. The students confidence intervals are

plotted in Display 8.36 as vertical lines.
How many of their confidence intervals
capture the population proportion of
0.20? What percentage is this?
c. Are most of the confidence intervals that
dont capture the population proportion
too close to 0 or too close to 0.5? Do they
tend to be shorter or longer than intervals
that do capture the population proportion?
d. Unfortunately, what you have discovered
here is true in general: If you use
the standard formula to construct
confidence intervals for a proportion,
your actual capture rate will be smaller
(and sometimes a lot smaller) than
advertised. List several ways in which
this formula is based on approximations
and so might have an actual capture rate
different from that advertised.
add 2 to the number of observed failures

(which adds 4 to the sample size). That is,
if x denotes the number of successes in a
sample of n items, the new estimator of the
population proportion, p, is
The Plus 4 confidence interval is then
a. Construct 90% confidence intervals for

n = 50 and x = 5 using both the standard
formula and the Plus 4 method. What
happened to the center of the interval?
The length of the interval?
b. Repeat the procedures and questions of
part a for n = 50 and x = 20. Compare
your answers to those in part a.
c. The students in AP9 used the same
samples as before to construct Plus 4
confidence intervals. These Plus 4
confidence intervals are given in Display
8.37. How many of the intervals
capture the population proportion
of 0.20? What percentage is this?
Display 8.36 Fifty 90% confidence intervals for

random samples of size 50 with p = 0.20
constructed using the standard formula.
AP10. The Plus 4 interval. In AP9, you saw that the

standard formula for a confidence interval
results in a capture rate that tends to be less
than advertised. The capture rate should
improve if the more extreme intervals
are moved toward the center and made
a little longer, as long as the remaining
intervals arent disturbed too much. There
is a method that will do exactly that, called
the Plus 4 method. You simply add 2 to
the number of observed successes and
Display 8.37 Fifty 90% confidence intervals for

the same random samples of size 50
with p = 0.20 constructed using the
Plus 4 method.
d. Describe how the Plus 4 confidence

intervals differ from the standard
intervals.
AP Sample Test
559
22-03-2009 21:17
Lesson
1 de 2
Inference for Means
Does it make a
difference whether you
take water samples
at mid-depth or near
the bottom when
studying pesticide
levels in a river?
Inference for differences
is fundamental to
statistical applications
because many surveys
and all experiments are
comparative in nature.
22-03-2009 21:17
Lesson
2 de 2
You now know how to make inferences for proportions based on categorical
data. But often the data that arise in everyday contexts are measurement data
rather than categorical data. Measurement data, as you know, are most often
summarized by using the mean. Mean income, mean scores on exams, mean
waiting times at checkout lines, and mean heights of people your age are all
commonly used in making decisions that affect your life.
In Chapter 8, you learned how to construct confidence intervals and
perform significance tests for proportions. Although the formulas youll use for
the inferential procedures in this chapter will change a little from those you just
learned, the basic concepts remain the same. These concepts are all built around
the question What are the reasonably likely outcomes from a random sample?
This chapter will follow the same outline as Chapter 8 except that the methods
are applied to means rather than proportions.
In this chapter, you will learn how to

construct and interpret a confidence interval for estimating an unknown
mean
perform a significance test for a single mean
construct and interpret a confidence interval for estimating the difference
between two means
perform a significance test for the difference between two means
use a confidence interval to estimate the mean of the differences from
paired samples
perform a significance test for the mean of the differences from paired
samples
22-03-2009 21:17
Lesson
1 de 18
9.1
A Confidence Interval for a Mean

In this section you will learn how to construct a confidence interval for a
population mean, . The structure of the confidence interval will be similar to
that in Section 8.1 for a proportion:
= statistic
Lets start with an example of a real-life situation in which it is important to locate

the mean of a population.
Example: Average Body Temperature

When you are sick and take your temperature, you probably compare your
temperature to the so-called normal temperature, 98.6F. How was that
temperature determined? In the 1860s, the German physician Carl Wunderlich
measured the temperatures of thousands of healthy people and reported the
mean as 37C (98.6F). Investigators have questioned his methodology and the
quality of his thermometers. To determine an up-to-date average, researchers
took the body temperatures of 148 people at several different times during two
consecutive days. A portion of these data, for ten randomly selected women, is
given here (in F):
97.8 98.0 98.2 98.2 98.2 98.6 98.8 98.8 99.2 99.4
[Source: Allen L. Shoemaker, Whats Normal?Temperature, Gender, and Heart Rate, Journal
of Statistics Education 4 (1996). Original source: P. A. Mackowiak et al., A Critical Appraisal of 98.6
Degrees F, the Upper Limit of the Normal Body Temperature, and Other Legacies of Carl Reinhold
August Wunderlich, Journal of the American Medical Association268 (September 1992): 157880.]
The mean body temperature, , for this sample of ten women is 98.52, and
the standard deviation, s, is 0.527. Are these statistics likely to be equal to the
mean and standard deviation for the population? How can you determine the
plausible values of the mean temperature of all women?
Solution
Neither nor s is likely to be exactly equal to the population parameters and .
(This would be true no matter how large the sample.)
Plausible values of the mean body temperature of all women, , are those
values that lie close to = 98.52, where close is defined in terms of standard
error. From Section 7.3, the standard error of the sampling distribution of a
sample mean is given by
=
, where is the standard deviation of the
population and n is the sample size. When the sample size is large enough or the
population is normally distributed, in 95% of all samples and are no farther
. So plausible values of lie in the interval
apart than 1.96
or
562 Chapter 9 Inference for Means
22-03-2009 21:18
Lesson
2 de 18
For the sample of body temperatures, you know and n, but you dont know
you seldom will know in real-life situations. What now? As you might be
thinking, you will have to use the sample standard deviation, s, as an estimate of .
The Effect of Estimating

To compute a confidence interval for the mean, you use the sample standard
deviation, s, as an estimate of . How will substituting s for affect your
confidence interval? Some samples give an estimate thats too small: s < . In this
case, the confidence interval will be too narrow. Others give an estimate thats too
large: s > . In this case, the confidence interval will be too wide.
On average, the sampling distribution of s has its center very near , so, on
average, the width of the confidence interval is about right. Thats a nice feature,
but confidence intervals are judged on their capture rate, not their average width.
Does replacing with s change your chance of capturing the unknown population
mean? You will investigate this matter in Activity 9.1a.
The Effect of Estimating the Standard Deviation

on Measuring Romantic Love
Nerve growth factor (NGF) is a neurotrophin found in blood plasma.
Researchers found that the mean NGF concentration of people truly, deeply
and madly in love for six months or less was significantly higher than that in
single people or in people in a long-term relationship. The NGF levels in those
who recently fell in love were normally distributed with mean 227 (pg/mL)
and standard deviation 107. Assume that these are the population parameters.
[Source: Enzo Emanuele et al., Raised Plasma Nerve Growth Factor Levels Associated with
Early-Stage Romantic Love, Psychoneuroendocrinology 31, no. 3 (2006): 28894.]
Suppose you must estimate the population mean from the NGF levels of a
random sample of four people who recently fell in love.
1. Use your calculator to generate a random sample of four NGF levels from
this normally distributed population, and store them into list 1. [See
Calculator Note 7A .]
2. Use your calculator to find the mean, , and standard deviation, s. Record
the value of s. [See Calculator Note 2F.]
3. Construct a confidence interval using the formula
Does your confidence interval capture the population mean, 227?

(continued)
563
22-03-2009 21:18
Lesson
3 de 18
(continued)
4. Follow the procedure in steps 1 through 3 until the students in your class
have generated a total of 100 samples. You will have 100 values of s and
100 decisions as to whether the confidence interval captured the
population mean, .
5. Make a dot plot of the 100 values of s and describe its shape, mean, and
spread. How close to is the mean of the values of s? How many times is s
smaller than ? Larger?
6. What percentage of your confidence intervals should capture the true
population mean, , of 227? What percentage of your confidence intervals
actually capture ? What can you conclude about the effect of using s to
estimate ?
s is smaller than s more

often than it is larger.
You learned in the activity that the sampling distribution of s is skewed right.
Although the average value of s is about equal to , s tends to be smaller than
more often than it is larger because the median is smaller than the mean for this
skewed distribution. This causes the confidence intervals to be too narrow more
than half the time, so the capture rate will be less than the advertised value of 95%.
Fortunately, the sampling distribution of s becomes less skewed and more
approximately normal as the sample size increases. The histogram on the left in
Display 9.1 shows the values of s for 1000 samples of size 4 taken from a normally
distributed population with = 107. For this small sample size, the distribution
is quite skewed. The histogram on the right shows the values of s for 1000 samples
of size 20. There is little skewness here, so s is smaller than about as often as it
is larger.
Display 9.1
Approximate sampling distribution of s for samples

of size 4 (left) and size 20 (right) for a normally
distributed population with = 107.

D1. Consider two extreme situations, n = 10 and n = 1000. If you use s for in
the formula,
1.96
, which sample size10 or 1000do you expect
to give a capture rate closest to 95%? Why?
22-03-2009 21:18
Lesson
4 de 18
D2. When you use s to estimate , the capture rate is too small unless you make
further adjustments. If an intervals capture rate is smaller than what you
want it to be, do you need to use a wider or a narrower interval to get the
capture rate you want?
How to Adjust for Estimating

Replacing s with s tends
to make the interval too
narrow. Compensate by
replacing z* with t*.
As you have seen, using s as an estimate of in a confidence interval lowers the

overall capture rate because s tends to be smaller than more often than it is
larger. To compensate, you will increase the width of the confidence interval by
replacing z* with a larger value, called t*:
Lets listen in on a discussion about t*.

Student: Where does the value of t* come from?
Statistician: In principle, you could find it using simulation. Set up an
approximately normal population, take a random sample, and
compute the mean and standard deviation. Do this thousands of
times. Then use the results to figure out the value of t* that gives a
95% capture rate for intervals of the form
.
Student: Wouldnt that take a lot of work?
Statistician: Yes, especially if you went about it by hand. Fortunately, this
work has already been done, long ago. Statistician W. S. Gosset
(18761937), who worked for the Guinness Brewery, actually
did this back in 1915. Four years later, geneticist and statistician
R. A. Fisher (18901962) figured out how to find values of t* using
calculus. If the population is nearly normal, it turns out that the
value of t* depends on just two things: how many observations
you have, and the capture rate you want.
Student: So t* doesnt depend on the unknown mean or unknown standard
deviation?
Statistician: No, it doesnt, which is very handy because in practice you dont
know these numbers. Suppose, for example, you have a sample of
size 5 and you want a 95% confidence interval. Then you can use
t* = 2.776 no matter what the values of and are.
Student: Where did you get that value for t*?
Statistician: From a t-table, although I could have found it from a calculator
or statistical software. A brief version of the table is shown in
Display 9.2. Table B on page 826 is more complete. For 95%
confidence, you go to the column with 95% at the bottom.
For the row, you need to know the degrees of freedom, df
for short.
565
22-03-2009 21:18
Lesson
5 de 18
Display 9.2
An abbreviated version of a t-table, with a sketch

showing the shaded tail areas.
Student: Degrees of freedom? Whats that?

Statistician: Theres a short answer, a longer answer, and a very long answer.
The longer answer will come in E40. The very long answer is for
another course. For the moment, heres the short answer: The
number of degrees of freedom is one less than the sample size.
(It is the number you use for the denominator when you calculate
the standard deviation, s.) So for these confidence intervals,
df = n 1, where n is the sample size. When n = 5, df = 4,
and you look in that row of the table. If you look in the row with
df = 4 and the column with confidence level 95%, youll find the
value 2.776 for t*.
Finding the Value of t* to Use in Your Confidence Interval
Turn to Table B on page 826. Find your confidence level in the last row of the
table. Find your df, n 1 where n is the sample size, in the column at the left.
Your value of t* is at the intersection of this row and column. If your df falls
between two of those listed in the column on the left, use the smaller df.
[To find t * using a calculator, see Calculator Note 9A.]
Example: Who Is Hotter?

Is normal body temperature the same for men and women? As you saw in the
previous example, medical researchers interested in this question collected data
from a large number of men and women. Random samples from that data are
recorded in Display 9.3. The womens temperatures are the same as those given in
the previous example.
a. Use a 95% Confidence interval to estimate the mean body temperature
of men.
b. Use a 95% Confidence interval to estimate the mean body temperature
of women.
22-03-2009 21:18
Lesson
6 de 18
Display 9.3
Two random samples of body temperatures. [Source:

Journal of Statistics Education Data Archive, www.amstat.org,
April 15, 2002.]
Solution
First, you need to check that these are random samples from relatively large
populations and that its reasonable to assume that the populations are normal.
(More on this is in Section 9.3.) The problem states that these are random
samples selected from a group of men and women examined by the researchers,
so your fi nal result generalizes only to this population. The researchers reported
that the distributions are normal. You should always plot the data, however, and
the stemplots in Display 9.4 give you no reason to suspect that the populations
are not normally distributed.
Display 9.4
Stemplots of body temperature data.
For a sample of size 10 (9 degrees of freedom) and a confidence level of 95%,

Table B on page 826 gives a value of t* of 2.262.
a. For the males, the sample mean is 97.88, and the sample standard deviation is
0.555. That gives a 95% confidence interval of
or (97.48, 98.28). Any population of body temperatures with a mean in this

interval could have produced a sample mean of 97.88 as a reasonably likely
outcome.
567
22-03-2009 21:18
Lesson
7 de 18
b. For the females, the sample mean is 98.52, and the sample standard deviation
is 0.527. You are 95% confident that the mean body temperature of women in
this population is in the interval
or
or (98.14, 98.90).
From the sample means it might appear that the average body temperature of
females is higher than that of males, but beware! There is some overlap in the
confidence intervals, so that conclusion might not be valid. Youll come back to
the question of comparing two means in Section 9.4.

D3. What parameter does the standard deviation, s, of a random sample tend to
approach as the sample size gets larger? Look at Table B on page 826. What
happens to the value of t* as the sample size increases? Find the row where
df equals . Do those numbers look familiar?
D4. Overall, the researchers found a mean body temperature of about 98.2F,
which is lower than Wunderlichs 98.6F that has become the standard.
Was Wunderlich wrong? Wunderlich worked in the metric system. He
rounded his mean body temperature to the nearest degree to get 37C. What
is the lowest his mean could have been in degrees Celsius? The highest?
Convert the lowest possible mean, the reported mean (37C), and the
highest possible mean to degrees Fahrenheit. The conversion formula is
F = C + 32. What can you conclude?
Constructing a Confidence Interval for a Mean

Whenever you construct a confidence interval based on t, there are three
conditions you must check. Officially, you need
a random sample (or random assignment of treatments to units)
an approximately normally distributed population or a large enough sample
size that the distribution of the sample mean is approximately normal
in the case of a survey, the size of the population should be at least ten times
the size of the sample
The fact that you must always have a random sample should be no surprise. A
normally distributed population is (theoretically) required because the values of t*
in Table B were computed by assuming that the population is normally distributed.
However, you can get away with less than a normally distributed population if the
sample size is large enough (this will be explained in Section 9.3). If the sample size
is more than about 10% of the population size, the estimated standard error of the
mean used in this chapter will tend to be larger than it needs to be for the specified
confidence level.
22-03-2009 21:18
Lesson
8 de 18
Thats the last of the new things to learn about constructing a confidence
interval for a mean. Heres a summary of what youve learned.
A Confidence Interval for a Mean
Check conditions.
You must check three conditions:
Randomness. In the case of a survey, the sample must have been randomly
selected. In the case of an experiment, the treatments must have been
randomly assigned to the experimental units.
Normality. The sample must look like its reasonable to assume that it
came from a normally distributed population or the sample must be large
enough that the sampling distribution of the sample mean is approximately
normal. There is no exact rule for determining whether a normally
distributed population is a reasonable assumption or what constitutes a
large enough sample, but you will learn some guidelines in Section 9.3.
Sample Size. In the case of a sample survey, the population size should be at
least ten times as large as the sample size.
Do computations.
A confidence interval for the population mean, , is given by
where n is the sample size, is the sample mean, s is the sample standard
deviation, and t* depends on the confidence level desired and the degrees of
freedom, df = n 1.
Give interpretation in context.
A good interpretation is of this form: I am 95% confident that the population
mean, , is in this confidence interval.
Of course, when you interpret a confidence interval, you will do it in
context, describing the population you are talking about.
Example: The Atomic Weight of Silver

The National Institute of Standards and
Technology (NIST) conducts research
that helps U.S. industry measure, test, and
standardize products and services. In order
to reduce uncertainty in the atomic weight
of silver, an NIST researcher measured the
atomic weight of a reference sample of
silver. The 24 measurements from one mass
spectrometer are given in Display 9.5. Note
that all 24 measurements begin with 107.868;
thus, it will be easier to work with the coded
values, the last four decimal places.
569
22-03-2009 21:18
Lesson
9 de 18
Construct and interpret a 95% confidence interval for the mean measurement
(using this mass spectrometer) of the atomic weight of silver.
Display 9.5
Check conditions.
Twenty-four measurements of a sample of silver.

[Source: L. J. Powell, T. J. Murphy, and J. W. Gramlich, The Absolute
Isotopic Abundance & Atomic Weight of a Reference Sample of
Silver, NBS Journal of Research 87 (1982): 919, www.itl.nist.gov.]
Solution
You can consider these measurements to be a random sample of all possible
measurements by this mass spectrometer. Display 9.6 shows that the distribution
of measurements from the sample is reasonably symmetric with no outliers, so it
is reasonable to assume that they could have come from a normal distribution.
Display 9.6
Do computations.
Give interpretation in
context.
A dot plot of the sample data.
The mean and standard deviation of the measurements are 1363.54 and
169.017, respectively. For 95% confidence and df equal to 24 1, or 23, use
t* = 2.069. The 95% confidence interval is
or about (1292, 1435).

You are 95% confident that the mean measurement of the atomic weight of silver
using this mass spectrometer is in the interval from 107.8681292 to 107.8681435.
22-03-2009 21:18
Lesson
10 de 18
You also can use your calculator to calculate the confidence interval for a
mean, given either the original data or the summary statistics of the mean and
standard deviation [See Calculator Note 9B]. Both approaches are shown here for
the data in the previous example.
Constructing a Confidence Interval for a Mean

D5. A student said, Lets be sure and capture by producing an interval with
a 100% confidence level. Is this possible? Would it be wise to shoot for a
99.5% confidence level?
D6. You have produced a confidence interval for your supervisor, and she says,
This interval is too wide to be of any practical value. What are your options
for producing a narrower interval in the next study of this same type? Which
option would you choose if the study might have consequences that are vital
to the future of the firm?
More on Interpreting a Confidence Interval

The capture rate of a
method for producing
confidence intervals
is the chance that the
interval will contain
the population mean.
The capture rate

is the same as the
confidence level.
As in Chapter 8, the capture rate of a confidence interval is the proportion

of random samples for which the resulting confidence interval captures the
(population) parameter. You can think of it as the chance that the method used
to produce a confidence interval actually will work correctly.
The Capture Rate of a Confidence Interval for a Mean
that capture is equal to
The proportion of intervals of the form
the confidence level. That is, a 95% confidence interval will have a 95% capture
rate, a 90% confidence interval will have a 90% capture rate, and so on.
This will be true provided that
the sample is random
the population is normally distributed
the size of the population is at least 10 times the sample size
The capture rate will be approximately correct for non-normally distributed
populations as long as the sample size is large enough.
571
22-03-2009 21:18
Lesson
11 de 18
Remember that the capture rate is a property of the method, not of the
sample. After the sample is selected, either the resulting interval will capture
the population mean or it wont. So it is incorrect to say about a specific interval
that there is a 95% probability (or 95% chance) that the true mean falls in this
interval. All you can say is that you are 95% confident that is in your interval,
meaning that the method produces a confidence interval that captures , the true
mean, 95% of the time.
The Meaning of 95% Confident
Saying you are 95% confident means that if you could take random samples
repeatedly from this population and compute a confidence interval from each
sample, in the long run 95% of these different intervals would contain (or
capture) .
In other words, a 95% confidence interval for a population mean, , includes
the plausible values of . A population with any one of these values for could
have produced the observed sample mean as a reasonably likely outcome.
In practice, samples might not be random, distributions might not be normal,
and survey questions might be worded so that people answer incorrectly. In such
situations, the capture rate may be far below what is advertised.

D7. Explain why it is incorrect to say you are 95% confident that the population
mean is in the confidence interval that you just constructed.
D8. The capture rate equals the confidence level. Explain why this statement
depends on
a. the randomness of the sampling
b. the normality of the population
Margin of Error
The quantity
The margin of error is

half the width of the
is called the margin of error. It is half the width of the confidence interval. When
your samples are random, larger samples provide more information than do
smaller ones. So, in general, as the sample size gets larger, the margin of error
gets smaller.
Example: Margin of Error for the Mean Atomic Weight of Silver

Find the margin of error for the mean atomic weight of silver in the example on
pages 569570.
22-03-2009 21:18
Lesson
12 de 18
Solution
The confidence interval for the mean atomic weight was
So the margin of error is 71.38.
Summary 9.1: A Confidence Interval for a Mean

In this section you have learned how to construct a confidence interval for a
population mean, . A confidence interval for is given by
where is the sample mean, s is the sample standard deviation, and n is the
sample size. Because you are using s to estimate , you must use values of t*
rather than values of z*. Find the value of t* in Table B on page 826 or on your
calculator, using df = n 1. For a given confidence level and sample size, the
value of t* is slightly larger than the corresponding value of z*, but this difference
becomes smaller with larger sample sizes. You must check that you have a
random sample from a relatively large population (or a random assignment
of treatments to units) and that the sample came from a distribution that is
approximately normal or, if not, that the sample is large enough.
Here is how to interpret this confidence interval:
You are 95% confident that the population mean , is in this confidence
interval.
This method of constructing a 95% confidence interval ensures that the
chance of getting a value of whose interval captures is 95%. This is called
the capture rate or confidence level.
Practice
P1. Aldrin in the Wolf River. Aldrin is a highly
toxic organic compound that can cause
various cancers and birth defects. Ten
samples taken from Tennessees Wolf River
downstream from a toxic waste site once used
by the pesticide industry gave these aldrin
concentrations, in nanograms per liter:
5.17 6.17 6.26 4.26 3.17
3.76 4.76 4.90 6.57 5.17
a. Calculate a confidence interval of the
, using s as an
form
estimate of .
b. In general, how do you expect this

estimateusing s for to affect the
center of a confidence interval? Will it
tend to make the interval too wide or too
narrow? Will it make the capture rate
larger than 95% or smaller than 95%?
P2. Using a t-Table. Use Table B on page 826 to
find the correct value of t* to use for each of
these situations.
a. a 95% confidence interval based on a
sample of size 10
573
22-03-2009 21:18
Lesson
13 de 18
b. a 95% confidence interval based on a

sample of size 7
c. a 99% confidence interval based on a
sample of size 12
d. a 90% confidence interval with n = 3
e. a 99% confidence interval with n = 44
f. a 99% confidence interval with n = 82
P3. For each situation, construct a 95%
confidence interval for the unknown mean.
a. n = 4
= 27 s = 12
b. n = 9
=6
s=3
=9
s = 48
c. n = 16
P4. Using the data in P1, construct a 95%
confidence interval for the mean level of
aldrin. Compare the width of this confidence
interval to your interval in P1.
Constructing a Confidence Interval
for a Mean
P5. An article in the Journal of the American
Medical Association included the body
temperatures of 122 men. (The data in the
example on pages 566567 were a random
sample taken from this larger group of men.)
The summary statistics for the entire sample
of mens temperatures were = 98.1,
s 0.73, and n = 122. [Source: P. A. Mackowiak et
al., A Critical Appraisal of 98.6 Degrees F, the Upper Limit
of the Normal Body Temperature, and Other Legacies of
Carl Reinhold August Wunderlich, Journal of the American
Medical Association 268 (September 1992): 157880.]
a. Construct a 95% confidence interval for

the mean of the population from which
these data were selected. You can assume
that the conditions have been met.
b. Write an interpretation of this confidence
interval.
c. Is 98.6F in this interval? What can you
conclude?
P6. The data in Display 9.7 are from a survey
taken in an introductory statistics class
during the first week of the semester. These
data are the number of hours of study per
week reported by each of the 61 students.
Because of the way students register for class,
these students can be considered a random

sample of all students taking this course.
Display 9.7 Number of hours of study per week.
a. You were told that the sample can be

considered a random selection from this
large population of students. Do the data
look as if they reasonably could have
come from a normal distribution?
estimate the mean study hours per week
for all the students taking this course,
with 90% confidence.
c. Explain the meaning of your confidence
interval.
P7. Refer to the 95% confidence interval you
constructed for the aldrin data in P4. Which
of statements AE are correct interpretations
of that interval?
A. If you take one more measurement from
the Wolf River, you are 95% confident
that this measurement will fall in the
B. If you take ten more measurements from
the Wolf River, you are 95% confident that
the sample mean of the ten measurements
will fall in the confidence interval.
C. There is a 95% chance that the mean
aldrin level of the Wolf River falls in the
22-03-2009 21:18
Lesson
14 de 18
D. You are 95% confident that the mean

aldrin level of the Wolf River falls in the
E. You are 95% confident that the sample
mean falls in the confidence interval.
P8. A 95% confidence interval for a population
mean is calculated for a random sample
of weights, and the resulting confidence
interval is from 42 to 48 lb. For each
statement, indicate whether it is a true
or false interpretation of the confidence
interval. Explain why the false statements
are false.
a. 95% of the weights in the population are
between 42 lb and 48 lb.
b. 95% of the weights in the sample are
between 42 lb and 48 lb.
c. The probability that the interval includes

the population mean, , is 95%.
d. The sample mean, , might not be in the
e. If 200 confidence intervals were generated
using the same process, about 10 of these
confidence intervals would not include
Margin of Error
P9. Refer to the confidence intervals for the
mean body temperature of men and women
in the example on pages 566567.
a. Find the margin of error for the men and
for the women.
b. Which margin of error is larger? Why is it
larger?
Exercises
E1. If you are lost in the woods and do not have
clear directional markers, will you tend to
walk in a circle? To study this question, some
students recruited 30 volunteers to attempt
to walk the length of a football field while
blindfolded. Each volunteer began at the
middle of one goal line and was asked to
walk to the opposite goal line, a distance of
100 yards. None of them made it. Display
9.8 shows the number of yards they walked
before they crossed a sideline.
a. Are the conditions met for finding a
confidence interval estimate of the
mean distance walked before crossing
a sideline? Explain any reservations
you have.
construct a 99% confidence interval for
the mean distance walked before crossing
a sideline.
c. Interpret your interval in part b.
Display 9.8
Number of yards walked by blindfolded

volunteers before going out of bounds.
[Source: Andrea Axtell, When Direction Vanishes:
Walking Straight or in Circles, STATS 39 (2004):
1821.]
575
22-03-2009 21:18
Lesson
15 de 18
E2. Walking Babies. Some babies start walking

well before they are a year old, while others
still havent taken their first unassisted steps
at 18 months or even later. The data in
Display 9.9 are from an experiment designed
to see whether a program of special exercises
for 12 minutes each day could speed up
the process of learning to walk. In all, 23
week-old male infants (and their parents)
took part in the study and were randomly
assigned either to the special exercise group
or to one of three control groups.
In order to isolate the effects of interest, the
scientists used three different control groups.
In the exercise-control group, parents were
told to make sure their infant sons exercised
at least 12 minutes per day but were given no
special exercises and no other instructions
about exercise. In the weekly report group,
parents were given no instructions about
exercise, but each week they were called to
find out about their babys progress. Parents
in the final report group also were given
no instructions about exercise, nor did they
receive weekly check-in calls. Instead, they
gave a report at the end of the study. The
response is the age, in months, when the
baby first walked without help.
Display 9.9
Age (in months) when 23 male infants

first walked without support. [Source:
Phillip R. Zelazo, Nancy Ann Zelazo, and Sarah
Kolb, Walking in the Newborn, Science 176 (1972):
31415 (via Larsen and Marx).]
a. You can assume that there is no problem

with the normality condition for
constructing a cofidence interval. Has
the other condition been met?
b. Find a 95% cofidence interval estimate

of the mean time to first unaided steps for
each treatment, assuming all 23 infants
could have received that treatment. The 23
infants participating in the study constitute
the population for this experiment.
c. Is it clear from your cofidence intervals
which treatment is best if you want baby
boys to walk earlier on average?
E3. To avoid any further hill climbing, Jack and
Jill have opened a water-bottling factory.
The distribution of the number of ounces
of water in the bottles is approximately
normal. The mean, , is supposed to be
16 oz, but the water-filling machine slips
away from that amount occasionally and
has to be readjusted. Jack and Jill take a
random sample of ten bottles from todays
production and weigh the water in each. The
weights (in ounces) are
15.91 16.08 16.08 15.94 16.02
15.94 15.96 16.03 15.82 15.96
Should Jack and Jill readjust the machine?
Use a statistical argument to support your
advice.
E4. Display 9.10 gives the hot dog prices for
three randomly selected National Basketball
Association (NBA) teams.
Display 9.10
A sample of hot dog prices. [Source:

www.teammarketing.com, March 2006.]
a. From this very small sample, is there any

reason to suspect that the conditions are
not met for constructing a cofidence
interval for the mean hot dog price for all
NBA teams?
b. Regardless of your answer in part a,
construct a 95% cofidence interval.
c. The mean hot dog price for all NBA
teams is $3.20. Is that value in your
22-03-2009 21:18
Lesson
16 de 18
confidence interval? If not, list two

reasons why this could have happened.
The plots of all 30 prices in Display 9.11
might help you.
Display 9.11
Dot plot and boxplot of NBA hot dog

prices.
d. There are only 30 teams in the NBA. If ten

teams had been sampled instead of three,
would the inferential methods of this
section still have worked satisfactorily? If
not, where do the potential problems lie?
E5. A statistics class at the University of
WisconsinStevens Point decided to
estimate the average mass of a small bag
of french fries sold at McDonalds. They
bought 32 bags during two different time
periods on two consecutive days at the same
McDonalds and weighed the fries. The data
a. Are the conditions met for constructing a
95% confidence interval for the mean?
construct a 95% confidence interval.
c. The target value set by McDonalds
for the mass of a small order of fries
was 74 g. Is there evidence that this
McDonalds franchise wasnt meeting
that target?
Display 9.12
Mass of small bags of McDonalds fries.

[Source: Nathan Wetzel, McDonalds French Fries.
Would You Like Small or Large Fries? STATS 43
(Spring 2005): 1214.]
E6. The statistics class from E5 decided to

estimate the average number of fries in a
small bag of McDonalds french fries. They
bought 32 bags during two different time
periods on two consecutive days at the same
McDonalds and counted the fries. The data
are given in Display 9.13 (on the next page).
a. Are the conditions met for constructing a
95% confidence interval for the mean?
construct and interpret a 95% confidence
interval.
577
22-03-2009 21:18
Lesson
17 de 18
Display 9.14 Reading level of pages of cancer

brochures. [Source: Thomas H. Short et al.,
Readability of Educational Materials for Patients
with Cancer, Journal of Statistics Education 3, no.
2 (July 1995).]
Display 9.13 Number of fries in small bags of

McDonalds fries.
E7. A page was randomly selected from each of

30 brochures for cancer patients published
by the American Cancer Society and the
National Cancer Institute. The pages were
judged for readability, using standard
readability tests. The reading grade levels
of the 30 pages selected are given in
Display 9.14.
a. Suppose you want to estimate the mean
reading level of cancer brochures. Do you
have the required random sample?
b. Using the data in Display 9.14, construct
a 90% confidence interval.
c. What is the best interpretation of your
interval?
E8. Your grade in your literature class isnt all

that you had hoped it would be. You took
five exams, each consisting of one essay
question worth 100 points. Your scores
were 52, 63, 72, 41, and 73. Your instructor
averages these scores, gets 60.2, and says you
have earned a D. You are, however, doing
very well in your statistics class and think
that what you have learned in this section
can convince your literature instructor
to give you a C, which is a mean score of
between 70 and 79. You remember that your
instructor told the class early in the semester
that the questions on the exams would be
only a sample of the questions that she could
ask. Write a paragraph to your teacher giving
your argument for a C. You can assume
that the instructor knows something about
statistics.
E9. Which confidence interval is widest in each
of parts ac? (You dont need to do any
computing.)
a. 95% interval with n = 4 and s = 10;
90% interval with n = 5 and s = 9;
99% interval with n = 4 and s = 10
22-03-2009 21:18
Lesson
18 de 18
b. 95% interval with n = 3 and s = 10;

c. 90% interval with n = 10 and s = 5;
E10. A large population of measurements
has unknown mean and standard
deviation. Two different random samples
of 50 measurements are taken from the
population. A 95% confidence interval for
the population mean is constructed for
each sample. Which statement would
most likely be true of these two confidence
intervals?
A. They would have identical values for the
lower and upper limits of the confidence
interval.
B. They would have the same margin of
error.
C. The confidence intervals would have the
same center but different widths.
D. None of the above is true.
E11. Suppose a large sample, with n = 100, is
going to be taken from a population of
6-year-old girls. A 90% confidence
interval will be constructed to estimate the
population mean height. A smaller sample,
with n = 50, will also be taken from the
same population of 6-year-old girls, and a
99% confidence interval will be constructed
to estimate the population mean height.
Which of these statements is true about the
chance of producing a confidence interval
that captures the population mean height?
A. The 90% confidence interval based on a
sample of 100 heights has a better
chance.
B. The 99% confidence interval based on a
sample of 50 heights has a better chance.
C. Both methods have an equal chance.
D. You cant determine which method will
have a better chance.
E12. Is the capture rate affected by

a. changes in the sample size?
b. the size of the population standard
deviation?
c. the confidence level?
E13. What happens to the margin of error if all
remains the same except that
a. the standard deviation of the sample is
increased?
b. the sample size is increased?
c. the confidence level is increased?
E14. Refer to the four confidence intervals you
constructed in E2.
a. Which confidence interval has the
smallest margin of error?
b. Why is this the case?
E15. When you do not know the value of
and have to estimate it with s, the capture
rate will be too small if you use z in your
confidence interval. Complete this exercise
to see part of the reason why.
a. In Activity 9.1a, you generated sample
standard deviations, s, from samples
of size 4 taken from a population with
known mean and standard deviation.
Display 9.15 (on the next page) displays
and summarizes 100 sample standard
deviations, s, for random samples of size
4 taken from a normal distribution with
mean, , 511 and standard deviation,
, 112. What is the shape of this
distribution? On average, how well
does s approximate ?
b. Which happened more often, a value of
s smaller than or a value of s larger
than ?
c. How does your answer to part b explain
why the capture rate tends to be too small
if you use z in your confidence interval?
579
22-03-2009 21:18
Lesson
1 de 22
Step 1: If the population is normally

distributed or the sample size is large
enough, 95% of all sample means fall
within 1.96 standard errors of the
population mean.
Step 2: Writing this algebraically, for 95% of
all sample means,
(This interval contains the values of

that are reasonably likely.)
Step 3: You can write this in another,
algebraically equivalent, way. For
95% of all sample means,
Display 9.15
Simulated sampling distribution of

s for samples of size 4 from a normal
distribution with = 112.
E16. Suppose you know , the standard deviation

of the population. (In practice, is almost
never known.) These steps give a justification
for the 95% confidence interval for the
population mean.
9.2
(This interval contains all the

values of that are plausible as the
population mean. In other words,
each population mean in the interval
1.96 SE plausibly could
produce the observed value of .)
Step 4: The inequality in step 3 can be
rewritten using the fact that the
standard error of the sampling
distribution of the sample mean, SE,
.
is or
a. Show algebraically why the inequality in
step 3 is equivalent to that in step 2.
b. Rewrite the inequality based on the
information in step 4.
c. Explain why the inequality in part b is
equivalent to a 95% confidence interval
for the population mean.
A Significance Test for a Mean

Significance tests all rely on the same basic idea: Be suspicious of any model that
assigns low probability to what actually happened. In this section you will learn
when your data tell you to be suspicious of any claimed value of the population
mean.
The basic procedure for testing a hypothesis is always the same: Compute
a test statistic from the data and compare it with a known distribution. As
23-03-2009 22:41
Lesson
2 de 22
in Sections 8.2 and 8.4, you will check conditions for the test, state the null
hypothesis and alternative hypothesis (two-sided or one-sided), find the P-value
to assess the strength of the evidence against the null hypothesis, decide whether
to reject the null hypothesis, and write a conclusion in context. Youll still have
the possibility of making a Type I or Type II error, and sometimes youll do a
fixed-level test with a given significance level .
Activity 9.2a will help you get into the spirit of making an inference about
a mean.
What Makes Strong Evidence?

The four data sets plotted in Display 9.16 each come from an experiment that
measured the differences in pulse rates for pairs of people standing and sitting.
In each experiment, two people with similar pulse rates were paired. One
person was randomly assigned to stand and the other to sit. Then the difference
in the two pulse rates, standing sitting, was computed.
Display 9.16 Differences in pulse rates, standing sitting.

1. Which experiment gives the strongest evidence in favor of a larger mean
pulse rate when people are standing than when they are sitting?
2. Rank the four experiments from strongest to weakest in terms of the
evidence provided. Explain your reasoning.
3. What else would you like to know about these data, in addition to what you
see in the boxplots, in order to improve your inference-making ability?
The Test Statistic

In Section 8.2, you computed this test statistic to test whether a hypothesized
population proportion p0 is consistent with the sample proportion, :
581
23-03-2009 22:41
Lesson
3 de 22
The size of z tells you how many standard errors your sample proportion lies
from the hypothesized proportion, p0. Note that the standard error,
,
can be computed because p0 is stated in the null hypothesis (H0: p = p0) and n is
the known sample size.
What will change if you are testing a mean, H0: = ? It would be nice if you
could use the z-statistic of Section 7.2 to measure how far a sample mean is from a
hypothesized mean :
The catch is that you dont know , the standard deviation of the population.
Perhaps you can guess whats coming next. You do the same thing you did in
Section 9.1. If you dont know the value of , you substitute s for and t for z. The
test statistic becomes
Example: Mean Weight of Pennies

The weights of newly minted U.S. pennies are approximately normally distributed
and are targeted to have a mean weight of 3.11 g. The random sample of nine
pennies shown in Display 9.17 was taken from a production line. Their mean
weight is 3.16 g with standard deviation 0.065. What is the value of the test
statistic for a test to determine whether the mean has moved away from the
target mean?
Display 9.17
Weights of newly minted pennies.
Solution
Constructing a t-Distribution
Before you can start interpreting the new test statistic t, you need to know how
it behaves when the null hypothesis is true. If the value of t from your sample fits
right into the middle of the distribution of t constructed by assuming the null
hypothesis is true, you have no evidence against the null hypothesis. If the value of
t from your sample is way out in the tail of the t -distribution, you have evidence
23-03-2009 22:41
Lesson
4 de 22
against the null hypothesis. In other words, if the value of t from your sample
would be a rare event given the hypothesized mean, be suspicious of that mean.
If the value of t from your sample is a reasonably likely event, the hypothesized
mean is plausible.
The following simulation, based on the distribution of IQ scores, will show
you how the distribution of t is different from that of z.
Some IQ tests are constructed so that scores are normally distributed with
mean = 100 and standard deviation = 15. Suppose you repeatedly take
random samples of size 9 from this distribution and compute the sample mean, .
According to the explanation in Section 7.2, the distribution of
will be normal even with this small sample size (because the population is
normally distributed). The histogram on the left in Display 9.18 shows 1000
values of z, each computed from a random sample of size 9 from the population
of IQs.
If you do not know (the usual situation), you have to use s to estimate it.
You would then compute
The histogram on the right in Display 9.18 shows 1000 values of t, each computed
from a random sample of size 9 from the population of IQs.
Display 9.18
(Left) 1000 values of z computed from random samples taken from a normally
distributed population with mean 100 and standard deviation 15; (right) 1000
values of t computed from random samples from the same population.
Note that the values of t are more spread out than are the values of z. That
makes sense. Not only does vary from sample to sample, but s also varies from
sample to sample. And, as you saw in Section 9.1, s tends to be smaller than
more often than it tends to be larger than , making t larger in absolute value than
the corresponding z more often than it is smaller. Thus, when you compute t, you
tend to get more values in the tails of the distribution than when you compute z.
583
23-03-2009 22:41
Lesson
5 de 22
The box summarizes the most important facts about t-distributions. The
values of t in Table B on page 826 that you used to construct confidence intervals
in Section 9.1 come from these distributions.
The t-Distributions
Suppose you draw random samples of size n from a normally distributed
population with mean and unknown standard deviation. The distribution of
the values of
is called a t-distribution. There is a different t-distribution for each degree of

freedom, df = n 1, where n is the sample size.
A t-distribution is mound-shaped, with mean 0 and a spread that depends
on the value of df. The greater the df, the smaller the spread. The spread of
any t-distribution is greater than that of the standard normal distribution.
Display 9.19 shows the t-distribution for df = 4 plotted on the same graph as
the standard normal distribution.
Display 9.19
The t-distribution for df = 4, and the standard

As the number of degrees of freedom gets larger, the t-distribution more

closely approximates a normal distribution. In Display 9.20, the t-distribution
with df = 9 is very much like the standard normal distribution.
Display 9.20
The t-distribution for df = 9, and the standard

You can explore graphs of t-distributions using your calculator. [See Calculator
Note 9C.]
23-03-2009 22:41
Lesson
6 de 22
P-Values
Now that you know the distribution of t when the null hypothesis is true, the
next step is to locate where the value of t computed from your sample lies on this
distribution. The standard way to do that is to report a P-value, sometimes called
an observed significance level, just as in Section 8.2. The next example will show
you how to find the P-value and how to interpret it.
Example: The P-Value for the Mean Weight of Pennies

As you saw in the example on page 582, the test statistic to see whether the mean
weight of newly minted pennies has moved away from the target, 3.11 g, is
Find and interpret the P-value that corresponds to this value of t.

Computation.
Solution
The P-value for a two-sided test is the area under the t-distribution with df = 9 1,
or 8, that lies above t = 2.31 and below t = 2.31, as shown in Display 9.21.
Display 9.21 P-value for t = 2.31 on a t -distribution with df = 8.
You can get this P-value from your calculator or software. Display 9.22 shows the
P-value 2(0.0248) = 0.0496, or approximately 0.05, on various printouts. [See
Calculator Note 9D to learn how to find a P-value given t and df.]
Display 9.22 P-values from Minitab, Fathom, and the TI-84 Plus.
585
23-03-2009 22:41
Lesson
7 de 22
Interpretation.
If it is true that the mean has not slipped off the target, 3.11 g, there is only a
0.0496 chance of getting an absolute value of t as large as or even larger than the
one from this sample (that is, t 2.31 or t 2.31). The small P-value, 0.0496,
indicates that the sample is somewhat inconsistent with the null hypothesis. It
looks as if the population mean has moved away from the target.
If you do not have a calculator or software that finds P-values, you can get
an estimate of the P-value from Table B on page 826. The next example shows
you how.
Example: Approximating the P-value from Table B

Suppose you have a two-sided test with df = 9 and t = 3.98 (see Display 9.23).
Approximate the P-value for this test using Table B and using a calculator.
Display 9.23
A test statistic, t, of 3.98.
Solution
Go to Table B on page 826 and find the row with 9 degrees of freedom. Go across
the row until you find the absolute value of your value of t, which probably will
lie between two of the values in the table. The partial t-table here shows the two
values of t that lie on each side of |t| = 3.98.
The tail probability gives the area that lies in each tail. Because you have a
two-sided test, you will double the tail probability. If you must use Table B to find
a P-value, all you can say is that the P-value is between 2(0.001), or 0.002, and
2(0.0025), or 0.005. That is, 0.002 < P-value < 0.005. To find the P-value more
precisely, use your calculator to get approximately 0.0032.
23-03-2009 22:41
Lesson
8 de 22
The smaller the P-value,

the stronger the
evidence against the
null hypothesis.
As always, the P-value is a probability, computed under the assumption that

the null hypothesis is true. It tells the chance of seeing a result from a sample that
is as extreme as or even more extreme than the one computed from your data
when the hypothesized mean is correct.
Using a P-Value in a Test of a Mean
The P-value is the probability of getting a random sample, from a distribution
with the mean given in the null hypothesis, that has a value of t that is as
extreme as or even more extreme than the observed value of t computed from
the sample you observed.
When the P-value is close to 0, you have convincing evidence against

the null hypothesis. When the P-value is large, the result from the sample is
consistent with the hypothesized mean and you dont have evidence against the
null hypothesis.
P-Values
D9. Study the formula for the test statistic when testing a hypothesis about a
population mean, and think about the relationship of the test statistic to the
P-value.
a. What happens to the P-value if the sample standard deviation increases
but everything else remains the same?
b. What happens to the P-value if the sample size increases but everything
else remains the same?
To Reject or Not to Reject

P-values are very informative about the strength of the evidence against the
null hypothesis, but sometimes it is advantageous to know at what level you will
reject the null hypothesis even before the data are collected. For what P-values
will you decide to reset the penny machine? For what P-values will you decide
to market the new drug? How much evidence will you require before replacing
the thermostat? When a yes/no or stop/go decision is needed, a fixed-level test of
significance, just as in Section 8.2, has two advantages:
You decide before collecting data when you will reject the null hypothesis in
favor of the alternative hypothesis so that you are not influenced by the data.
You can control the probability of rejecting a true null hypothesis (Type I
error).
587
23-03-2009 22:41
Lesson
9 de 22
When you are asked to do a fixed-level test, you will be given the required level of
significance (or significance level or simply level), , which typically is 5%. To
decide whether to reject the null hypothesis in favor of the alternative hypothesis,
you compare your P-value to .
Fixed-Level Testing
When you use fixed-level testing, you reject the null hypothesis in favor of the
alternative hypothesis if your P-value is less than the level of significance, .
If your P-value is greater than or equal to the level of significance, you do not
reject the null hypothesis.
The significance level is equal to the probability of rejecting the null
hypothesis when it is true (making a Type I error).
The smaller the significance level you choose, the stronger you are requiring
the evidence to be in order to reject H0. The stronger the evidence you require, the
less likely you are to make a Type I error (to reject H0 when it is true). However,
the stronger the evidence you require, the more likely you are to make a Type II
error (to fail to reject H0 when it is false).
Example: Fixed-Level Testing of the Mean Penny Weight

State the null and alternative hypotheses for a test of significance to determine
whether the mean penny weight has moved away from the target, 3.11 g (see the
examples on pages 582 and 585). Can you reject the null hypothesis at a level of
significance, , of 0.01? At = 0.05? At = 0.10?
Solution
The hypotheses are
H0: = 3.11, where denotes the mean weight of all pennies on this
production line
Ha:
3.11
From the example on page 585, the P-value for this test is 0.0496.
Because the P-value is greater than = 0.01, you do not reject the null
hypothesis at that level of significance. Because the P-value is less than = 0.05,
you do reject the null hypothesis at that level of significance. Because the P-value is
less than = 0.10, you do reject the null hypothesis at that level of significance.
Note that, in the previous example, if your calculator or software rounded the
P-value to 0.05, you would not have rejected the null hypothesis at a significance
level, , of 0.05. Thats the major drawback of fixed-level testing: It boils down all
the data to the two choices reject or dont reject. To see what gets lost, think
about a court trial where not guilty can mean anything from This guy is so
innocent he should never have been brought to trial to Everyone on the jury
thinks the defendant did it, but the evidence presented isnt quite strong enough.
In the same way, do not reject communicates only a small fraction of the
23-03-2009 22:41
Lesson
10 de 22
information in the data. Thats why reporting the P-value has become standard in
modern statistical practice.
The t-Test
Tests of significance for means, called t-tests, have the same general structure as
significance tests for proportions, although some details are a bit different.
Components of a Significance Test for a Mean
1. Name the test and check conditions. For a test of significance for a mean,
the methods of this section require that you check three conditions:
Randomness. In the case of a survey, the sample must have been
randomly selected. In the case of an experiment, the treatments must
have been randomly assigned to the experimental units.
Normality. The sample must look like its reasonable to assume that it came
from a normally distributed population or the sample size must be large
enough that the sampling distribution of the sample mean is approximately
normal. There is no exact rule for determining whether a normally
distributed population is a reasonable assumption or for what constitutes a
large enough sample size, but you will learn some guidelines in Section 9.3.
Sample/population size. In the case of a sample survey, the population size
should be at least ten times as large as the sample size.
2. State your hypotheses. The null hypothesis is that the population mean, ,
has a particular value . This is typically abbreviated
The alternative hypothesis can have any one of three forms:
Be sure to define in context.

3. Compute the test statistic, find the P-value, and draw a sketch. The test
statistic is the distance from the sample mean, , to the hypothesized value,
, measured in standard errors:
The P-value given by your calculator or software (or estimated from

Table B) is the probability of getting a value of t that is as extreme as or
even more extreme than the one computed from the actual sample if the
null hypothesis is true.
(continued)
589
23-03-2009 22:41
Lesson
11 de 22
(continued)
4. Write your conclusion linked to your computations and in the context
of the problem. If you are using fixed-level testing, reject the null
hypothesis if your P-value is less than the level of significance, . If the
P-value is greater than or equal to , do not reject the null hypothesis. (If
you are not given a value of , you can assume that is 0.05.)
Write a conclusion that relates to the situation and includes an interpretation
of your P-value.
The next example demonstrates a fixed-level significance test with all four
steps included.
Example: Mass of a Large Bag of Fries

The statistics class at the University of WisconsinStevens Point (see E5 on page
577) also estimated the average mass of a large bag of french fries at McDonalds.
They bought 30 bags during two different half-hour periods on two consecutive
days at the same McDonalds and weighed the fries. The data are given in Display
9.24. McDonalds target value for the mass of a large order of fries was 171 g. Is
there evidence that this McDonalds wasnt meeting that target?
Display 9.24 Mass of large bags of McDonalds fries. [Source: Nathan

Wetzel, McDonalds French Fries. Would You Like Small or Large
Fries? STATS 43 (Spring 2005): 1214.]
23-03-2009 22:41
Lesson
12 de 22
Check conditions and

name the test.
State H0 and Ha.
Solution
The dot plot in Display 9.24 gives no indication that the large population of
masses of large bags of french fries isnt normally distributed. These certainly
arent a random sample of the bags of fries, because they were collected on two
specific days during two half-hour time periods on each day. The mass of the
fries might vary with the person bagging them, and the same person might bag
the fries at those times while other people bag the fries on other days or at other
times. With that in mind, we will proceed with a one-sample t-test for a mean.
The hypotheses are
H0: = 171, where denotes the mean mass, in grams, of all large bags of
french fries produced by this McDonalds
171
Ha:
The statistics from the sample are = 144.07, s = 12.28, and n = 30. The test
statistic is
Compute the test

statistic, find the P-value,
and draw a sketch.
Write a conclusion linked

to your computations
and in the context of the
situation.
The P-value from a calculator is 8.85 1013.

A P-value this close to 0 is very strong evidence against the null hypothesis. A
sample that gives a value of t of 12.01or one even more extremeis almost
never going to occur if the mean mass of the bags is 171 g. So you can reject the
null hypothesis that the mean mass is 171 g.
Even though this was not a true random sample, the sample of bags collected
by these students might be typical of the bags produced during those times or
on that day or by that shift of employees. So any generalizations should be made
cautiously, with careful attention to describing the population to which the results
of the analysis might apply.
You can use your calculator to perform a significance test for a mean. [See
Calculator Note 9E.]
In the next example, you will learn one technique for dealing with outliers.
Example: Distance from the Sun

Earth is 93 million miles from the Sun.
You have probably heard that assertion.
But is it true? Display 9.25 shows
15 measurements made of the average
distance between Earth and the Sun
(called the astronomical unit, or A.U.),
including Simon Newcombs original
measurement from 1895. Test the
hypothesis that the data are consistent
with a true mean A.U. of 93 million miles.
591
23-03-2009 22:41
Lesson
13 de 22
Display 9.25 Measurements of the astronomical unit (A.U.).

[Source: W. J. Youden, Experimentation and Measurement (National
Science Teachers Association, 1985), p. 94.]
Check conditions.
Solution
This type of problem is commonly referred to as a measurement error problem.
There is no population of measurements from which this sample was selected.
The measurements were, however, independently determined by different
scientists and can be thought of as a random sample taken from a conceptual
population of all such measurements that could be made.
The stemplot in Display 9.25 shows that one measurement is extremely large
compared to the others. This sample does not look as if it could reasonably have
been drawn from a normally distributed population. The large measurement,
93.28 million miles, is Newcombs original measurement from 1895 and thus is
different from the more modern measurements. There might be a good scientific
reason to remove it from the data set. So the analysis will be done twice, once with
Newcombs measurement and once without it.
Letting denote the true value of the astronomical unit, the hypotheses are
H0:
State H0 and Ha.
From the data,
= 93 and Ha:
93
= 92.93 and s = 0.112. The test statistic then is given by
Compute the test

and draw a sketch.
23-03-2009 22:41
Lesson
14 de 22

to your computations
situation.
From a calculator, the P-value is 0.0297.

If it is true that the A.U. is equal to 93 million miles, the chance of getting a
mean from a random sample that is as far away as or even farther from 93 million
miles as the mean from this sample is only 0.0297. This is convincing evidence
against the null hypothesis. If there is no systematic bias in the measurements,
you are pretty sure that the astronomical unit differs from 93 million miles.
What about the influence of the outlier? Removing Newcombs original 1895
measurement, 93.28 million miles, gives a new observed test statistic of
The P-value is now 0.0000181. This value is even stronger evidence against the
hypothesized value, 93 million miles.
Whether the outlier is included or not, you come to the same conclusion: The
A.U. is not 93 million miles.
The t-Test
D10. Compare the lists of components of significance tests for a proportion (page
498) and for a mean (page 589). List all the similarities you find. What are
the main differences?
The Meaning of Reject and Do Not Reject
A Type I error is rejecting

a true H0.
Large t and Small P: Reject the Null Hypothesis

A test statistic that is large in absolute value tells you that there is a large
discrepancy between your sample mean and the hypothesized mean, . There
are three ways you can get a large value of t (and therefore a small P-value): The
model is wrong, a rare event occurred, or the null hypothesis is false.
Wrong model: The model you used (that you have a random sample from
a normally distributed population or a random sample with a large enough
sample size) doesnt fit the data.
Rare event: The model fits and the null hypothesis is true, but just by chance
an unlikely event occurred and gave you an unusually large value of the test
statistic. In this situation, you will make a Type I error, rejecting a true null
hypothesis.
Reject is right: The model fits, but the true population mean is not , so
you are right to reject H0.
Once youve checked the normality of the population as best you can or have a
large random sample, you know you have a reasonable modelso either a rare
event occurred or the null hypothesis is false. Standard practice is to reject the
null hypothesis.
593
23-03-2009 22:41
Lesson
15 de 22
A Type II error is failing to

reject a false H0.
Power is the ability to
reject H0.
Small t and Large P: Do Not Reject the Null Hypothesis

What about a small absolute value of the test statistic? Assuming that your
sample fits the conditions for your test, there are two ways you can get a small
value of t (and therefore a large P-value): The hypothesized mean is right, or the
hypothesized mean is wrong but your test doesnt have enough power to allow
you to reject it.
Do not reject is right: The null hypothesis is true, and your sample is
consistent with it. You dont reject the null hypothesisthe correct decision.
Not enough power: When your sample size is small, there can be a lot of
variability in . Thus, the sample mean might end up being fairly close
to the erroneous mean stated in the null hypothesis. You cant reject this
false null hypothesis, and so you have made a Type II error. This occurs
because a significance tests performance (its power) is measured by its
chance of rejecting a hypothesized value. You want the power to be large for
null hypothesized values that are not true. As you learned in Section 8.2,
increasing your sample size is the best way to get more power in your test.

D11. Suppose you are doing a fixed-level test at the 1% significance level. Which
of these questions can be answered?
A. What is the chance that the test will reject the null hypothesis?
B. If the null hypothesis is true, what is the probability that it will be
rejected? That it will not be rejected?
C. If the null hypothesis is false, what is the probability that it will be
rejected? That it will not be rejected?
One-Sided Tests
Sometimes it makes
sense to consider
alternatives that depart
from the standard in
only one direction.
So far in this section, every test has had a two-sided alternative hypothesis: The
true mean, , is not equal to the standard, . But there are two other possible
alternatives: might be less than the standard, or might be greater than the
standard. In real applications, you sometimes can use the context to rule out one
of these two possibilities as meaningless, impossible, uninteresting, or irrelevant.
In Martin v. Westvaco (Chapter 1), a statistical analysis compared the ages
of the workers who were laid off with the ages of workers in the population of
employees working for Westvaco at the time of the layoff. An average age for the
fired workers that was greater than the population mean would tend to support
a claim of age discrimination. On the other hand, the opposite inequality is
not relevant: An average age for the fired workers that was less than the overall
average would not be evidence of age discrimination, because younger workers
arent protected under the law.
When the context tells you to use a one-sided alternative hypothesis, your
P-value is the area on only one side of the t-distribution. Such P- values are called
one-sided or one-tailed P-values, and the corresponding test of significance is
called a one-sided or one-tailed test, just as in Chapter 8.
23-03-2009 22:41
Lesson
16 de 22
Example: One-Sided P-Value

A book about different colleges reports that the mean time students at a particular
university study each week is 1015 minutes. A dean says she believes the mean
is greater than 1015 minutes. To test her claim, she takes a random sample of 64
students and finds that the sample mean is 1050 minutes, with standard deviation
150 minutes. Is this strong evidence in favor of her claim?
Check conditions.
State H0 and Ha.
Solution
A random sample of size 64 would produce a nearly normal sampling distribution
of the sample mean even if the population were not normally distributed. The
university undoubtedly has more than 10(64), or 640, students. This situation
requires a one-sided test, because the dean is interested only in the alternative
(research) hypothesis that the population mean might be larger than reported.
The hypotheses are
H0: = 1015, where denotes the mean number of minutes students study
each week for all the universitys students
Ha: > 1015
From the data, = 1050, s = 150, and n = 64. The test statistic then is given by
Compute the test

and draw a sketch.

to computations and
in the context of the
situation.
The P-value is the area to the right of 1.87 under the t-distribution with
df = 64 1, or 63. From a calculator, the P-value is about 0.0331.
A P-value of only 0.0331 is fairly strong evidence against the null hypothesis
that the mean amount of time spent studying each week is 1015 minutes. Thus,
the data support the deans claim that the mean is greater than 1015 minutes.
Summary 9.2: A Significance Test for a Mean
A claim is made about the value of a population mean, , and you want to
conduct a test of the significance of this claim by taking a random sample from
the population. Then
the null hypothesis gives the hypothesized, or standard, value of
the alternative (research) hypothesis, which can be one- or two-sided,
conjectures how the true mean differs from the standard given in the
null hypothesis
You must check that you have a random sample that is not more than one-tenth
the population size (or a random assignment of treatments to experimental units)
and that it is reasonable to assume either that the sample came from a distribution
that is approximately normal or that the sample size is large enough (to be
explained in Section 9.3).
595
23-03-2009 22:41
Lesson
17 de 22
Just as in constructing a confidence interval for a mean, when you use s to

estimate you must use t rather than z in doing a significance test for a mean.
You can use the test statistic
Instead of doing a fixed-level test with given significance level , you can
report a P-value in order to provide more information about the strength of
evidence against the null hypothesis than you provide by simply saying reject
or dont reject.The P-value gives the probability of seeing a test statistic t as
extreme as or even more extreme than the one computed from the data when
the null hypothesis is true. It serves as a measure of the evidence against the null
hypothesis. The smaller the P-value, the greater the evidence in the sample against
the null hypothesis. You can get exact P-values from a calculator or statistical
software. You can get approximate P-values from Table B on page 826.
If you reject a true null hypothesis, you have made a Type I error. If you fail
to reject a false null hypothesis, you have made a Type II error. The power of your
test depends on its ability to reject the null hypothesis. The larger the sample size
and the larger the level of significance, the greater the power of your test to reject
a false null hypothesis.
Practice
The Test Statistic
P10. The thermostat in your classroom is set
at 72F, but you think the thermostat isnt
working well. On seven randomly selected
days, you measure the temperature at
your seat. Your measurements (in degrees
Fahrenheit) are 71, 73, 69, 68, 69, 70, and 71.
What is the test statistic for a significance
test of whether the mean temperature at your
seat is different from 72F?
P11. In 2003, Pell Grants totaling $166,733,957
were awarded to 76,525 Minnesota college
students. You believe that the mean amount
has changed since 2003. You take a random
sample of 35 Minnesota college students
who currently receive Pell Grants and get a
mean amount of $2317.14, with a standard
deviation of $754.00. What is the test statistic
for a test to determine whether the mean
Pell Grant amount for Minnesota college
students has changed since 2003? [Source:
www.ohe.state.mn.us.]
Constructing a t-Distribution
P12. One of the distributions in Display 9.26 is a
t-distribution, and the other is the standard
normal distribution. Which is which?
Explain how you know.
Display 9.26 A t-distribution and a standard normal

distribution.
P13. You plan to use simulation to construct an

approximate t-distribution with 6 degrees
of freedom by taking random samples from
a population of IQ scores that are normally
distributed with mean, , 100 and standard
deviation, , 15.
23-03-2009 22:41
Lesson
18 de 22
a. Describe how you will do one run of this

simulation.
b. Produce three values of t using your
simulation.
P-Values
P14. In P10, you computed the test statistic
for testing that the mean temperature at your
seat isnt 72F. Find and interpret the P-value.
P15. In P11, you computed the test statistic for
testing whether the mean Pell Grant amount
for Minnesota college students has changed
since 2003. Find and interpret the P-value.
To Reject or Not to Reject
P16. In P14, you found the P-value for testing that
the mean temperature at your seat isnt 72F.
a. State the hypotheses.
b. Would you reject the null hypothesis at the
10% level? The 5% level? The 1% level?
P17. In P15, you found the P-value for testing
whether the mean Pell Grant in Minnesota
has changed since 2003.
b. Would you reject the null hypothesis
at the 10% level? The 5% level? The
1% level?
P18. You get the SAT scores of a random sample
of nine students from State University
and calculate a sample mean of 1535 with
standard deviation 250. A university official
states that the mean SAT score for students
at the school is 1700. You are asked to
validate his claim.
b. Would you reject the hypothesis that
State University has a mean SAT score of
1700 at the 10% level? The 5% level? The
1% level?
The t-Test
P19. For the aldrin data of P1 on page 573, carry
out a test of whether the true mean differs
from 4 nanograms per liter. Use = 0.05.
P20. The question about mean body temperature
addressed with the data in Display 9.3 on
page 567 could have been phrased differently:

Does the mean body temperature differ from
98.6F? Do these data provide convincing
evidence that the population mean for
males differs from 98.6F? Answer the same
question for the female population. State
your conclusions in terms of P-values.
P21. Return to the significance tests indicated and
tell whether a Type I error, a Type II error,
both errors, or neither error might have been
made.
a. french fries example on pages 590591
b. astronomical unit example on pages
591593
c. aldrin, P19
P22. In each situation below, assume the null
hypothesis is actually false. Tell which test
will have more power, all else being equal.
a. a test with n =12; a test with n = 16
b. a test with = 0.05; a test with = 0.01
One-Sided Tests
P23. The board of the Gainesville Home Builders
Association knows the selling price of all
new houses sold in the city last month and
believes that the mean selling price has gone
up this month. To check this claim, the
board obtains a random sample of the selling
prices of houses sold this month.
a. Should the alternative hypothesis be
one-sided or two-sided?
b. Tell what and are for the problem.
c. State the null and alternative hypotheses.
P24. The situation is the same as in P10 on page
596. The thermostat in your classroom is set
at 72F, but this time, before you collect your
data, you are convinced that the room tends
to be colder. On seven randomly selected
days, you measure the temperature at your
seat. Your measurements are 71, 73, 69, 68,
69, 70, and 71. Perform all four steps of a test
of whether the mean temperature at your
seat is lower than 72F.
597
23-03-2009 22:41
Lesson
19 de 22
Exercises
E17. In E3 on page 576, Jack and Jill opened a
water-bottling factory. The distribution of
the number of ounces of water in the bottles
is approximately normal. The mean, , is
supposed to be 16 oz, but the water-filling
machine slips away from that amount
occasionally and has to be readjusted. Jack
and Jill take a random sample of ten bottles
from todays production and weigh the water
in each. The weights (in ounces) are
15.91 16.08 16.08 15.94 16.02
15.94 15.96 16.03 15.82 15.96
Should Jack and Jill readjust the machine?
Do all four steps in a test of significance,
writing the conclusion in terms of the
P-value.
E18. In E5 on page 577, a statistics class decided
to check the weights of bags of small fries at
a local McDonalds to see if, on average, they
met the target value of 74 g. They bought
32 bags during two different time periods on
two consecutive days and weighed the fries.
The data are given in Display 9.12 on page
577. Is there evidence that this McDonalds
wasnt meeting its target? Do all four steps in
a test of significance, writing the conclusion
in terms of the P-value.
E19. Fifteen students were given pieces of paper
with five vertical line segments and asked to
mark the midpoint of each segment. Then,
using a ruler, each student measured how
far each mark was from the real midpoint
of each segment. The students then each
averaged their five errors. If the average is
positive, the student tended to place the
midpoint too high. If the average is negative,
the student tended to place the midpoint too
low. The results are given in Display 9.27.
Is there evidence of statistically significant
measurement bias? That is, on average, do
students tend to place the midpoints either
too high or too low? Do all four steps in a
test of significance, writing the conclusion in
terms of the P-value.
Display 9.27 Average error in estimating the

location of the midpoint of vertical line
segments. (A positive value indicates
that the average estimate was too
high.) [Source: Gretchen Davis, Beware of
Measurement Bias, CMC ComMuniCator 22,
no. 4 (June 1998): 5456.]
E20. Refer to E19. Fourteen different students

were given the same instructions, but with
pieces of paper with five horizontal line
segments. The results are given in Display
9.28. Is there evidence of statistically
significant measurement bias? That is,
on average, do students tend to place the
midpoints either too far to the right or too
far to the left? Do all four steps in a test of
significance, writing the conclusion in terms
of the P-value.
23-03-2009 22:41
Lesson
20 de 22
E22. At sea level, water boils at 212F. Theory

says that, at high altitude, water boils at a
lower temperature than it does at sea level.
You plan to test this theory by flying to the
mile-high Denver airport and observing the
boiling point of water there.
E23. Munchies Potato Chip Company claims that
the weight of the contents of a 10-oz bag of
chips is normally distributed, with mean
10 oz. A consumer group, Snack Munchers
for Truth (SMFT), says that the average
weight is less than this. SMFT weighs the
contents of 15 randomly selected bags of
potato chips and gets the weights in Display
9.29. Test SMFTs claim, including all four
steps of a significance test and writing the
conclusion in terms of the P-value.
Display 9.28 Average error in estimating the

location of the midpoints of horizontal
line segments. (A positive value
indicates that the average estimate
was too far to the right.)
E21. Theory says that if you dissolve salt in

water, the freezing point will be lower
than it is for pure water (32F). To test the
theory, you plan to dissolve a teaspoon of
salt in a bowl of water, put a thermometer
into the mixture, and then put the bowl in
your freezer, checking periodically so you
can observe the temperature at which the
solution begins to freeze. You will repeat
this ten times. (If you actually try this, you
need to check often enough to be sure to
remove the thermometer once ice begins to
form, or you might end up with a broken
thermometer.)
Display 9.29 Weights of bags of chips.
E24. A sample of 15 monthly rents for twobedroom apartments was selected from
a recent edition of the Gainesville Sun. The
15 values are symmetric, with no outliers.
9.2 A Significance Test for a Mean 599
23-03-2009 22:41
Lesson
21 de 22
The mean of this sample is $636 with

standard deviation $121.
a. A group of students thinks the mean rent
for two-bedroom apartments is at least
$650. Test their claim by doing a test of
significance, writing the conclusion in
terms of the P-value.
b. A newspaper reports that the average
monthly rent for two-bedroom
apartments last year was $500. Is there
statistical evidence of a change in mean
rent from last year to this year? Test this
claim by doing a test of significance,
writing the conclusion in terms of the
P-value.
E25. The heights of young women in the
United States are approximately normally
distributed, with mean 64.8 in. The heights
of the 11 players on a recent roster of the
WNBA Chicago Sky basketball team are (in
inches) 72, 74, 76, 71, 76, 68, 69, 76, 71, 73,
and 70. [Source: www.wnba.com.] Is there sufficient
evidence to say that the mean height of
these 11 players is so much larger than the
population mean that the difference cannot
reasonably be attributed to chance alone?
(You might want to refer to Inference for an
Observational Study in Section 8.5.)
E26. The heights of young men in the United
States are approximately normally
distributed, with mean 70.1 in. The heights
of the young men in the Hamilton family are
(in inches) 71, 72, 71, 72, and 70.
a. Is there sufficient evidence to say that the
difference between the mean height of
these family members and the population
mean cannot reasonably be attributed to
chance alone? (You might want to refer to
Inference for an Observational Study in
Section 8.5.)
b. Has the null hypothesis been rejected
primarily because the sample size is so
small, because the mean is so close to
that of all young men, or because the
variability is so small in the Hamilton
family?
E27. Return to the significance test in E17.

Rewrite the conclusion for a fixed-level test
with equal to 0.10.
Rewrite the conclusion for a fixed-level test
with equal to 0.04.
a. Rewrite the conclusion for a fixed-level
test at the 0.05 level.
b. In fact, these values were selected at
random from a normal distribution with
mean 9.9. Has an error been made? If so,
which type?
c. Could you have reached the conclusion
in part a based only on a confidence
interval? Explain.
E30. Return to the significance test in E24, part b.
a. Rewrite the conclusion for a fixed-level
test with equal to 0.02.
b. Could you have reached the conclusion
in part a based only on a confidence
interval? Explain.
E31. Select the best answer. A P-value measures
A. the probability that the null hypothesis is
true
B. the probability that the null hypothesis is
false
C. the probability that an alternative
hypothesis is true
D. the probability of seeing a value of t at
least as extreme as the one observed,
given that the null hypothesis is true
E32. Select each correct description of the power
of a test.
A. gives the probability of rejecting the null
hypothesis
B. gives the probability of failing to reject
the null hypothesis
C. increases by increasing the sample size
D. increases by increasing the level of
significance
E. isnt an issue if the null hypothesis is true
23-03-2009 22:41
Lesson
22 de 22
E33. Suppose you have a sample of size 23 and the

value of your test statistic is 1.645.
a. Draw a sketch of a t-distribution and
shade the area that corresponds to the
P-value for a two-sided test.
b. Identify the part of the shaded area that
corresponds to evidence that > 0.
What is the relationship between the
two-sided and one-sided P-values for
a given value of the test statistic?
c. If all else is equal and the alternative
hypothesis is in the right direction, will
the P-value be larger for a one-sided test
or a two-sided test?
E34. For the aldrin data of P1 on page 573, find
the P-value for each alternative hypothesis.
a. Ha: 4
b. Ha: > 4
c. Ha: < 4
E35. Suppose H0 is true and you are using a
significance level of 5%. Which gives a larger
chance of a Type I error, a one-sided test or a
two-sided test? Explain.
E36. If the null hypothesis is true, does using
= 0.05 or = 0.01 give a larger chance of
a Type I error?
E37. In each part, the null hypothesis is actually
false. Which test will have more power, all
other things being equal?
a. a test with = 0.01; a test with = 0.10
b. a test with n = 45; a test with n = 29
c. a one-sided test; a two-sided test
E38. Suppose > 0 and you are using a
significance level of 5%. Which of these
three tests has the greatest power (probability
of rejecting the false null hypothesis that
= 0)?
A. a one-sided test of H0 versus Ha: 0
B. a two-sided test
C. a one-sided test of H0 versus Ha: 0
E39. Histograms A and B in Display 9.30 were
generated from 200 random samples of size
4, each selected from a normal distribution
with = 100 and = 20. One histogram

shows the 200 z-values (one for each sample,
using ) for testing the hypothesis that the
population mean is, in fact, 100, and the
other shows the 200 t-values (using s).
Display 9.30 Histograms of the t-values and

z-values computed for 200 random
samples of size 4 taken from a normal
distribution with mean 100 and
standard deviation 20.
a. Compare the shapes, centers, and spreads

of these two distributions.
b. Choose which is the distribution of
t-values, and give the reason for your
choice.
E40. Degrees of freedom tell how much
information in your sample is available for
estimating the standard deviation of the
population. In a t-test, you use s to estimate
. The more deviations from the mean,
(x ), you have available to use in the
formula for s, the closer s should be to .
But, as you will see in this exercise, not all
of these deviations give you independent
information.
a. For a sample of size n, how many
deviations from the mean are there?
What is their sum? (See Section 2.3.)
b. Suppose you have a random sample of
size n = 1 from a completely unknown
population. Call the sample value x1.
Does the value of x1, by itself, give you
information about the spread of the
population?
c. Next suppose you have a sample of
size n = 2, with values x1 and x2. If one
deviation, x1 , is 3, what is the other
deviation?
d. Now suppose you have a sample of size
n = 3. Suppose you know two of the
601
23-03-2009 22:41
Lesson
1 de 14
deviations, x1 = 3 and x2 = 1
What is the third deviation?
e. Finally, suppose you have a sample of size
n. Show how to find the final deviation
once you know all the others. With a
sample of size n, how many deviations
9.3
give you independent information

about the size of and how many are
redundant?
f. Explain what your answers to parts ae
have to do with degrees of freedom.
When Things Arent Normal

If you think about what youve done so far in terms of the big three for
distributionscenter, spread, and shapeyoull see that theres more work
to do.
1. Center: The population mean is the goal of our inference. You dont know
the mean, but thats the whole point: If you knew the mean, you wouldnt be
trying to estimate it.
2. Spread: In real applications, you almost never know the population standard
deviation in confidence intervals and significance tests, so you have to use s
as an estimate of . In Section 9.1, you saw that, for small samples, using s in
place of has a pretty big effect on confidence intervals. Unless you make a
suitable adjustment to the width of the interval by using t* instead of z*, its
capture rate will be way off.
3. Shape: Officially, inference using the t-distribution requires a normally
distributed population. But official rules dont always have to be followed
exactly. You can safely get away with less than a normally distributed
population, because the t-procedures are robust (not very sensitive) to
departures from normality. Youll still have to check a condition about shape,
which will be described on pages 607608.
First, you will see the reason for the condition requiring a normal distribution
by investigating what happens to the capture rate when small samples are taken
from a skewed population.
The Effect of Skewness and Outliers

In the examples so far, the shape of the population has been approximately
normally distributed or the sample size has been large, but often thats not the
case. Display 9.31 shows the brain weights of 68 species of animals used in a study
of sleep. These 68 species will serve as the population. This distribution is highly
skewed toward the larger values. In Activity 9.3a, you will observe the effect the
skewness has on the capture rate of confidence intervals.
22-03-2009 21:20
Lesson
2 de 14
Display 9.31 Brain weights for a selection of species. [Source:

T. Allison and D. V. Cicchetti, Sleep in Mammals: Ecological and
Constitutional Correlates, Science 194 (1976): 73234.]
603
22-03-2009 21:20
Lesson
3 de 14
The Effect of Skewness on Confidence Intervals

What youll need: the data set of Display 9.31
Your goal is to see what happens to the capture rate when you construct many
confidence intervals based on random samples taken from the population of
brain weights in Display 9.31.
1. Decide on a way to get a random sample of five of the species. Select the
sample and record the brain weights of the five species in your sample.
2. Construct a 95% confidence interval estimate of the mean brain weight of
all the species using the sample mean, the sample standard deviation, and
the appropriate value of t*.
3. Does your confidence interval include the true mean weight, 394.49 g? If
not, does the interval lie below the true mean or above it?
4. Repeat steps 1 through 3 with a second sample of size 5.
5. Combine your answers with those of other groups in the class to compute
an estimate of the overall capture rate. Is it close to 95%?
6. Looking at the capture rate from your simulation and at the locations of the
intervals that do not capture the population mean, write a brief statement
about what happens to confidence intervals when techniques based on
the normal distribution are used with distributions that are highly skewed
toward large values.
The lesson to be learned from Activity 9.3a is that if you have data from a
distribution that is far from normal, confidence intervals and hypothesis tests
might not behave the way they are supposed to.
Skewness means
that the capture
rate will be too
small.
The Effect of Skewness and Outliers

If the sample size is small and if the underlying distribution is highly skewed
rather than normal or has extreme outliers, then the capture rate for an interval
of the form
might be substantially lower than the advertised
capture rate, for example, 95%, and a test of significance based on the normal
distribution will falsely reject H0 substantially more often than the advertised
rate, for example, 5%.
What If My Population Is Not Normal?
Always plot your data to

check for skewness and
outliers.
What can you do about skewness and outliers? Most important, before you do
any tests or construct any intervals, always plot your data to see their shape. If the
plot looks as if the data came from a normal distribution, you dont need to worry
about shape. On the other hand, if your plot shows any major deviations from a
normal shape, you will have to decide whether to try one of the three approaches
discussed next.
22-03-2009 21:20
Lesson
4 de 14
I. Try a Transformation.
Fortunately, skewed distributions almost always can be made much more nearly
symmetric by transforming them to a new scale. If you can find a change of scale
that makes your data look roughly normal, once again you dont have to worry
about shape. Sometimes a change of scale will also take care of what at first
looked liked outliers, making them more like just part of the herd. [See
Calculator Note 3J to review how to perform shape-changing transformations.]
For a distribution that

is skewed right (a
common shape), try a
log transformation.
Example: The Log Transformation

Take the logarithm of each brain weight in Display 9.31. Is it now safe to proceed
with a confidence interval based on t?
Solution
Display 9.32 shows a histogram of the brain weights after taking the natural
logarithm of each value. Notice how the outliers have been drawn in toward
the center of the data in a way that makes the distribution much more nearly
symmetric. This distribution still is not normal, but it is symmetric enough to
make the distribution of the sample means for 100 samples of size 5 in Display
9.33 look nearly symmetric. Thus, the confidence interval using t will work well
on the transformed data.
Display 9.32 Logarithms of brain

weights.
Display 9.33 Sample means for

100 samples of size 5
from the logarithms
of brain weights.
Example: The Reciprocal Transformation

Gasoline mileage for cars typically is computed in miles per gallon. But why
not use gallons per mile? Display 9.34 shows estimated city miles per gallon
(mi/gal) for a random sample of compact car models, along with the gallons
per mile (gal/mi). Which distribution is better suited for inference? Construct
and interpret a 95% confidence interval for the population mean of that
distribution.
605
22-03-2009 21:20
Lesson
5 de 14
Display 9.34 City gas mileage for a random sample of compact

models. [Source: EPA, www.fueleconomy.gov.]
Check conditions.
Solution
The boxplots show that the gallons per mile distribution doesnt have an outlier
and is less skewed than the miles per gallon distribution and thus should be better
suited to inference.
The second condition for constructing a confidence interval for a mean also
is metyou were told this is a random sample taken from the population of
compact car models. The third condition is met if there are at least 10(15), or 150,
models of compact cars. Because this isnt the case, the confidence interval will be
wider than necessary for 95% confidence.
0.0405, s 0.00669, and n = 15. For
The statistics from the sample are
df = 15 1, or 14, use t* = 2.145. A 95% confidence interval for the sample mean is
Show computations.
Give conclusion in
context.
You are 95% confident that the mean gallons per mile of all models of compact cars
lies in the interval from 0.0368 to 0.0442. So, any mean gal/mi in that interval could
have produced the result from the sample as a reasonably likely outcome.
Applying the methodology of this chapter to gal/mi measurements will,
over many random samples, produce intervals with a capture rate closer to the
nominal 95% than would using miles per gallon measurements.
22-03-2009 21:20
Lesson
6 de 14
When the data are

ratios, try a reciprocal
transformation.
One lesson of the previous example is that whenever data come in the form
of a ratio, think about what might happen if you inverted the ratio. Similarly, a
rate like 30 customers are served per hour might just as naturally be expressed
as 2 minutes per customer.
II. Do the Analysis With and Without the Outliers.
Dont remove outliers

without a good reason.
A robust procedure
works okay even if
certain conditions are
not met.
If changing the scale doesnt take care of the outliers, do two parallel analyses, one
with all the data and the other with the outliers removed. Thats what was done
in the example on pages 591593. If both analyses lead to the same conclusion,
youre all set. But if the two conclusions differ, you need more data: You dont want
your conclusion to depend on what you assume about one or two observations!
III. Get a Large Sample Size.
The worst cases are small samples with extreme skewness or with extreme outliers.
For moderate sample sizes, you can rely on the robustness of the t-procedures:
The t-procedures are comparatively insensitive to departures from normality,
especially for larger samples. What this means in practice is that the true capture
rates and significance levels will be close to the advertised values except in extreme
situations. For large samples, the value of s wont vary much from sample to sample
and the Central Limit Theorem says that the sampling distribution of will be
approximately normal, so it is safe to use the t-procedure.
As a rough guide, you can rely on the 15/40 rule, described in the box.
15/40 Guideline for Inference Using t-Procedures

First, plot your data. A boxplot is especially helpful in identifying shape and
outliers. If your random sample looks as if it reasonably could have come from
a normal distribution, you can proceed. If you suspect that the data didnt come
from a normal distribution, follow the guideline for your sample size.
If your sample size is less than 15: Be very careful. Your data or transformed
data must look as if they came from a normal distributionlittle skewness,
no outliers.
If your sample size is between 15 and 40: Proceed with caution. Strongly
skewed distributions should be transformed to a scale that makes them
more nearly symmetric before you use a t-procedure. If you have extreme
outliers, a transformation might be in order. If you dont transform the data
or if the outliers remain even after a change of scale, do two versions of
your significance test or confidence interval, one with and one without the
outliers. Dont rely on any conclusions that depend on whether you include
the outliers.
If your sample size is over 40: Youre in good shape. Your sample size is large
enough that skewness will not reduce capture rates or alter significance
levels enough to matter. Still, if your sample shows strong skewness, it is
worth asking whether a change of scale would make the usual summary
(continued)
607
22-03-2009 21:20
Lesson
7 de 14
(continued)
statistics (especially the standard deviation) more meaningful. Even though
outliers might not have much effect on capture rates or significance levels,
you still should check by doing two versions of your t-procedure.
Display 9.35 The 15/40 guideline for inference based on the

t-distribution.

D12. Discuss whether a transformation might be necessary in order to estimate
the population mean in each scenario.
a. You are taking a sample of size 10 from the population of prices of
single-family houses in your city.
b. You are taking a sample of size 100 from the population of prices of
single-family houses in your city.
c. You are taking a sample of size 5 from the SAT scores of students
entering a college as freshmen.
d. You are taking a sample of size 20 from the waiting times of customers at
a bank drive-up window.
Summary 9.3: When Things Arent Normal

Small samples from a skewed population will tend to produce confidence intervals
that are too narrow and off-center. The capture rate of such intervals will be less than
the confidence level. If the sample size is small, be careful! If the sample size is large,
however, the skewness of the population is of little consequence. If the sample size is
moderate, then the sampling distribution might be slightly skewed, but t-procedures
are robust to slight departures from normality; that is, they still will work well.
Always plot your data first and then proceed with a confidence interval or test
of significance using the 15/40 guideline:
If the sample size is less than 15, make sure your data or transformed data
have little skewness and no outliers.
If the sample size is between 15 and 40, your data or transformed data
should be nearly symmetric, with no extreme outliers. (If you dont transform
the data or if the outliers remain even after a change of scale, do two versions
22-03-2009 21:20
Lesson
8 de 14
of your significance test or confidence interval, one with and one without the
outliers. Dont rely on any conclusions that depend on whether you include
the outliers.)
If the sample size is over 40, its safe to proceed. If your sample shows strong
skewness, it still is worth asking whether a change of scale would make the
usual summary statistics (especially the standard deviation) more meaningful.
With small samples, in addition to constructing plots, think carefully about
the nature of the population before concluding that approximate normality is a
reasonable assumption.
Practice
P25. Pretend that each data set (AD) is a random
sample and that you want to do a significance
test or construct a confidence interval for
the unknown mean. Use the sample size and
the shape of the distribution to decide which
description (IIV) best fits each data set.
A. weights of bears (Display 2.6 on page 33)
B. number of passengers at airports (Display
7.53 on page 462)
C. speed of mammals (Display 2.25 on
page 44)
D. female life expectancy in Africa and
Europe (Display 2.53 on page 69)
I. There are no outliers, and there is no
evidence of skewness. Methods based
on the normal distribution are suitable.
II. The distribution is not symmetric, but
the sample is large enough that it is
reasonable to rely on the robustness
of the t-procedure and construct
a confidence interval, without
transforming the data to a new scale.
III. The shape suggests transforming. With
a larger sample, this might not be
necessary, but for a skewed sample of
this size transforming is worth trying.
IV. It would be a good idea to analyze this
data set twice, once with the outliers and
once without.
P26. Display 9.36 shows the number of fries in
30 large bags, similar to the example on
page 590. Suppose you want to test that the
mean number of fries in a large bag is 90.
Display 9.36 Number of fries in a large bag. [Source:

Nathan Wetzel, McDonalds French Fries. Would
You Like Small or Large Fries? STATS 43 (Spring
2005): 1214.]
a. How would you suggest handling the

analysis in this case? Give your reasons.
b. Do your analysis and report your results.
609
22-03-2009 21:20
Lesson
9 de 14
Exercises
For E41 and E42: Pretend that each data set
described is a random sample and that you want
to do a significance test or construct a confidence
interval for the unknown mean. Use the sample
size and the shape of the distribution to decide
which of these descriptions (IIV) best fits each
data set.
I. There are no outliers, and there is no
evidence of skewness. Methods based
on the normal distribution are suitable.
II. The distribution is not symmetric, but
the sample is large enough that it is
reasonable to rely on the robustness
of the t-procedure and construct
a confidence interval, without
transforming the data to a new scale.
III. The shape suggests transforming. With
a larger sample, this might not be
necessary, but for a skewed sample of
this size transforming is worth trying.
IV. It would be a good idea to analyze this
data set twice, once with the outliers and
once without.
E41. See the instructions above.
A. weights, in ounces, of bags of potato chips
B. per capita gross domestic product (GNP)

for various countries
C. batting averages of American League

players
D. self-reported grade-point averages of

67 students
E42. See the instructions above.

A. mean number of people per room for
various countries
B. record low temperatures of national

capitals
C. student errors in estimating the midpoint

of a segment
22-03-2009 21:20
Lesson
10 de 14
D. ages of employees
E43. Health insurance companies look for ways

to lower costs, and one way is to shorten the
length of hospital stays. A study to compare
two large insurance companies on length
of stay (LOS) for pediatric asthma patients
randomly sampled 393 cases from Insurer A.
Summary statistics and a histogram of the
data are shown in Display 9.37.
Display 9.37 Summary and data plot for lengths of

hospital stays for Insurer A. [Source:
R. Peck, L. Haugh, and A. Goodman, Statistical
Case Studies, 1998, ASA-SIAM, 4564.]
a. Is it appropriate to construct a confidence

interval without transforming these data?
Explain.
estimate the mean LOS for Insurer A in a
90% confidence interval.
c. Would you be more concerned about
constructing a confidence interval
without a transformation if the sample
size was 40 instead of nearly 400? What
about a sample size of 4 instead of
nearly 400?
E44. An independent random sample of 396 cases
from Insurer B gave the results for length of
stay summarized in Display 9.38.

hospital stays for Insurer B.
a. Is it appropriate to construct a confidence

interval without transforming these data?
Explain.
estimate the mean LOS for Insurer B in a
c. Would you be more concerned about
constructing a confidence interval
without a transformation if the sample
size was 40 instead of nearly 400? What
about a sample size of 4 instead of
nearly 400?
d. Compare your confidence intervals in
part b of E43 and E44. Does it appear that
the two insurers differ in mean LOS?
E45. Display 9.39 (on the next page) gives the
location and mass of 31 black holes. Assume
that these can be considered a random
sample of all black holes in the universe.
a. Have the conditions been met for
constructing a confidence interval for a
mean? If not, how would you recommend
that the analysis proceed?
b. Construct and interpret a 90% confidence
interval for a mean, proceeding in the
way you recommended in part a.
611
22-03-2009 21:20
Lesson
11 de 14
E46. The ages, in years, of a random sample of

16 pennies are 1, 1, 1, 1, 1, 1, 2, 2, 3, 4, 4, 11,
11, 12, 26, and 30. You want to test that the
mean penny age is 7 years.
a. The boxplot in Display 9.40 shows that 30
is an outlier. Should it be removed from
the data set? Explain why or why not.
Display 9.40 Boxplot of the ages of 16 pennies.
b. Are the conditions met for a significance

test of the mean age of all pennies? If not,
what transformation would you suggest?
Give your reason.
c. Try your transformation. Are the
conditions better for inference?
E47. Look again at the brain weight data in
Display 9.31. This time, consider it a random
sample taken from all animal species.
a. If you omit the three obvious outliers,
you get the boxplot in Display 9.41. Has
removing the three outliers improved the
situation?
Display 9.41 Boxplot of brain weights with outliers

removed.
b. Construct the 95% confidence interval for

the mean using the set of all 68 species.
394.5
The summary statistics are
and s 1207.0. Then construct the 95%
confidence interval for the mean with
the three original outliers removed. The
summary statistics are
149.4 and
s 323.2.
Display 9.39 Location and mass of 31 black holes.
22-03-2009 21:20
Lesson
12 de 14
c. How do the centers of the two confidence

intervals in part b compare? The
widths? What does this tell you about
the sensitivity of confidence intervals to
outliers?
d. If you take the natural log of each of
the original 68 brain weights, you get
the boxplot in Display 9.42 and these
summary statistics:
2.977 and
s 2.686. Construct and interpret a 95%
confidence interval for the population
mean on this transformed scale.
lakes deserved the headline? Justify your

answer statistically.
Display 9.43 Mercury content of fish in Maine lakes.

[Source: R. Peck, L. Haugh, and A. Goodman,
Statistical Case Studies, 1998, ASA-SIAM, 114.]
Display 9.42 Boxplot of ln(brain weights).
E48. Fish absorb mercury as water passes through

their gills, and too much mercury makes the
fish unfit for human consumption. In 1994,
the state of Maine issued a health advisory
warning that people should be careful
about eating fish from Maine lakes because
of the high levels of mercury. Before the
warning, data on the status of Maine lakes
were collected by the U.S. Environmental
Protection Agency (EPA) working with
the state. Fish were taken from a random
sample of lakes, and their mercury content
was measured in parts per million (ppm).
Display 9.43 shows a subset of those data
from a random sample of 35 lakes.
a. Are there any problems with computing a
confidence interval estimate of the mean
mercury levels in fish for the lakes of
Maine using the values in the table?
If so, how do you recommend handling
the analysis?
b. One newspaper headline proclaimed,
Mercury: Maine Fish Are Contaminated
by this Deadly Poison. Most states
consider mercury levels of 0.5 ppm as the
borderline for issuing a health advisory.
Does it appear that, on average, Maine
E49. Almost all of the rain in Los Angeles falls in

the winter, so a season of rainfall is measured
from July 1 through the end of June.
Display 9.44 gives a summary of the rainfall
in Los Angeles for the last 128 seasons.
Display 9.44 Los Angeles rainfall, in inches per

season, for 128 seasons.
a. You want to estimate mean seasonal

rainfall in a confidence interval and
can afford to check a random sample
of only seven seasons. If you suspect
that the population has a shape like that
in the histogram in Display 9.44, can
you proceed, or should you consider a
transformation?
b. Display 9.45 shows a reciprocal
transformation of the rainfall data. The
613
22-03-2009 21:20
Lesson
13 de 14
mean of the distribution is about 0.081.

Does this mean have a meaningful
interpretation? Are you satisfied with the
conditions for inference now?
E50. Display 9.47 shows a random sample of

7 seasons taken from the 128 seasons of
Los Angeles rainfall in E49.
Display 9.45 Los Angeles rainfall, in years per inch,

for 128 seasons.
c. Display 9.46 shows a log transformation

of the rainfall data. Does this
transformation do a better or worse job
than the reciprocal transformation in
satisfying the conditions for inference?
Display 9.46 Log transformation of inches per

season of Los Angeles rainfall.
Display 9.47 Los Angeles rainfall in a random

sample of 7 of the last 128 seasons.
a. Assuming you havent seen the

population in E49, are the conditions met
for constructing a confidence interval for
mean seasonal rainfall?
c. The mean seasonal rainfall is 15.06 in.
Did you capture the true mean in your
confidence interval in part b?
d. Is the skewed shape of the population
reflected in this small sample? If you take
the log of each rainfall amount, are the
conditions for inference better satisfied
or worse? (What to do when you have a
small sample isnt always clear. Fortunately,
the t-procedure is fairly robust.)
d. If, instead of taking a random sample of

size 7, you were able to consider these
128 seasons to be a random sample of
Los Angeles rainfall over, say, the last
2000 seasons, would you be concerned
about constructing a confidence
interval for the mean seasonal rainfall
directly from the original data (with no
transformation)?
22-03-2009 21:20
Lesson
14 de 14
E51. How should you measure the relationship

between housing units (H) and the number
of people living in them (P)as persons
per housing unit, or as housing units per
person? The data in Display 9.48 give the
population and number of housing units
for a random sample of ten counties in
Florida.
E52. According to the popular press, some

kinds of thinking (visual tasks) are
right-brained, whereas others (verbal tasks)
are left-brained. There is a fair amount
of scientific support for this theory. For
example, one experiment invented by L. R.
Brooks involves two kinds of tasks. [Source:
L. R. Brooks, Spatial and Verbal Components of the Act of
Recall, Canadian Journal of Psychology 22 (1968): 34968.]
The verbal task was to scan a sentence such

as The pencil is on the desk and decide
whether each word is or is not a noun. (The
correct response is No Yes No No No Yes.)
The visual task was to scan a block letter like
the F shown here, starting at the arrow and
moving clockwise, and decide whether each
corner is an outside corner. (The correct
response is Yes Yes Yes Yes No No Yes Yes
No Yes.)
Display 9.48 Population and housing units in a

sample of Florida counties. [Source: U.S.
Census Bureau, Census 2000.]
a. Make a stemplot of the variable persons

per housing unit (P/H) and another of
housing units per person (H/P).
interval for the mean number of persons
per housing unit.
c. Construct and interpret a 95% confidence
interval for the mean number of housing
units per person.
d. Turn the confidence interval in part c into
one for the mean number of persons per
housing unit by taking the reciprocals of
the endpoints of the interval. How does
this interval compare with the one in
part b?
e. From parts a, b, and d, which variable
do you think is better for measuring the
relationship between housing units and
their occupants?
Brooks also devised two ways to report, one

verbal and one visual. To report verbally, you
would simply say Yes or No out loud; to
report visually, you would point in sequence
to Yes or No on a piece of paper.
The theory predicted that when the task
and the report were of the same kind, they
would interfere with each other in memory
and slow the response time. If visual and
verbal tasks are handled independently, then
a visual task with a verbal report or a verbal
task with a visual report would be easier and
so would take less time than a verbal task
with a verbal report or a visual task with a
visual report.
Display 9.49 (on the next page) shows the
data from a version of the experiment run by
a psychology lab at Mount Holyoke College.
615
22-03-2009 21:20
Lesson
1 de 24
a. The means of the four combinations of

task and report are given in the last row
of the table. Discuss the meaning of the
pattern in light of the two predictions.
b. Boxplots showing the times for each of
the four treatment groups are shown
in Display 9.50. Are the distributions
suitable for constructing a confidence
interval for the mean, or would you
recommend a transformation first? Does
a reciprocal transformation make sense
here? (To decide, compute the reciprocal
of the visual task/visual report mean and
interpret it.)
Display 9.50 Boxplots of the times of each of the

four treatment groups.
Display 9.49 Data on visual/verbal report times

(in seconds) from the Mount Holyoke
experiment.
9.4
c. Using the procedure you think best,

estimate each of the four treatment means
in 90% confidence intervals. Is there any
evidence against the theory?
Inference for the Difference Between Two Means

Most scientific studies involve comparisons. For example, for purposes of
studying pesticide levels in a river, does it make a difference whether you take
water samples at mid-depth, near the bottom, or at the surface of the river? Do
special exercises help babies learn to walk sooner? Inference about comparisons
is more often used in scientific investigations than is inference about a single
parameter. In fact, almost all experiments are comparative in nature. The study
of inference for differences, then, is fundamental to statistical applications.
22-03-2009 21:20
Lesson
2 de 24
Warm-up: How Strong Is the Evidence?

Before you begin to develop formal methods of inference for differences between
means, you can develop your intuition and your ability to make informal
judgments based on comparing means by completing Activity 9.4a. It revisits the
Wolf River data of P1 on page 573.
The Strength of Evidence

The Wolf River in Tennessee flows past an abandoned site once used by
the pesticide industry for dumping wastes, including hexachlorobenzene
(chlordane), aldrin, and dieldren. These highly toxic organic compounds can
cause various cancers and birth defects. The standard method to test whether
these poisons are present in a river is to take samples at six-tenths depth, that
is, six-tenths of the way from the surface to the bottom. Unfortunately, there
are good reasons to worry that six-tenths is the wrong depth. The organic
compounds in question dont have the same density as water, and their
molecules tend to stick to particles of sediment. Both these facts suggest
that youd be likely to find higher concentrations near the bottom than near
mid-depth.
1. Display 9.51 (on the next page)
shows eight back-to-back stemplots.
The first plot shows actual data. It
compares concentrations of aldrin
(in nanograms per liter) for 20 water
samples taken from the Wolf River
downstream from the dump site.
Ten of the samples were taken at
mid-depth, and ten were taken at
the bottom. The other seven plots,
numbered 1 through 7, show hypothetical data.
Compare the plot of each hypothetical data set, 1 through 7, with the plot
of the actual data and evaluate the strength of the evidence of a difference
in concentration at the two depths. Is the evidence stronger in the actual
data, stronger in the hypothetical data, or about the same for both? State
your choice, and give the reason for your answer.
2. With everything else the same, is the evidence of a difference in
concentration stronger when the difference in the means is smaller or
larger? When the spread is smaller or larger? When the sample size is
smaller or larger?
617
22-03-2009 21:20
Lesson
3 de 24
Display 9.51 Concentrations of aldrin (in tenths of a nanogram per liter) for samples from
the Wolf River: actual data and seven hypothetical data sets. [Source: P. R. Jaffe,
F. L. Parker, and D. J. Wilson, Distribution of Toxic Substances in Rivers, Journal
of Environmental Engineering Division 108 (1982): 63949 (via Robert V. Hogg and
Johannes Ledolter, Engineering Statistics (New York: Macmillan, 1987).]
Most people can simply look at hypothetical data set 1 in Display 9.51 and
correctly conclude that the mean concentration at the bottom probably isnt
equal to the mean concentration at mid-depth. However, with the actual data
and the other hypothetical data sets, its not clear if there really is a difference.
And to estimate the size of any difference, you need to compute a confidence
interval.
A Confidence Interval for the Difference Between

Two Means
A confidence interval for the difference between two means has the standard form
you have seen before:
statistic (critical value) (standard deviation of statistic)
The procedure, given in the box, is also standard. Use this procedure when you
want to estimate the size of the difference between the mean of one population
and the mean of another population.
22-03-2009 21:20
Lesson
4 de 24
Confidence Interval for the Difference Between Two Means

(Two-Sample t-Interval)
Check conditions.
The two samples were randomly and independently selected from two
different populations. In the case of an experiment, the two treatments
were randomly assigned to the available experimental units.
The two samples look as if they came from normally distributed populations
or the sample sizes are large enough that the sampling distributions of the
sample means will be approximately normal. The 15/40 guideline (Section
9.3, pages 607608) can be applied to each sample or treatment group,
although it is a bit conservative.
In the case of sample surveys, the population size should be at least ten
times larger than the sample size for both samples.
Do computations.
A confidence interval for the difference between the means of two populations,
, is given by
where 1 and 2 are the respective means of the two samples, s1 and s2 are
the standard deviations, and n1 and n2 are the sample sizes. It is best to use a
calculator or statistics software to find this confidence interval because the
value of t* depends on a complicated calculation.
Give interpretation in context and linked to computations.
For a 95% confidence interval, for example, you are 95% confident that if you
knew the means of both populations, the difference between those means,
, would lie in the confidence interval.
For an experiment, the interpretation is like this: If all experimental units
could have been assigned each treatment, you are 95% confident that the
difference between the means of the two treatment groups would lie in the
Of course, when you interpret a confidence interval, you do it in context,
describing the two populations.
The standard error of the difference,
, comes directly from the rules
for random variables in the box in Section 6.1 on page 372. That is, if you have
two independent random variables with variances
then the variance of the sampling distribution of their difference is found by
adding the two variances,
. Then take the square root to get the standard
error,
619
22-03-2009 21:20
Lesson
5 de 24
Example: A Confidence Interval for the Aldrin Data

Construct a 95% confidence interval for the difference between the mean bottom
measurement of aldrin in the Wolf River and the mean mid-depth measurement.
The data from the first stemplot in Display 9.51 are repeated in Display 9.52.
Display 9.52 Actual aldrin concentrations, in nanograms per liter.
Name the procedure

and check conditions.
Solution
You will construct a confidence interval for the difference between two means
(also called a two-sample t-interval). You have two samples from two populations,
but you have no information about how randomly they were taken with respect
to either their location in the river or time. Further, you dont know if the samples
were taken independently of one another. If, for example, a pair of bottom and
mid-depth measurements were taken at the same time from the same spot, the
measurements would not be independent (and you would use the techniques of
the next section). Neither sample is skewed or has outliers, and each looks like it is
reasonable to assume that it came from a normally distributed population.
Here are the summary statistics:
Using a calculator (which does not give t*), you get a 95% confidence interval of
Do computations.
context and linked to
computations.
or (0.3009, 2.2809), where df = 16.10.

You are 95% confident that the true difference between the mean aldrin
, is in the interval
levels in the two depths in the Wolf River,
(0.3009, 2.2809). This interval overlaps 0 slightly, so there is insufficient
evidence to say that the true mean of the bottom measurements differs from the
true mean of the mid-depth measurements (but it is close). However, just as for
any confidence interval, unless the conditions are satisfied, there is no automatic
guarantee that the capture rate will be equal to the advertised confidence
level. Thus, the conclusion must be limited to the population of potential
measurements from which these samples were taken.
22-03-2009 21:20
Lesson
6 de 24
The two random

samples must be taken
independently.
Degrees of freedom is a
fraction?
More on Independence. The aldrin samples might not have been taken
independently. That would be the case if two measurements were taken at the
same time at the same spot in the river, one at mid-depth and one at the bottom,
and this procedure was repeated at nine other locations. When the measurements
of the two samples are taken in pairs (such as measurements from husband/wife
or left foot/right foot), they are not independent, and you should use the methods
of the next section.
More on Sample Sizes. The 15/40 guideline for inference for the difference
of two means is a bit conservative. In other words, the minimum sample sizes of
15 and 40 are higher than they need to be, in most cases. The reason is that, with
skewed populations, the sampling distribution of the difference of two means tends
to be more symmetric than the two separate sampling distributions of the sample
mean. Subtracting two sample means brings in the tails. The net effect is that the
sampling distribution of the difference of two means will look approximately
normal for smaller sample sizes than would be the case for the means themselves.
More on Degrees of Freedom. In the aldrin example, the calculator gave a value
for degrees of freedom, 16.10, that isnt a whole number. Whats that all about?
The matter is fairly complicated, as you might have guessed, and thats why it
is better to let the calculator find the confidence interval. The basic idea is that,
unlike in the one-sample case, the sampling distribution of the statistic for the
difference of two samples doesnt have a t-distribution. The exact distribution
isnt even known. However, it is known that the distribution is reasonably close
to a t-distribution if the right number of degrees of freedom is used. If you
are interested in the formula your calculator uses to compute these degrees of
freedom, see D14 on page 624.
In Which Order Do You Subtract? Usually, it doesnt matter whether you
compute a confidence interval for or for , so, if it seems reasonable,
subtract in the order that makes the difference positive and thus easier to
interpret. Just be sure to keep the SDs and sample sizes with their respective
means, and be sure your interpretation reflects the order of subtraction you used.
Example: Walking Babies 1

Well use the same scenario and experimental data as in E2 on page 576. The
effects of special exercises to speed up infant walking were isolated using three
different control groups. Display 9.53 shows the data again.
Display 9.53 Age (in months) when 23 male infants first walked without support.
[Source: Phillip R. Zelazo, Nancy Ann Zelazo, and Sarah Kolb, Walking
in the Newborn, Science 176 (1972): 31415 (via Larsen and Marx).]
Use these data to find a 95% confidence interval estimate of the difference
between mean walking times for the special exercises group and the exercise
control group if all babies could have had each treatment.
621
22-03-2009 21:20
Lesson
7 de 24
Name the procedure

and check conditions.
Solution
Two treatments were randomly assigned to the babies. You should plot the data
from the two groups to see if there is any reason to doubt that they resemble
samples from normally distributed populations. Display 9.54 shows that there is
some doubt, because each group has one relatively large value. Further, we were
told that some babies are 18 months old before they walk, so the population of
ages must be somewhat skewed right. However, the sampling distribution of the
difference of the means tends to be more symmetric than that of either group
mean by itself, so a confidence interval for the difference of two means should
work for this amount of skewness.
Display 9.54 Dot plots of two samples.

Do computations.
Special exercises: 1 = 10.125
n1 = 6
s1 = 1.447
2 = 11.375
n2 = 6
s2 = 1.896
Exercise control:
The two-sample t-interval function of a calculator gives df 9.35 and the

confidence interval (3.44, 0.94). [See Calculator Note 9F to learn how to calculate
a two-sample t-interval. You can start with the actual data or summary statistics.]
context.
You are 95% confident that if all the babies could have been in the special
exercises group and all the babies could have been in the exercise control group,
the difference in the mean age at which they would learn to walk is in the interval
from 3.44 months to 0.94 month. Because 0 is in the confidence interval, you
have no evidence that there would be any difference if you were able to give each
treatment to all the babies. However, it is important to note that any conclusion
is subject to doubt about the appropriateness of this procedure due to skewness
of the population of potential measurements, so it would be good to confirm this
finding with more data.
Display 9.55 shows printouts for the two-sample t-interval for the walking
babies experiment from three commonly used statistical software packages.
22-03-2009 21:20
Lesson
8 de 24
Data Desk
Fathom
Minitab
Display 9.55 Printouts of estimation from three statistical

software packages.
A Confidence Interval for the Difference Between

Two Means
D13. Here are summary statistics for the hypothetical data sets of Activity 9.4a.
Assume samples are random, observations are independent, and populations
are normally distributed.
i. Construct a 95% confidence interval for the difference between the two
means, bottom mid-depth, for each data set.
ii. Are the results consistent with the intuitive judgments you made about
the strength of the evidence in Activity 9.4a?
a. Hypothetical Data Set 1
Bottom:
s1 = 1.6
1 = 7.04 n1 = 10
Mid-depth:
2 = 5.05 n2 = 10
s2 = 1.1
623
22-03-2009 21:20
Lesson
9 de 24
b. Hypothetical Data Set 2
c. Hypothetical Data Set 3
d. Hypothetical Data Set 4
e.
Hypothetical Data Set 5
f.
Hypothetical Data Set 6
g. Hypothetical Data Set 7
D14. Your calculator or statistics software uses this formula to find df when
doing a two-sample t-procedure. You might be curious as to why we need
this complicated rule to calculate degrees of freedom. The simple answer
is that using t in place of z does not provide quite the right adjustment in
the two-sample case, as it does in the one-sample case, unless we make this
additional adjustment to the degrees of freedom. (The whole theoretical
story is complicated.)
a. Verify the value of df given in the aldrin and walking babies examples on
pages 620 and 621.
b. If n1 = n2, derive a simplified version of the formula for df.
c. If n1 = n2 and, in addition, s1 = s2, derive an even simpler rule for df.
A Significance Test for the Difference of Two Means

As you know, the other way to approach an inference problem is through
significance testing. Heres a summary of significance testing for the difference
between two means.
22-03-2009 21:20
Lesson
10 de 24
Components of a Test for the Difference Between Two Means

(the Two-Sample t-Test)
1. Name the test and check conditions. For a test of significance of the
difference between two means, the methods of this section require the
same conditions as those for a confidence interval:
Samples have been randomly and independently selected from two
different populations. In the case of an experiment, the two treatments
were randomly assigned to the available experimental units.
The two samples look as if they come from normally distributed
populations or the sample sizes are large enough that the sampling
distributions of the sample mean will be approximately normal. The
15/40 guideline can be applied to each sample, although it is a bit
conservative. As noted in the conditions for confidence intervals, you
can get by with smaller sample sizes when taking a difference.
In the case of sample surveys, the population size should be at least ten
times larger than the sample size for both samples.
2. State your hypotheses. The null hypothesis is, ordinarily, that the two
population means are equal. In symbols, H0: = or, in terms of the
difference between the means, H0: = 0. Here is the mean of the
first population and is the mean of the second. There are three forms of
the alternative or research hypothesis:
3. Compute the test statistic, find the P-value, and draw a sketch. Compute
the difference between the sample means (because the hypothesized mean
difference is zero), measured in estimated standard errors:
Use the two-sample t-test function of your calculator or statistics software

to get the P-value.
4. Write your conclusion, linked to your computations and in the context
of the problem. If you are using fixed-level testing, reject the null
hypothesis if your P-value is less than the level of significance, . If the
P-value is greater than or equal to , do not reject the null hypothesis.
(If you are not given a value of , you can assume that is 0.05.) Write a
conclusion that relates to the situation and includes an interpretation of
your P-value.
625
22-03-2009 21:20
Lesson
11 de 24
Example: A Significance Test for the Aldrin Data

Youve been given the responsibility to analyze the Wolf River data in Display 9.51
on page 618 to test whether the true mean aldrin concentrations at the bottom
and at mid-depth might differ. That is, you want to set up a test of significance for
the difference between two population means based on data from independent
random samples. Use = 0.10.
Check conditions.
State your hypotheses.
Solution
Conditions were checked in the example on page 620, and the conclusion here is
subject to those concerns.
In terms of the difference between two means, your null hypothesis is
is the mean aldrin concentration at the bottom of the Wolf River

where
and
is the mean concentration at six-tenths depth.
You are looking for a difference in either direction, so the alternative
hypothesis is two-sided:
Compute the test

and draw a sketch.
Here are the summary statistics for the aldrin concentrations:
The value of the test statistic is
From a calculator, you get an approximate df of 16.10 and a P-value of 0.1236.

[See Calculator Note 9G.]
Give conclusion in
context.
Conclude that, because the P-value for a two-sided test is greater than
= 0.10, you do not reject the null hypothesis. There is insufficient evidence
to claim that the mean aldrin concentration at the bottom of the Wolf River is
22-03-2009 21:20
Lesson
12 de 24
different from the mid-depth concentration. In other words, although it appears

from the stem-and-leaf plot that aldrin concentrations are greater near the
bottom, with these sample sizes the difference is not large enough to rule out
chance variation as a possible explanation.
Display 9.56 shows computer printouts from three commonly used statistical
software packages for the analysis in this example.
Data Desk
Fathom
Minitab
Display 9.56 Printouts of significance tests from three statistical

software packages.
For the aldrin data, it makes sense to use a two-sided alternative. (Researchers
wanted to know whether samples taken at mid-depth would give essentially
the same results as samples taken near the bottom.) For the walking babies
627
22-03-2009 21:20
Lesson
13 de 24
experiment, a one-sided alternative makes more sense because the researchers

wanted to test whether the special exercises helped babies walk sooner. They
werent interested in the possibility that the special exercises would slow down
the babies.
Example: Walking Babies 2

Test the null hypothesis that the mean age at first unaided steps is the same for
the special exercises and exercise control groups against the alternative hypothesis
that the mean age is smaller for the special exercises group. Use = 0.05.
Check conditions.
Solution
The conditions were checked in the example Walking Babies 1 on page 621.
The null hypothesis is
or, equivalently,
where
is the mean age that the babies in the experiment would first walk
if they all could have been given the special exercises and
is the mean age
that the babies would first walk if all babies in the experiment could have received
the exercise control treatment. (You can also use the symbols and as long as
you define the symbols.)
For this one-sided test, the alternative hypothesis is
or, equivalently,
Here are the summary statistics:
Special exercises:
Exercise control:
Compute the test

and draw a sketch.
Give conclusion linked

to computations and in
context.
From a calculator, df is 9.35 and the P-value is 0.115.

Because the P-value, 0.115, is greater than = 0.05, you do not reject the null
hypothesis. The evidence isnt convincing that babies who are given the special
exercises walk at an earlier age than babies who are given the control exercise. In
other words, if the researchers had been able to give both the special exercises and
the control exercises to all the babies in the experiment, they would have been
reasonably likely to get groups as different as these even if the special exercises
made no difference.
22-03-2009 21:20
Lesson
14 de 24
Increasing Power in the Two-Sample t-Test

Power is the ability
to reject the null
hypothesis.
As always, the best way to get more power to reject a false null hypothesis is to
increase the sample sizes. However, sometimes you have the resources to get, say,
only 40 measurements total between the two samples. How should you divide
them up20 in each sampleor some other way?
Getting the Most Power Out of Your Two-Sample t-Test
The best way to get more power is to have larger sample sizes.
If you have reason to believe that the population standard deviations are
about equal, make the sample sizes equal.
If you have reason to believe that one populations standard deviation is
larger than the others, allocate your resources so that you take a larger sample
from the population with the larger standard deviation. (Choose the sample
sizes to be proportional to the estimated standard deviations.)
Follow these same rules to get the smallest margin of error for a confidence
interval.
Why Not Two Separate Confidence Intervals? Significance tests for the
difference of two means take a little getting used to. It is natural to ask why you
cant simply compute two separate confidence intervals, one for and one for ,
and check to see if they overlap. This method will tell you if there are any values
that are plausible means for both populations. If so, then you wouldnt reject the
null hypothesis that the difference in the means is 0. The difficulty is that the
method is too conservative, meaning that you wont reject a false null hypothesis
often enough. In other words, you have sacrificed power.
However, you can use this rule: If you construct two separate confidence
intervals for the means of two populations (at confidence level 1 ) and they
dont overlap, you are safe in rejecting the null hypothesis that the means are equal
at significance level . If the intervals overlap, you can come to no conclusion.
A Special Case: Pooling When
Suppose you are taking two independent samples from the same population for
the purpose of comparing means. This happens, for example, when you randomly
divide available experimental units into two groups for the purpose of comparing
two treatments. If the true treatment means really do not differ (the usual null
hypothesis), then the true variances of the sample measurements should not differ
either. You then have two sample variances (that probably will differ) to estimate
the single population variance. One way to combine these two estimates is simply
to average the sample variances and use this average to estimate the population
variance. (This average should be a weighted average if the sample sizes are
not equal.) This process, called pooling, gives you another way to calculate a
confidence interval or a test statistic based on t.
Your calculator and statistics software give you the choice of pooled or
unpooled when doing two-sample t-procedures. Pooled should be used only
when you have a good reason to believe that the population standard deviations
629
22-03-2009 21:20
Lesson
15 de 24
Almost always select the

unpooled procedure.
are equal. Even if you know this to be true, the two-sample (unpooled) procedure
discussed in this chapter works almost as well as the pooled procedure, especially
if the sample sizes are equal. The only situation in which the pooled procedure has
a definite advantage is when the populations have equal standard deviations but
your sample sizes are unequal. Therefore, we wont cover the pooled procedure in
this text. Unless you encounter a problem that specifically tells you to assume that
and
are equal, choose the unpooled procedure.
You should know, however, that there are extensions of the two-sample
procedures that allow comparisons among more than two means. The most
common of these procedures (analysis of variance, or ANOVA) is a generalization
of the pooled procedure for two samples, so it is good to know when pooling
works and when it doesnt.

D15. To pool or not to pool? Here are three different situations for two
independent samples.
I. n1 = 5, n2 = 25,
= 10,
= 10
II. n1 = 10, n2 = 10,
= 10,
= 10
= 10,
=1
III. n1 = 5, n2 = 25,
For one of these situations, pooling is wrong. For a second situation,
pooling, though not wrong, is not likely to offer an advantage over the
unpooled approach. Finally, in the third situation, pooling not only is
appropriate but also will likely give narrower intervals than the unpooled
approach. Which is which?
Summary 9.4: Inference for the Difference Between

Two Means
In this section, you learned how to construct a confidence interval for
and to perform a significance test of = . Whether you are doing a confidence
interval or a significance test, the conditions to check are the same:
Samples have been randomly and independently selected from two different
populations (or two treatments were randomly assigned to the available
experimental units).
The two samples look as if they came from normally distributed populations
or the sample sizes are large enough. (The 15/40 guideline can be applied to
each sample, although it is a bit conservative.)
In the case of sample surveys, the population size should be at least ten times
larger than the sample size for both samples.
A confidence interval for the difference between the means of two
populations, , has the form
22-03-2009 21:20
Lesson
16 de 24
The test statistic for a test of significance is
This statistic is called t, but it doesnt have exactly a t-distribution. Fortunately, it

is reasonably accurate to proceed using the t-distribution with df approximated by
a rather complicated rule. A calculator or statistics software is the most accurate
method of computing a confidence interval and of finding a P-value. Always
select the unpooled t-procedure unless you are asked to do otherwise.
Practice
A Confidence Interval for the Difference
Between Two Means
P27. As you read in E1 on page 575, some
students recruited 30 volunteers to attempt
to walk the length of a football field while
walk to the opposite goal line, a distance
of 100 yards. The dot plot and summary
statistics in Display 9.57 show the distance
at which the volunteer crossed a sideline
of the field and whether the volunteer was
left-handed or right-handed.
a. Assuming that these volunteers can be
considered independent random samples
from the populations of left-handed
volunteers and right-handed volunteers,
are the conditions met for a confidence
interval for the difference of two means?
construct a 95% confidence interval,
using your calculator or statistics
software.
c. You are 95% confident that something is
in the interval you constructed in part b.
Describe exactly what that something is.
d. Do you have statistically significant
evidence that left- and right-handed
volunteers differ in the mean number
of yards they can walk before crossing
a sideline?
Display 9.57 Yard-line data, separated by

handedness. [Source: Andrea Axtell, When
Direction Vanishes: Walking Straight or in Circles,
STATS 39 (2004): 1821.]
P28. The data in Display 9.58 (on the next page)

are from a survey done in an introductory
statistics class during the first week of a term
(see P6 on page 574). You can assume that
this class is a random sample taken from
all students in this course. The data are the
number of hours of study per week, classified
by gender.
Estimate the difference in mean study hours
per week for all the males and females taking
this course, with confidence level 0.90.
Interpret this interval.
631
22-03-2009 21:20
Lesson
17 de 24

P31. The actual aldrin data appear in Display
9.59. Your boss sends you out to get ten more
measurements. Would you suggest getting
five additional measurements at each depth,
or something different? Explain.
Display 9.58 Two stemplots of study hours per

week, classified by gender.
A Significance Test for the Difference

of Two Means
P29. Refer to P28. Test to see if there is evidence
of a statistically significant difference in
mean weekly study hours for males and
females at the 5% level of significance.
P30. Refer to the walking babies data in Display
9.53 on page 621.
a. Test to see if the special exercises group
produces a statistically significant
decrease in mean walking age over the
weekly report group. Use 0.05.
b. Test to see if the exercise control group
produces a statistically significant
decrease in mean walking age over the
weekly report group. Use 0.05.
c. State a conclusion about comparisons
among the three groups.
Display 9.59 Actual aldrin concentrations.
P32. Refer to the example Who Is Hotter?

on page 566. From that example, the 95%
confidence intervals for the mean body
temperature of males, (97.48, 98.28), and the
mean body temperature of females, (98.14,
98.90), overlap. For example, if you look at
the two genders separately, 98.2 is a plausible
mean temperature for both (but is near
opposite ends of the confidence intervals).
a. Construct and interpret a 95% confidence
interval for the difference in mean
temperatures. You can assume that the
conditions are met.
b. What can you conclude about whether
males and females have the same mean
body temperature?
c. When using the confidence intervals to test
the hypothesis that there is no difference
in population means, which procedure
had more power: comparing two separate
confidence intervals, or constructing one
confidence interval for the difference?
Exercises
E53. Kelly randomly assigned eight golden
hamsters to be raised in long days or short
days. She then measured the concentrations
of an enzyme in their brains. (Refer to
page 244 for more about Kellys hamster
experiment.) The resulting measurements of

enzyme concentrations (in milligrams per
100 milliliters) for the eight hamsters (shown
in Display 9.60) were
Short days: 12.50011.625 18.275 13.225
Long days: 6.62510.375 9.900 8.800
22-03-2009 21:20
Lesson
18 de 24
Display 9.60 A dot plot of Kellys hamster data.
a. Are the conditions met for inference

about the difference of two means?
c. You are 95% confident that something is
in the interval you constructed in part b.
Describe exactly what that something is.
d. Does Kelly have statistically significant
evidence to back her claim that
the observed difference in enzyme
concentrations between the two groups
of hamsters is due to the difference in the
hours of daylight?
E54. Suppose Kellys means and the shapes of the
distributions are the same as in E53 but the
enzyme concentrations are more variable, as
Short days: 9.500 8.625 27.275 10.225
Long days: 4.625 12.375 11.900 6.800
Display 9.61 Dot plots of altered hamster data.
a. With the altered data, are the conditions

met for inference about the difference of
two means?
construct and interpret a 95% confidence
interval.
c. If the values had been this variable,
would Kelly have had statistically
significant evidence to back her claim
that the observed difference in enzyme
concentrations between the two groups

of hamsters is due to the difference in the
hours of daylight?
E55. The inflammation caused by osteoarthritis
of the knee can be very painful and can
inhibit movement. Leech saliva contains
anti-inflammatory substances. To study the
therapeutic effect of attaching four to six
leeches to the knee for about 70 minutes,
51 volunteers were randomly assigned to
receive either the leech treatment or a topical
gel, diclofenac. [Source: Andreas Michalsen et al.,
Effectiveness of Leech Therapy in Osteoarthritis of the Knee:
A Randomized, Controlled Trial, Annals of Internal Medicine
139, no. 9 (November 4, 2003): 72430.]
a. This summary table gives the results of a

pretreatment measure of the amount of
pain reported by the two groups, before
beginning therapy. A higher score means
more pain. The researchers hoped that
the randomization would result in two
comparable groups with respect to this
variable. Construct and interpret a 95%
confidence interval for the difference in
means. Is there statistically significant
evidence that the randomization failed to
yield groups with comparable means?
Leech:
= 53.0 s1 = 13.7
n1 = 24
Topical gel:
= 51.5 s2 = 16.8
n2 = 27
b. Because a high body mass can stress

the knee, the researchers hoped that
the randomization would result in two
comparable groups with respect to this
variable as well. This summary table gives
the body mass index of the two treatment
groups, before beginning therapy.
Construct and interpret a 95% confidence
interval for the difference in means. Is
there statistically signicant evidence that
the randomization failed to yield groups
with comparable means?
Leech:
= 27.6 s1= 3.7
n1= 24
Topical gel:
= 27.1 s2 = 3.7
n2 = 27
633
22-03-2009 21:20
Lesson
19 de 24
E56. Refer to E55. So far the researchers have

been lucky, and the two treatment groups
have comparable means. However, the
researchers compared the two treatment
groups on 12 initial variables. Another
variable was an initial measure of stiffness.
a. This summary table gives summary
statistics for the stiffness scores of the
two groups, before beginning therapy.
A higher score means more stiffness.
Construct and interpret a 95% confidence
interval for the difference in means. Is
there statistically significant evidence that
with comparable means?
Leech:
= 63.3 s1 = 19.0
n1 = 24
Topical gel:
= 48.6 s2 = 22.2
n2 = 27
b. If you can consider the initial variables

to be independent, how many of the 12
variables would you expect to show a
statistically significant difference between
two randomly assigned groups?
c. If you can consider the initial variables to
be independent, what is the probability
that none of the 12 variables shows a
two randomly assigned groups?
d. At the end of the experiment, the leech
treatment was shown to be significantly
better than the topical gel. Does your
result in part a tend to invalidate this
conclusion?
E57. In E19 and E20 on page 598, students were
given pieces of paper with five vertical or
horizontal line segments and asked to mark
the midpoint of each segment. Then, using
a ruler, each student measured how far each
mark was from the real midpoint of each
segment. He or she then averaged the five
errors. (We arent told if the assignments of
the two treatments were random, but assume
they were because this experiment was done
in a statistics class.)
The results are given in Display 9.62.
Display 9.62 Average error in locating midpoints of

horizontal and vertical line segments.
[Source: Gretchen Davis, Beware of Measurement
Bias, CMC ComMuniCator 22, no. 4 (June 1998):
5456.]
a. Before beginning this experiment, the

instructor believed that students tend to
make larger errors (in absolute value)
when the line segments are vertical than
when the line segments are horizontal.
Make the necessary changes to the data in
the table so you can test the instructors
belief.
b. Conduct a significance test of the
instructors claim, showing all four steps.
E58. It has been speculated that the mean amount
of calcium in the blood is higher in women
than in men. A retrospective review of the
medical charts of subjects tested in a certain
city produced the summary statistics on
calcium (in millimoles per liter) shown in
Display 9.63.
a. Discuss whether the conditions are met
for doing a two-sample t-test of the
difference in the means.
22-03-2009 21:20
Lesson
20 de 24
b. Find and interpret the P-value for this

test.
b. Regardless of your answer to part a, carry

out an appropriate test. Carefully write
your conclusion in the context of the
study.
c. What are your main concerns about the
conclusion you reached in part b? What
advice would you give the experimenter?
Display 9.63 Calcium in the blood of women and

men, in millimoles per liter. [Source: JSE
Data Archive, www.amstat.org.]
E59. In warm and humid parts of the world, a

constant battle is waged against termites.
Scientists have discovered that certain
tree resins are deadly to termites, and thus
the trees producing these resins become a
valuable crop. In one experiment typical
of the type used to test the protective power
of a resin, two doses of resin (5 mg and
10 mg) were dissolved in a solvent and
placed on filter paper. Eight dishes were
prepared with filter paper at dose level 5 mg
and eight with filter paper at dose level
10 mg. Twenty-five termites were then
placed in each dish to feed on the filter
paper. At the end of 15 days, the number of
surviving termites was counted. The results
are shown in Display 9.64. In parts ac,
youll determine if there is a statistically
significant difference in the mean number of
survivors for the two doses.
a. Are the conditions for a two-sample t-test
met here?
Display 9.64 Number of termites, out of 25,

surviving after being placed in a dish
with 5 mg or 10 mg of a resin. [Source:
lib.stat.cmu.edu.]
635
22-03-2009 21:20
Lesson
21 de 24
E60. The fact that your food usually tastes good is

no accident. Food manufacturers regularly
check taste and texture by recruiting
taste-test panels to measure palatability.
A standard method is to form a panel with
50 persons25 men and 25 womento
do the tasting. In one such experiment,
simplified here, coarse versus fine texture
was compared. Panel members were
assigned randomly to the treatment groups
as they were recruited. There were 16 panels
of 50 consumers each. The variables in
Display 9.65 are these:
Total palatability score for the panel of 50: A
higher total score means the food was rated
more palatable by the panel.
Texture: 0 (coarse), 1 (fine)
E61. Is there suffcient evidence from the random

samples of heart rates for men and women
under normal conditions in Display 9.66
to say that the mean heart rates differ for
the two groups? Analyze the data in two
different ways (confidence interval and
test of significance) before coming to your
conclusion, and then state your conclusion
carefully.
Display 9.66 Heart rates. [Source: Journal of Statistics

Education Data Archive, www.amstat.org,
April 15, 2002.]
E62. The data on mercury content of fish in

the lakes of Maine in E48 on page 613 are
augmented in Display 9.67 by two other
variables: whether the lake is formed behind
a dam and the oxygen content of the water.
Display 9.65 Food ratings. [Source: E. Street and M. G.
Carroll, Preliminary Evaluation of a Food
Product, in Statistics: A Guide to the Unknown,
Judith M. Tanur, ed. (San Francisco: Holden-Day,
1972), pp. 22038.]
Is there a statistically significant difference

in mean palatability score between the
two texture levels? Show all four steps in
a significance test when answering this
question.
22-03-2009 21:20
Lesson
22 de 24
Display 9.67 Mercury content of fish in Maine lakes.

[Source: R. Peck, L. Haugh, and A. Goodman,
Statistical Case Studies, 1998, ASA-SIAM, 114.]
a. Some environmentalists claim that dams

are a cause of high mercury levels. Is
there suffcient evidence to conclude
that the mean mercury level of the lakes
behind dams is greater than the mean
mercury level of the lakes without dams?
b. Type 1 lakes are oligotrophic (balanced
between decaying vegetation and living
organisms), type 2 lakes are eutrophic
(high decay rate and little oxygen), and
type 3 lakes are mesotrophic (between the
other two states). Estimate the difference
in mean mercury level for type 1 and
type 2 lakes?
c. Is there sufficient evidence to conclude
that the mean mercury level differs for
lakes of type 2 compared to lakes of
type 3?
E63. Refer to Kellys hamster experiment in E53

on page 632.
a. What kind of error could Kelly be making
in a significance test of the difference in
mean enzyme concentration?
b. What should Kelly do in a similar
experiment if she wants to reduce the
probability of an error of this type?
c. What about this situation gave Kelly
power to reject the null hypothesis even
with such small treatment group sizes?
E64. Refer to the taste test experiment in E60.
a. What kind of error might investigators be
making in this significance test?
b. What should the investigators do in a
similar experiment if they want to reduce
the probability of this type of error?
E65. As you read in E1 and P27, some students
recruited 30 volunteers to attempt to
walk the length of a football field while
walk to the opposite goal line, a distance of
100 yards. Display 9.68 shows the distance at
which the volunteers crossed a sideline of the
field and whether they were male or female.
Display 9.68 Boxplot and stemplot of yard-line data,

classified by gender.
a. Is this study more like an experiment or

more like a survey?
637
22-03-2009 21:20
Lesson
23 de 24
b. Conduct a test of significance for the

difference in the mean number of yards
walked for males and females. Use the
summary statistics in the printout in
Display 9.69.
Display 9.69 Summary statistics of yard-line data for

males and females.
c. Display 9.70 shows the relationship

between the height of the volunteer, in
inches, and the number of yards he or she
walked before crossing a sideline. What
lurking variable can help explain your
result in part b?

hospital stays for Insurer A. [Source:
R. Peck, L. Haugh, and A. Goodman, Statistical
Case Studies, 1998, ASA-SIAM, 4564.]
An independent random sample of 396 cases

from Insurer B gave the results on length of
stay summarized in Display 9.72.

hospital stays for Insurer B.
Display 9.70 Scatterplot of yard-line data,

categorized by gender.
E66. As you saw in E43 on page 611, a study to

compare two insurance companies on length
of stay (LOS) for pediatric asthma patients
randomly sampled 393 cases from Insurer A.
Summary statistics and a histogram for the
data are shown in Display 9.71.
a. Estimate the difference between the mean

lengths of stay for the two insurance
companies in a 95% confidence interval.
Is there evidence of a difference between
the population means for the two
companies?
b. Many other variables could contribute to
the difference in mean length of stay for
the two insurers. One is the number of
full-time staff per bed. The sample means
for this variable are 4.63 for Insurer A and
6.13 for Insurer B. The respective sample
standard deviations are 1.70 and 2.40.
Is this difference in means statistically
significant? If so, provide a practical
explanation of how this difference might
be related to the difference in mean
length of stay.
c. Another contributing variable is the
percentage of private hospitals versus
public hospitals among the patients in
each insurers sample. For Insurer A,
93.6% of the sampled patients are in
22-03-2009 21:20
Lesson
24 de 24
private hospitals; for Insurer B, that

percentage is 73.0%. Is this a statistically
significant difference? If so, how might
this difference be related to the difference
in mean length of stay?
E67. There is much controversy about whether
coaching programs improve scores on SAT
exams by more than a minimal amount.
Although this exercise wont settle the
argument, it does present a practical use of a
test of significance.
The College Board reports that coaching
programs improve SAT mathematics scores
by about 25 points on average. A sample of
50 students in a control group who took
the math portion of the SAT as juniors
and then again as seniors had an average
gain of 13 points (reported to be about the
national average gain without coaching),
with standard deviation of about 30 points.
A sample of 9 students who were coached
between their two exams had an average
gain of 60 points, with a standard deviation
of 42. [Source: Jack Kaplan, An SAT Coaching Program
That Works, Chance 15, no. 1 (2002): 1217.]
a. Is there evidence that the difference in

mean point gain between the coached
group and the control group exceeds
the 25 points that is to be expected
(according to the College Board)?
b. Another sample of 12 students from a
coaching program had an average gain
of 73 points, with a standard deviation of
42. Is there evidence that the difference
in mean point gain here exceeds the 25
points that is to be expected?
c. What further questions would you like to
ask about this study?
E68. An important factor in the performance of
a pharmaceutical product is how fast the
product dissolves in vivo (in the body). This
is measured by a dissolution test, which
yields the percentage of the label strength
(%LS) released after certain elapsed times.
Laboratory tests of this type are conducted
in vessels that simulate the action of the
stomach. Display 9.73 shows %LS at certain
time intervals for an analgesic (painkiller)

tested in laboratories in New Jersey and
Puerto Rico. Time is measured in minutes.
Display 9.73 Percentage of label strength of

analgesic for dissolution tests in two
laboratories. [Source: R. Peck, L. Haugh, and
A. Goodman, Statistical Case Studies, 1998,
ASA-SIAM, 3744.]
a. Use a 90% confidence interval to

estimate the difference in mean %LS at
40 minutes for New Jersey compared to
Puerto Rico.
b. Use a 90% confidence interval to
estimate the difference in mean %LS at
60 minutes for New Jersey compared to
Puerto Rico.
c. Good manufacturing practices call for
equivalence limits of 15 percentage
points for dissolution percentages
below 90% and 7 percentage points for
dissolution percentages above 90%. That
is, if the 90% confidence interval for the
mean difference is within the equivalence
limits (within an interval from 15% to
+ 15% for the lower percentages), then
the two sets of results are accepted as
equivalent. Will the results in parts a and
b be accepted as equivalent at 40 minutes?
At 60 minutes?
639
22-03-2009 21:20
Lesson
1 de 23
9.5
Paired Comparisons
Now that youve seen the methods for comparing means, its time to put them
to work. In this section, youll see confidence intervals and significance tests in
action. Keep in mind that a t-test is no smarter than a chainsaw. Neither has any
brains of its own. A chainsaw cant tell whether its cutting an old dead tree into
firewood or turning a valuable antique table into scrapwood. A t-test is every bit
as oblivious. The difference between thoughtful and careless use is up to you, the
operator. This final section looks at four issues on which you need to be clear in
order to use your statistical tools with care.
Do you really have two independent samples, or do you have only one sample
of paired data?
What if shapes arent normal?
Is it meaningful to compare means?
Does your inference have the chance it needs?
Two Independent Samples, or Paired Observations?

One of the recurring themes in statistics is how important it is to pay attention
to data production. This theme was developed in Chapter 4, where the entire
emphasis was on designing studies. Now the theme returns in the context of
inference for means, where one of the key questions is Do I have paired data or
independent samples? Activity 9.5a will help you see how pairing changes the
standard error.
Hand Spans
What youll need: a ruler marked in millimeters
Detective Sherlock Holmes amazed a man by relating obvious facts about
him, such as that he had at some time done manual labor: How did you know,
for example, that I did manual labour? It is as true as gospel, for I began as a
ships carpenter. Sherlock replied, Your hands, my dear sir. Your right hand
is quite a size larger than your left. You
have worked with it, and the muscles are
more developed. [Source: Sir Arthur Conan Doyle,
The Adventures of Sherlock Holmes, ed. Richard Lancelyn
Green, Oxford World Classics (Oxford and New York,
1988).] In fact, peoples right hands tend to
be bigger than their left, even if they are

left-handed and even if they havent done
manual labor, but the difference is small.
To detect it, you will have to design your
study carefully.
(continued)
22-03-2009 21:21
Lesson
2 de 23
1. Measure your left and right hand spans, in millimeters. (An easy way to do
this is to spread your hand as wide as possible, place it directly on a ruler,
and get the distance between the end of your little finger and the end of
your thumb.) Record the data for each student in your class in a table with
a column for the left hand span and another column for the right. There
should be one row for each person.
2. For each row in the table, calculate the difference, right left. Find the
mean of the differences and the standard error of the mean difference using
the formula sd =
3. Now make a new table, but this time randomize the order of the right hand
spans so that peoples left hand spans are no longer matched with their
right. Then repeat step 2 with the new table.
4. Finally, treat the left hand spans and right hand spans as independent
samples. Calculate the difference between the two sample means and the
standard error of that difference using the formula
5. Compare the standard errors from steps 2, 3, and 4. Which is smallest?

Which two are most nearly the same size?
6. Suppose you make scatterplots of the data in your two tables from steps
1 and 3, with the data for the left hand spans on the horizontal axis.
Which would you expect to have the higher correlation? Why? Make the
scatterplots, and calculate the correlations to check your answer.
The lesson of Activity 9.5a is this: Paired observations can greatly reduce
variation over independent samples and produce a much more powerful test
and a more precise confidence interval estimate of the true mean difference. The
reduction in variation is greatest when the underlying measurements vary greatly
from pair to pair but the differences do not.
In Activity 4.4a, you compared sitting and standing pulse rates using three
different designs. In the completely randomized design, you randomly selected
half of the class to sit and the other half to stand. These data can be analyzed using
the techniques for two independent samples, as in Section 9.4. In the matched
pairs design, you matched pairs of students on a preliminary measure of pulse
rate. Then you randomly assigned sitting to one student in each pair and standing
to the other. In the repeated measures design, you had each person sit and stand,
with the order randomly assigned. The data from the last two designs should not
be analyzed using the techniques for two independent samples. In this section,
you will learn how to analyze these data using the differences in pulse rate for
each pair. Display 9.74 (on the next page) shows the data for pulse rates, in beats
per minute, from a class that worked through this activity.
641
22-03-2009 21:21
Lesson
3 de 23
Completely
Randomized
Design
Matched Pairs
Design
Repeated Measures
Design
Summary Statistics
Display 9.74 Pulse measurements, in beats per minute, from a

class experiment.
22-03-2009 21:21
Lesson
4 de 23
Completely Randomized Design (Two Independent Samples)

You will now work through the analysis of each of the three sets of experimental
results, beginning with the completely randomized design. Display 9.75 shows the
relationship between the sitting and standing pulse rates for the arbitrary pairing
of values from the two independent samples in the completely randomized design
table in Display 9.74. The correlation is near 0, as it should be, because these are
independent measurements taken in arbitrary order.
Display 9.75 Scatterplot of sitting versus standing pulse rates,

in beats per minute, in a completely randomized
design.
To analyze the data from this completely randomized design, use the methods
of Section 9.4. A 95% confidence interval (df 24.31) for the difference between
the mean pulse rates for sitting and standing is given by
or 8.96 <
< 14.67.This interval overlaps 0, so you cant conclude
that one of these treatments would produce a higher mean than the other if every
subject were given both treatments.
Matched Pairs Design

Display 9.76 shows the relationship between the sitting and standing pulse rates
for the matched pairs design.
Display 9.76 Scatterplot of sitting versus standing pulse rates, in

beats per minute, in a matched pairs design.
643
22-03-2009 21:21
Lesson
5 de 23
This plot gives a hint of a linear trend (r = 0.48), as it should, because these are
dependent measurements based on pairing people with similar resting pulse rates.
The matched pairs design has dependent observations within a pair, so the
two-sample t-procedure is not a valid option. You can, however, look at the
differences between the standing and sitting pulse rates for each pair and estimate
the mean difference with a one-sample procedure. Observe in Display 9.74 that
the difference between the sample means, 77.57 73.86, is the same as the mean
of the differences, 3.71, so the latter is a legitimate estimator of the true difference
between the population means. Thus, the two sets of measurements are reduced
to one set of differences. The summary statistics for the observed differences are
Mean: d = 3.71
Standard deviation: sd = 12.38
With 13 degrees of freedom, the critical value, t*, for a 95% confidence interval is
2.160. The confidence interval estimate of the true mean difference is
t*
or
3.71
2.160
which yields the interval ( 3.44, 10.86). Any value of the true mean difference
between standing and sitting pulse rates in this interval could have produced the
observed mean difference as a reasonably likely outcome. This interval includes 0,
so there is not suffcient evidence to say that the standing mean would differ from
the sitting mean if all subjects were measured under both conditions. Although
this interval overlaps 0, 0 is proportionally closer to the endpoint than it is in the
confidence interval for the completely randomized design.
Repeated Measures Design
The scatterplot for the data from the repeated measures design (Display 9.77)
shows a strong linear trend (correlation 0.93). These are paired measurements
from the same person and should be highly correlated.
Display 9.77 Scatterplot of sitting versus standing pulse rates, in
beats per minute, in a repeated measures design.

A 95% confidence interval for the mean difference (df = 27 ) is
d
t*
or
8.36
2.052
22-03-2009 21:21
Lesson
6 de 23
or (6.31, 10.41). This estimate of the mean does not overlap 0, so the evidence
supports the conclusion that the mean standing pulse rate is higher than the mean
sitting pulse rate.
Checking Normality
You might be wondering why we did not check conditions by looking at a plot of
the data. We will do that now. For the analyses of differences, it is the differences,
not the two original samples, that must come from an approximately normal
distribution. Display 9.78 provides boxplots of these differences for both the
matched pairs design and the repeated measures design. They show that there is
no obvious reason to be concerned about non-normality in either case.
Display 9.78 Boxplots of differences in pulse rates.
A Confidence Interval for Paired Observations

The box summarizes how to construct a confidence interval for the mean difference.
A Confidence Interval for the Mean Difference from
Paired Observations
Check conditions.
Randomness. In the case of a survey, you must have a random sample from
one population, where a unit might consist of, say, a pair of twins or the
same persons two feet. In the case of an experiment, the treatments must
have been randomly assigned within each pair. If the same subject is assigned
both treatments, the treatments must have been assigned in random order.
Normality. The differences must look like its reasonable to assume that
they came from a normally distributed population or the sample size
must be large enough that the sampling distribution of the sample mean
difference is approximately normal. The 15/40 guideline can be applied to
the differences.
Sample Size. In the case of a sample survey, the size of the population of
differences should be at least ten times as large as the sample size.
Do computations.
A confidence interval for the mean difference, d, is given by
d
t*
(continued)
645
22-03-2009 21:21
Lesson
7 de 23
(continued)
where n is the sample size, d is the mean of the differences between pairs of
measurements, sd is the sample standard deviation, and t* depends on the
confidence level desired and the degrees of freedom, df = n 1.
Give interpretation in context.
An interpretation is of this form: I am 95% confident that the mean of the
population of differences, d, is in this confidence interval.
Of course, when you interpret a condfience interval, you will do it in
context, describing the population you are talking about.
A Significance Test for Paired Observations
You could also decide whether standing increases the mean pulse rate by doing a
test of significance. The next example illustrates the components of such a test.
Example:Testing the Mean Difference for the Repeated Measures Design

Use the data in the repeated measures design of Display 9.74 to test the research
hypothesis that standing increases the mean pulse rate.
Check conditions.
Solution
Because the order in which each subject receives the two treatments is
randomized, you can treat this as a random assignment of treatments to subjects.
Although Display 9.78 shows that the distribution of differences is not quite
symmetric, there is no reason to rule out the normal distribution as a possible
model for producing these differences.
The hypotheses are
H0: d = 0 versus Ha: d > 0
Compute the test

statistic, findnd the P-value,
and draw a sketch.
Here d is the theoretical mean difference between standing and sitting pulse rates
for this group of subjects if each subject could have received each treatment in
both orders.
t=
With 27 degrees of freedom, the P-value for this large t-statistic is essentially 0.
The very small P-value indicates that there is sufficient evidence to conclude
that the mean difference in pulse rates is positive. This implies that the mean
pulse rate is higher for persons standing than it is for those same persons sitting.
The result applies only to the people (subjects) in this experiment and cannot be
generalized to other people based on these data alone.
22-03-2009 21:21
Lesson
8 de 23
The procedure is summarized in this box.

Components of a Test of the Mean Difference Based
on Paired Observations
1. Name the test and check conditions. For a test of significance of a mean
difference based on paired observations, the method requires the same
conditions as those for a confidence interval. You must check three
conditions:
Randomness. In the case of a survey, you must have a random sample
from one population, where a unit might consist of, say, a pair of
twins or the same persons two feet. In the case of an experiment, the
treatments must have been randomly assigned within each pair. If the
same subject is given both treatments, the treatments must have been
assigned in random order.
Normality. The differences must look like its reasonable to assume
that they came from a normally distributed population or the sample
size must be large enough that the sampling distribution of the sample
mean difference is approximately normal. The 15/40 guideline can be
applied to the differences.
Sample Size. In the case of sample surveys, the size of the population
of differences should be at least ten times as large as the number of
differences in the sample.
2. State the hypotheses. The null hypothesis is, ordinarily, that in the entire
population the mean of the differences is 0. In symbols, H0: d0 = 0
There are three forms of the alternative or research hypothesis:
H0: a
Ha: d < 0
Ha: d > 0
3. Compute the test statistic, find the P-value, and draw a sketch. Compute
the difference between the mean difference, d, from the sample and the
hypothesized difference d0, and then divide by the estimated standard
error:
The P-value is based on n 1 degrees of freedom, where n is the number

of differences.
4. Write your conclusion linked to your computations and in the context of
the problem. If you are using fixed-level testing, reject the null hypothesis
if your P-value is less than the level of significance, . If the P-value is
greater than or equal to , do not reject the null hypothesis. (If you are not
given a value of , you can assume that is 0.05.) Write a conclusion that
relates to the situation and includes an interpretation of your P-value.
647
22-03-2009 21:21
Lesson
9 de 23
Two Independent Samples, or Paired Observations?

D16. Based on what you recall from Chapter 4 and what you saw in the analysis
of the pulse rate data in this section, discuss the relative merits of completely
randomized, matched pairs, and repeated measures designs. Under what
conditions will the analysis of differences between pairs pay bigger dividends
than the analysis of differences of independent means?
What If the Differences Dont Appear Normal?

You discovered in Section 9.3 that the true capture rate of a nominal 95%
confidence interval is less than 95% if the population is skewed and the sample
size is not large enough for the sampling distribution of to be approximately
normal. The same is true of the confidence interval for the mean difference. If the
sample differences appear to come from a skewed distribution, it is usually better
to transform the original data to a new scale that reduces the skewness.
Example: Transforming Paired Data

The data in Display 9.79 are counts of leprosy bacilli colonies at specified
sites on human subjects. The columns labeled B and A show the counts on
the same subjects before and after an antiseptic was applied. The order of the
measurements in each pair cannot be randomized, but these subjects were
randomly selected from a larger group. The research question is Is the antiseptic
effective in reducing bacteria counts?
Display 9.79
Check conditions.
Counts of leprosy bacilli colonies. [Source: G. Snedecor

and W. Cochran, Statistical Methods, 6th ed. (Ames, Iowa: Iowa
State University Press, 1967), p. 422.]
Solution
This particular group of subjects is a random sample taken from a larger group,
so the randomness condition is satisfied. But the distributions of the original
counts and the distribution of their differences dont appear normal, as you
can see in Display 9.80. Counts of this type are notorious for having skewed
22-03-2009 21:21
Lesson
10 de 23
Transform the original

values, not the
differences.
distributions because they must be non-negative but can get very large. (Think
of counting the number of insects on each plant in a garden.) Display 9.80 also
shows that these particular counts have positively skewed distributions, whereas
their differences are skewed in the negative direction. A transformation is
needed. Although the condition is that the distribution of differences must
be normal, statisticians typically transform the original values rather than the
differences.
B A
A
B
Display 9.80 Boxplots of the original counts, B and A, and of the

differences, B A.
If you take the square root of each of the counts in B and A, you get the more
symmetric boxplots in Display 9.81. The square root transformation typically
does a good job of making the distributions of counts more symmetric.
Display 9.81 Boxplots of the square roots of the counts and of

the differences of the square roots of the counts.
The purpose of applying the antiseptic was, of course, to reduce the bacteria
counts. Evidence that the treatment was effective can be produced through a t-test
of the differences of the square roots. The data in Display 9.79 have the after
counts subtracted from the before counts, so the hypotheses are
H0: d = 0 and Ha: d > 0

where d is the mean of the population of differences of the square roots of the
number of bacteria colonies before and after the antiseptic is applied.
The statistical summary for the differences of the square roots is
Mean of differences of square roots: d = 0.96
Standard deviation of differences of square roots: sd = 0.77
649
22-03-2009 21:21
Lesson
11 de 23
Compute the test

and draw a sketch.
The formula for the test statistic has the same pattern as in the t-test for a single
mean:
t=
=
= 3.943
Give conclusion in
context.
With df = 9, the P-value is 0.0017.

The small P-value, 0.0017, gives ample evidence to support the conclusion
that the bacteria counts were lower after the antiseptic was applied; that is, the
true mean difference is positive. However, because there was no randomization in
the order of treatments, perhaps you cant attribute the decrease to the antiseptic.
There is always the possibility that the bacteria count might have gone down over
time anyway. (See E78 on page 659 to find out how the experimenters eliminated
this possibility.)
What If the Differences Dont Appear Normal?

D17. In analyzing differences taken on skewed data, the data could be
transformed before the differences are calculated, or the differences could
be calculated first and then transformed. Does the order of doing these tasks
matter? Why or why not?
Is It Meaningful to Compare Means?

When your only tool is a hammer, everything looks like a nail. By now you
have quite a variety of statistical tools, but its still worth reminding yourself to be
thoughtful about when and how you use them. Before you start any computations,
always ask yourself, What is the question of interest? Then ask whether it makes
sense to try to answer that question by comparing means.
Example: Comparing Aptitude Tests

Business firms often give aptitude tests to job applicants to see if they have the
necessary skills and interests to adapt well to any training they might have to
complete. Suppose a company desiring to use such a test has two versions it
wants to compare, Test A and Test B. To see how well the exams predict aptitude
for a job, the company decides to try both exams on a sample of recently hired
employees who have completed the training (and whose aptitude they now know).
Is the company interested in comparing the mean scores of the two groups?
Solution
Probably not. The two tests might be graded on different scales, so a comparison
of means would be meaningless anyway. More important, the company really is
interested in how well the tests select the applicants with the greatest aptitude, so
22-03-2009 21:21
Lesson
12 de 23
they might want to compare, say, the top 10% of test takers on each exam to see if
they really are the employees with the greatest aptitude. In general, the variability
of the test scores is more important than the mean. Those with strong aptitude
should score high; those with weak aptitude should score low, with enough
variability in scores so that the two groups are substantively different.
Does Your Inference Have the Chance It Needs?
In statistics, exploratory methods look for patterns but make no assumptions
about the process that created the data; with exploratory methods, what you see is
what you get. Inference can deliver more than exploration, going beyond simply
saying Here are some interesting patterns. For inference to be justified, however
you need the right kind of data. Provided your numbers come from random
samples or randomized experiments, you can use probability theory and the
predictable regularities of chancelike behavior to draw conclusions not only about
the data you see but also about the unseen population from which they came or
about the treatments that were assigned to the experimental units.
Is It Meaningful to Compare Means??

D18. Make a list of ways your data can fall short of the requirements for statistical
inference on means.
Summary 9.5: Paired Comparisons

Textbook illustrations of statistical procedures are always neater than the realworld applications of those same procedures. When confronted with a real
problem involving two samples that has not been sanitized for textbook use, you
should ask some key questions.
Are these paired observations, or two independent samples? If the data are
paired, they are likely to be correlated, and the two-sample procedures are not
the correct method. The better method is to analyze the differences between
pairs as a single random sample using the methods of this section.
How skewed is the population from which these differences came? If
you have paired differences, check to see if its reasonable to assume that
the differences are a random sample taken from a normally distributed
population. If not, try a transformation on the original values (not on the
differences).
Is this a meaningful comparison? Anyone armed with statistical software
can compare the means of any two sets of data. But some might result in
comparing the proverbial apples and oranges, and some might involve
comparing samples that were constructed to have different means in the
first place.
651
22-03-2009 21:21
Lesson
13 de 23
Is there any chance mechanism underlying the selection of these data?

Statistical inference is based on probability theory, and the procedures
work only under the condition that the data come from random samples or
randomized experiments. You can then draw conclusions not only about the
data you see but also about the unseen population from which they came or
about the reason why the treatment groups differ.
Practice
Two Independent Samples, or Paired
Observations?
P33. Display 9.82 shows a set of data on pulse
rates, in beats per minute, from another
class experiment employing the same three
designs as in Display 9.74 on page 642.
Completely Matched
Randomized Pairs
Design
Design
Repeated Measures
Design
a. Display 9.83 shows the sitting pulse rate

plotted against the standing pulse rate for
each of the three designs. For which design
is the relationship strongest? Weakest?
Explain why that should be the case.
Summary
Statistics
Display 9.82 Data tables and boxplots of pulse rate data from another class experiment.
22-03-2009 21:21
Lesson
14 de 23
Display 9.84
Brain weights of a sample of birds and

fish.
a. When testing whether the mean brain

weight is the same for birds and fish, are
these independent samples or paired
observations?
b. What transformation of the brain weights
Display 9.83 Scatterplots of pulse rate data.
improves the conditions for inference?
c.
Assuming these are random samples,
b. Perform a significance test that standing
is there evidence to say that there is a
increases heart rate for the completely
difference in mean brain weight for the
randomized design.
two groups?
c. Perform a significance test that standing
P35.
Radioactive
twins. (Further details of this
increases heart rate for the matched pairs
study are given in Chapter 4 on page 282.)
design.
Most people believe that country air is better
d. Perform a significance test that standing
to breathe than city air, but how would you
increases heart rate for the repeated
prove
it? You might start by choosing a
measures design.
response that narrows down what you mean
e. Compare the P-values for the three
by better. One feature of healthy lungs is
designs in parts bd. Do they show the
that they are quick to get rid of any nasty
pattern you would expect?
stuff they breathe in, such as particles of dust
and smoke. This study used as its response
What If the Differences Dont Appear
variable the rate of tracheobronchial
Normal?
clearance, that is, how quickly the lungs got
P34. Fish or fowl, which are smarter? Display 9.84
rid of nasty stuff.
lists brain weights of a sample of birds and
fish selected from the data in Display 9.31 on
page 603.
653
22-03-2009 21:21
Lesson
15 de 23
<
Investigators managed to find seven

pairs of identical twins who satisfied two
requirements: (1) One twin from each
pair lived in the country and the other
lived in a city, and (2) both twins in the
pair were willing to inhale an aerosol of
radioactive Teflon particles. That was how
the investigators measured tracheobronchial
clearance. The level of radioactivity was
measured twice for each twinright a
inhaling and then again an hour later. The
response was the percentage of original
radioactivity still remaining 1 hour after
inhaling. The data are shown in Display 9.85.
volume of a dozen hens eggs. Display 9.86

gives the lengths and widths, in inches. It
would be possible to use these numbers to
test the null hypothesis that the lengths
and widths are equal. Do you think that
would be a sensible comparison? Why or
why not?
Display 9.86 Length and width measurements, in

inches, of a dozen eggs. [Source: A. P.
Dempster, Elements of Continuous Multivariate
Analysis (Reading, Mass.: Addison-Wesley, 1969),
p. 151.]
P37. An educational researcher is interested in

determining whether calculus students who
study with music playing perform differently
from those who study with no music playing.
a. She asks for volunteers to participate in
Display 9.85 Twin data on radioactivity, in
the study and finds a large number of
percentage of radioactivity remaining.
students who study with music playing
[Source: Per Camner and Klas Phillipson, Urban
and a large number who study with no
Factor and Tracheobronchial Clearance, Archives
of Environmental Health 27 (1973): 82.]
music playing. She then compares the
mean scores on the next calculus exam,
a. Discuss whether a statistical test is
using a t-test. Is this an appropriate
appropriate.
design and analysis for the problem at
b. Regardless of your answer to part a, carry
hand? Explain your reasoning.
out the test.
b. Realizing that the abilities and
c. Tell how to design a study that uses
backgrounds of the students might be
two independent samples (as an
important in this study, the researcher
alternative to paired samples) to compare
pairs students (one student to each
tracheobronchial clearance rates for people
treatment group) from the volunteer
living in urban and rural environments.
groups on the basis of their current grade
d. What are the advantages of using paired
in the class. She then compares the mean
data from a single sample?
scores on the next calculus exam, using
e. What are the advantages of using two
a t-test. Is this an appropriate design and
independent samples?
analysis for this problem? Is it better than
the design in part a?
Is It Meaningful to Compare Means?
c. Explain how you would design a study to
P36. Hens eggs. In the late 1960s, Harvard
compare the effects of the two treatments
statistician Arthur Dempster went into his
on students performance in calculus class.
kitchen and measured the length, width, and
22-03-2009 21:21
Lesson
16 de 23
Exercises
E69. You would expect that teams launching
gummy bears would get better with practice.
Part of the bears-in-space data from Chapter
4 (page 258) is given in Display 9.87. You will
test whether there is statistically significant
evidence that teams improved from launch 1
to launch 10. Use = 0.01.
Display 9.87 Selected bears-in-space data.
a. Should you use a two-sample test of the

difference of two means or a one-sample
test of the mean of the differences?
Explain. Should this be a one-sided test
or a two-sided test?
b. State hypotheses for your test, check
conditions, do the computations, and
write a conclusion in context.
c. Does the result of the test seem sensible?
What type of error might you have made?
d. If you had used the other test, would the
conditions have been satisfied?
E70. Refer to E69. This time you will test to see
if there is statistically significant evidence
that teams improved from launch 5 to launch
10. Use = 0.01. The data are given in
Display 9.88.
Display 9.88 Selected bears-in-space data.
a. Should you use a test of the difference of

two means or a test of the mean of the
differences? Explain. Should this be a
one-sided test or a two-sided test?
b. State hypotheses for your test, check
conditions, do the computations, and
write a conclusion in context.
c. What type of error might you have made?
d. Does the result of the test seem sensible
compared with the result in E69?
E71. Display 9.89 gives the average distance
soared by the gummy bears launched by the
six teams under two conditions: one book
under the ramp and four books under the
ramp. This was a randomized block design,
with a team being the block. Whether the
team did one book first or four books first
was decided by the flip of a coin. You want to
estimate the mean difference of the distance
soared ( four books one book) with 95%
confidence.
Display 9.89 Gummy bear launch data.
a. Should you use a one-sample procedure

or a two-sample procedure? Explain.
b. Are the conditions satisfied for
constructing a confidence interval using
your procedure?
c. Construct and interpret the 95%
E72. An undesirable side effect of some
antihistamines is drowsiness, which can
be measured by the flicker frequency of
patients (number of flicks of the eyelids
per minute). Low flicker frequency is
655
22-03-2009 21:21
Lesson
17 de 23
related to drowsiness because the eyes

stay shut too long. One study reported
data for nine subjects (Display 9.90), each
given meclastine (A), a placebo (B), and
promethazine (C), in random order. At
the time of the study, A was a new drug
and C was a standard drug known to cause
drowsiness.
mid-depth measurements in the same row in

Display 9.91 were taken at the same time and
in the same place.
Display 9.91 Paired aldrin data.
a. Construct a 95% confidence interval

for the mean difference in the aldrin
concentrations. Do you have statistically
Display 9.90 Flicker frequency of subjects, in eyelid
significant evidence that the mean
flicks per minute. [Source: D. J. Hand et al., A
difference is not 0?
Handbook of Small Data Sets (London: Chapman
and Hall, 1994), p. 8.]
b. Do you reach the same conclusion as in
the example on page 626?
a. Are the conditions met for doing
inference on the mean difference between
c. Suppose the observations are paired
drugs A and B? Between drugs B and C?
and so are not independent and the null
Between drugs A and C?
hypothesis is false. Which gives more
power, a one-sample test of the mean
b. Using a 95% confidence interval, estimate
difference or a two-sample test of the
the mean of the differences between
difference of the means?
drugs A and B. Save your conclusion for
part e.
E74. In the fitting of hearing aids to individuals,
it is standard practice to test whether a
c. Using a 95% confidence interval, estimate
particular hearing aid is right for a patient
by playing a tape on which 25 words
drugs B and C. Save your conclusion for
are pronounced clearly and then asking
part e.
the patient to repeat the words as heard.
d. Using a 95% confidence interval, estimate
Different lists are used for this purpose,
and, in order to make accurate checks of
drugs A and C. Save your conclusion for
hearing, the lists are supposed to be of equal
part e.
difficulty with regard to understanding
e. Write a summary of what you learned
them correctly. The research question of
about the three treatments in this
interest is Are the test lists equally di cult
analysis.
to understand in the presence of background
E73. Suppose the aldrin data (see page 620)
noise? Display 9.92 shows the number of
actually had been collected so that the
words identified correctly out of a list of
data were paired. That is, the bottom and
50 words by 24 people with normal hearing,
in the presence of background noise. The
22-03-2009 21:21
Lesson
18 de 23
two different lists were presented to each

person in random order.
Display 9.92 Number of correctly identified words

in two lists. [Source: lib.stat.cmu.edu.]
a. First, suppose the data are not paired.

For example, in the first row, the 28
words in list 1 and the 20 words in list
2 are not for the same person. Use the
information in Display 9.93 to decide
whether the conditions appear to be
met for conducting a two-sample test to
determine whether the mean number of
words understood differs between the
two lists.
b. Do the computations for the two-sample

t-test. Record the P-value.
c. Now use the fact that each row consists
of the two measurements for the same
person. Are the conditions met for
conducting a test of significance for the
mean difference?
d. Do the computations for the one-sample
t-test. Record the P-value.
e. Which test is more powerful, the
two-sample t-test or the one-sample
t-test?
E75. Aerobic exercise is good for the heart. A
group of college students interested in the
effect of stepping exercises on heart rate
conducted an experiment in which subjects
were randomly assigned to a stepping
exercise on either a 5.75-in. step (coded 0)
or an 11.50-in. step (coded 1). Each
subject started with a resting heart rate
and performed the exercise for 3 minutes,
at which time his or her exercise heart rate
was recorded. ( This is a simplification of
the actual design.) The data are shown in
Display 9.94.
Display 9.93 Boxplots for word list data.

Display 9.94 Resting and exercise heart rates (in
beats per minute), categorized by
height of step. [Source: lib.stat.cmu.edu.]
657
22-03-2009 21:21
Lesson
19 de 23
a. You want to test whether the higher step

resulted in a larger mean gain in heart
rate than did the lower step. Is this a
one-sample test or a two-sample test?
Explain.
b. Check conditions and then test whether
the higher step resulted in a larger mean
increase in heart rate.
E76. Refer to E75. Construct and interpret a
95% confidence interval to help you decide
whether the initial random assignment of
treatments to subjects did a satisfactory job
of equalizing the mean resting heart rates
between the two treatment groups.
E77. Bee stings. Beekeepers sometimes use smoke
from burning cardboard to reduce their risk
of getting stung. It seems to work, and you
might suppose the smoke acts like an insect
repellentbut thats not the case. When J. B.
Free, at Rothamsted Experimental Station,
jerked a set of muslin-wrapped cotton
balls in front of a hive of angry bees, the
number of stingers left by the bees made it
quite clear they were just as ready to sting
smoke-treated balls as they were to sting the
untreated controls. Yet in tests using control
cotton balls and cotton balls treated with the
repellent citronellol, the bees avoided the
repellent. On 70 trials out of 80, the control
cotton balls gathered more stings. What,
then, is the effect of the smoke?
One hypothesis is that smoke masks some
other odor that induces bees to sting; in
particular, smoke might mask some odor a
bee leaves behind along with its stinger when
it drills its target, an odor that tells other bees
Sting here. To test this hypothesis, Free
suspended 16 cotton balls on threads from
a square wooden board, in a four-by-four
arrangement. Eight cotton balls had been
freshly stung; the other 8 were pristine and
served as controls. Free jerked the square
up and down over a hive that was open at
the top and then counted the number of
new stingers left behind in the treated and
the control cotton balls. He repeated this
whole procedure eight more times, each time
starting with 8 fresh and 8 previously stung
cotton balls and counting the number of

new stingers left behind. For each of the nine
occasions, he lumped together the 8 balls of
each kind and took as his response variable
the total number of new stingers. His data
are shown in Display 9.95.
Display 9.95 Number of new bee stingers left

behind in cotton balls. [Source: J. B. Free,
The Stinging Response of Honeybees, Animal
Behavior 9 (1961): 19396.]
a. The research hypothesis was that the

previously stung cotton balls would
receive more stings. State the null and
alternative hypotheses for a one-sample
test of significance of the mean difference.
b. Are the conditions met for doing a test of
the significance of the mean difference?
If not, how do you suggest that you
proceed?
c. Conduct your analysis and give your
conclusion.
22-03-2009 21:21
Lesson
20 de 23
E78. In the experiment described in the leprosy

bacilli example on page 648, a control group
received a treatment that had no medical
value. Which subjects were in the antiseptic
group and which were in the control group
was determined by random assignment. The
results for the control group are given in
Display 9.96.
Display 9.97 Bacteria levels in hospital rooms.

[Source: W. G. Walter and A. Stober, Microbial
Air Sampling in a Carpeted Hospital, Journal of
Environmental Health 30 (1968): 405.]
Display 9.96 Bacteria counts before and after

control treatment.
a. Does a square root transformation

improve conditions for inference?
b. Construct a 95% confidence interval to
determine if the control treatment is
effective in reducing bacteria counts.
E79. Hospital carpets. If you had to spend time
in a hospital, would you want your room to
have carpeting or a bare tile floor? Carpeting
would keep down the noise level, which
certainly would make the atmosphere more
restful, but bare floors might be easier to
keep free of germs. One way to measure the
bacteria level in a room would be to pump
air from the room over a growth medium,
incubate the medium, and count the number
of colonies of bacteria per cubic foot of air.
This method was in fact used to compare
the bacteria levels in 16 rooms at a Montana
hospital. Eight randomly chosen rooms had
carpet installed; the floors in the other eight
rooms were left bare. The data are shown in
Display 9.97.
a. Which of the two designs (two
independent samples or paired data)
is the more appropriate model? Why?
b. Tell how to run the experiment using the

other design.
E80. Hospital carpets. Conduct the test that is
appropriate for the design you selected in
part a of process described in E79.
E81. Too tight? You might have lived your entire
life (until now) without once wondering
what statistics has to do with the screw
cap on a bottle of hair conditioner. (Brace
yourself!) The machine that puts on the
cap must apply just the right amount of
turning forcetoo little, and conditioner
can leak out; too much, and the cap might
be damaged and be hard to get off. Imagine
that you are in charge of quality control for a
hair conditioner manufacturer. An engineer
suggests that new settings on the capping
machine will reduce the variability in the
amount of force applied to the bottle caps so
that fewer will be either too tight or too loose
and more will be in the acceptable range. He
offers to set up a comparison: Ten batches
of bottles will be capped by each process,
and the force needed to unscrew the cap
will be measured. To make the comparison
scientific, statistical methods will be used
to test
H0: new = old
versus
H0: new > old
where in each case is the underlying true

mean for the amount of force. What do you
say to the engineer?
659
22-03-2009 21:21
Lesson
21 de 23
E82. One of the important measures of quality

for a product that consists of a mixture of
ingredients, such as cake mix, lawn fertilizer,
or powdered medications, is how well the
components are mixed. This is tested by
taking small samples of the product from
various places in the production process
and measuring the proportions of various
components in those samples. The sample
data then will be a set of proportions for
component A, a set of proportions for
component B, and so on. Does a comparison
of the means of these sets of proportions help
measure the degree of mixing? If not, suggest
another way to assess the degree of mixing.
E83. Guessing distances. Imagine guessing the
width of your classroom. Do you think youd
be as accurate guessing distances in meters as
in feet? Not long after the metric system was
offcially introduced into Australia, a college
professor asked a group of students to guess
the width of their lecture hall to the nearest
meter. He asked a different group to guess
the width to the nearest foot. Display 9.98
presents summary data for their guesses.
Display 9.98 Summary data for guessing the width

of a lecture hall. [Source: M. Hills and the
M345 course team, M345 Statistical Methods,
Unit 1: Data, Distributions, and Uncertainty,
Milton Keynes, The Open University (1986).]
a. Do the numbers come from a single

sample of paired values, or from two
independent samples?
b. Tell how to gather data in the other format.
c. Why do you think the study was done
the way it was instead of the way you
described in part b?
d. A test of the null hypothesis that the

two population means are equal rejects
that hypothesis with a P-value less than
0.0001. Is this a meaningful comparison?
Why or why not?
E84. Buying jeans. Would you guess that people
tend to buy more jeans when the weather is
warm or when it is cold? Display 9.99 shows
the total sales, in thousands of pairs, for
jeans sold in the United Kingdom in January
and in June for the years 198085. A test
of the null hypothesis H0: Jan = Jun gives
a P-value of 0.06. Is this a reasonable use of
significance testing?
Display 9.99 Number of jeans sold in different

seasons, in thousands of pairs. [Source:
S. Conrad, Assignments in Applied Statistics
(Chichester, U.K.: John Wiley and Sons, 1989),
p. 78.]
E85. Old Faithful. The Old Faithful geyser, in

Yellowstone National Park, got its name
from the predictability of its eruptions,
which used to occur about every 66 minutes.
For decades, visitors throughout the warm
months of the year have crowded the large
circle of benches surrounding the geyser
as each eruption time approaches, waiting
to see a 150-ft-high spout of superheated
steam and water shoot into the air. In recent
years, however, scientists have noticed
that the times between eruptions have
become somewhat more variable, and the
distribution of times in fact appears to be
bimodal, as seen in Display 9.100. Additional
study suggests that the time until the next
eruption might depend on the duration
of the previous eruption. To test this
hypothesis, a science class decides to classify
eruptions as long or short and then compare
the mean time until the next eruption for
long eruptions and for short eruptions.
22-03-2009 21:21
Lesson
22 de 23
E87. Auditors often are required to use sampling

to check various aspects of accounts they are
auditing. For example, suppose the accounts
receivable held by a firm are being audited.
Auditors typically will sample some of these
accounts and compare the value the firm
has on its books (book value) with what they
agree is the correct amount (audit value),
usually corroborated by talking with the
customer. (Obtaining the audit values is
a fair amount of work, hence the need for
Display 9.100 Histogram based on a sample of 222
sampling.) Data of this sort are privileged
inter-eruption times taken during
information in most firms, but Display 9.101
August 1978 and August 1979. [Source:
(on the next page) shows what a typical data
S. Weisberg, Applied Linear Regression, 2nd ed.
set might look like. Note that many of the
(New York: John Wiley and Sons, 1985), pp. 231
and 234.]
book values and audit values agree, as they
should. There is no question about the fact
a. Is this a meaningful comparison? Why or
that these are paired data, so an estimate of
why not?
the mean difference between the book values
b. Are the data on times between eruptions
and audit values should proceed from the
paired or unpaired?
differences. The question is what to do with
the differences that equal 0.
a. Should the estimate of the mean
difference between book values and audit
values be based on the entire sample of
40, or on only the 15 nonzero values?
Answer this question by working through
these steps.
i. Construct a 95% confidence interval
estimate using all 40 differences.
ii. Construct a 95% confidence interval
estimate using only the 15 nonzero
differences.
iii. How do the intervals in parts i and
ii compare? Which do you think
provides the better answer? Why?
E86. Old Faithful revisited. Suppose you have
b. If the estimate of the mean difference
the data in Display 9.100 on times between
is based on the nonzero values in the
eruptions but you dont have the data on the
sample, explain how you would use the
corresponding durations. Aha! you think.
result to estimate the total amount by
We want to know whether the difference
which the book values exceed the audit
between the two modes is a real pattern, that
values for all accounts in the population.
is, too big to be due just to chance variation.
The firm knows how many accounts
To test this, Ill just divide the histogram in half
receivable it has; assume this number is
at the low point between the two modes and
1000. (Keep in mind that the total could
compare the means of the upper and lower
be negative if audit values tend to exceed
halves of the data using a statistical test. Is this
book values.)
a reasonable application of hypothesis testing?
661
22-03-2009 21:21
Lesson
23 de 23
E88. In E7 on page 578, you read about the

readability of cancer brochures. Can patients
read the information they are given? Based
on suspicions that the reading level of patient
brochures is higher than the reading level of
patients, a study was conducted to compare
the two. A page was randomly selected from
each of 30 brochures for cancer patients
published by the American Cancer Society
and the National Cancer Institute. The pages
were judged for readability using standard
readability tests. Then 63 patients were
tested for their reading level. The results are
summarized in Display 9.102.
Display 9.102 Frequency of various reading levels,

as measured for brochures and
patients. [Source: Thomas H. Short et al.,
Readability of Educational Materials for
Patients with Cancer, Journal of Statistics
Education 3, no. 2 (July 1995).]
Display 9.101 Book values versus audit values.
a. Is there evidence that the materials are

at a higher mean reading level than the
patients mean reading level? To answer
this question, should you do a onesample test or a two-sample test?
b. Perform the test you selected in part a
and state your conclusion.
c. The question in part a isnt really the
important one. Answer this more
22-03-2009 21:21
Lesson
1 de 9
important question: If a patient reads at

the 10th-grade level and is given one of the
booklets at random, what is the probability
that the booklet will be at a higher reading
level than that of the patient?
d. If each of these patients is given one

booklet at random, how many patients do
you expect to get a booklet that is at too
high a reading level for them?
Chapter Summary
To use the confidence intervals and significance tests of this chapter, you need
either
a random sample from a population (consisting of either single values or
paired values)
two independent random samples from two distinct populations or, in the
case of an experiment, two treatments randomly assigned to the available
subjects
When there is no randomness involved, you proceed with the test only after
stating loudly and clearly the limitations of what you are doing. If you reject the
null hypothesis in such a case, all you can conclude is that something happened
that cant reasonably be attributed to chance.
You use a confidence interval if you want to find a range of plausible
values for
, the mean of your single population
, the difference between the means of your two populations
Both of the confidence intervals you studied have the same form:
statistic
The degrees of freedom for the two-sample t-procedure must be approximated

from a rather cumbersome formula. For that reason, the two-sample t-procedures
should be done with the aid of a calculator or statistics software.
All the significance tests you have studied include the same steps:
1. Justify your reasons for choosing this particular test. Discuss whether the
conditions are met and, if they are not strictly met, decide whether it is okay
to proceed.
2. State the null hypothesis and the alternative hypothesis.
statistic parameter
3. Compute the test statistic, __________________ , and find the P-value. Draw
standard deviation of statistic
a sketch of the situation.
4. Use the computations to decide whether to reject or not reject the null
hypothesis by comparing the P-value to the level of significance, . Then
state your conclusion in terms of the context of the situation. (Reject H0
is not sufficient.) Mention any doubts you have about the validity of that
conclusion.
Chapter Summary
663
22-03-2009 21:21
Lesson
2 de 9
Theres only one way you can be confident of drawing correct conclusions
from data: Use sound methods of data production, either random samples or
randomized experiments. At the other extreme, there are many ways you can
be confident that the conclusions might not be valid. For example, voluntary
response samples are worthless when it comes to inference. In reality, many
situations fall between the two extremes. What then? For example, what if you
use your class as a sample instead of taking a random sample? There are no
formal rules; the value of the inference methods is rarely all-or-nothing. The
more reasonable it is to regard your data as coming from a random sample or a
randomized experiment, the more reasonable it is to trust conclusions based on
the inference methods. For many data sets, making a careful judgment about this
issue is the hardest aspect of a statisticians job.
Review Exercises
E89. To find an estimate of the number of hours
that highly trained athletes sleep each night,
a researcher selects a random sample of
15 highly trained athletes and asks each
how many hours of sleep he or she gets each
night. The results are given in Display 9.103.
E91. Seventy students were randomly assigned

to launch a gummy bear, by themselves,
using either one book or four books (see
page 258). Display 9.104 gives the results
of all 70 first launches by these students. Is
there statistically significant evidence of a
difference in the mean distance? Show all
four steps of a test of significance, being sure
to name the test you are using.
Display 9.103 Hours of sleep for highly trained

athletes.
a. Are the conditions satisfied for

computing a confidence interval for the
mean?
interval for the mean.
E90. For the sleep data in E89, carry out a test
of the claim that the population mean is 8
hours of sleep per night.
a. State the hypotheses, defining any
symbols that you use.
b. Calculate the value of the test statistic,
showing the formula and the values
substituted in.
c. Find the P-value from your calculator
and interpret it in the context of this
situation.
Display 9.104 Summary information for gummy s
bear launches.
22-03-2009 21:21
Lesson
3 de 9
E92. Old Faithful. (Refer to E85 on page 660 for

a description of the data.) Display 9.105
presents summary statistics and side-by-side
boxplots comparing the distributions of
the times until the next eruption following
short eruptions (duration < 3 min) and long
eruptions (duration > 3 min).
Display 9.106 Life expectancies (in years) for a

sample of African countries. [Source:
CIA Factbook, 2006.]
Display 9.105 Summary statistics and boxplots of

times until next eruption following
short and long eruptions.
a. State appropriate null and alternative

hypotheses in words. Then define
notation and restate H0 and Ha in
symbols.
b. Tell whether the design of the study
justifies the use of a probability model,
and give reasons for your answer.
c. Based on the summary statistics and
boxplots, tell whether the shapes of the
distributions raise doubts about using a
t-test.
d. Carry out the computations, Find the
P-value, and tell what your conclusion is
if you take the results at face value.
e. Now tell what you think the results of
the test really mean for this particular
data set.
E93. The data in Display 9.106 show the life
expectancies (in years) of males and females
for a random sample of African countries.
Is there evidence of a significant difference
between the life expectancies of males and
females in the countries of Africa? Give
statistical justification for your answer and a
careful explanation of your analysis.
E94. Altitude and alcohol. On every commercial

passenger flight, there is an announcement
that tells what to do if the cabin loses
pressure. At high altitudes, air has less
oxygen, and you can lose consciousness
in a short time. In 1965, the Journal of the
American Medical Association published a
paper reporting the effects of alcohol on the
length of time subjects stayed conscious at
high altitudes.
There were ten subjects. Each was put in
an environment equivalent to an altitude
of 25,000 ft and then monitored to see
how long the subject could continue to
perform a set of assigned tasks. As soon as
performance deteriorated (the end of useful
consciousness), the time was recorded and
the environment was returned to normal.
Three days later, each subject drank a dose
of whiskey based on body weight1 cc of
100-proof alcohol for every 2 lband then,
after waiting 1 hour for the whiskey to take
effect, was returned to the simulated altitude
of 25,000 ft for another test. Display 9.107
gives the times (in seconds) until the end of
useful consciousness under both conditions.
a. How can you tell from simply the
description of the data, even before seeing
the numbers, that this study gives you
paired data from one sample rather than
two independent samples?
Chapter Summary
665
22-03-2009 21:21
Lesson
4 de 9
Display 9.107 The effect of alcohol consumption

on time of useful consciousness at
high altitudes. [Source: Jay Devore and Roxy
Peck, Statistics: The Exploration and Analysis
of Data (Minneapolis, Minn.: West Publishing
Co., 1986). Original source: Effects of Alcohol
on Hypoxia, Journal of American Medical
Association (December 13, 1965): 135.]
b. Carry out an exploratory analysis.

Include stemplots or boxplots of the
recorded times under each of the two
conditions and a stemplot or boxplot of
the differences, as well as a scatterplot.
Describe the patterns in words, and tell
what questions, if any, the patterns raise
about the validity of formal inference
based on the normal distribution.
true difference, C A. What does
the true difference refer to in this
context? What do you conclude about
the effect of alcohol on the time of useful
consciousness at high altitude?
E95. Select the best answer. The P-value measures
A. the probability that the null hypothesis is
true
B. the probability that the confidence
interval captures the true parameter
C. the probability of getting a result from the
sample as extreme as the one from the
actual data
D. None of the above is a good
interpretation.
E96. Select the best answer. The confidence level

measures
A. the fraction of times the confidence
interval will capture the parameter it
is estimating in repeated use of the
procedure on different samples
B. the fraction of times the confidence
interval will fail to capture the parameter
it is estimating in repeated use of the
procedure on different samples
C. the fraction of times the confidence
interval captures the sample statistic on
which it is based
D. None of the above is a good interpretation.
E97. The distribution of annual incomes of the
employees of a large firm is highly skewed
toward the larger values because of the
high salaries of the upper-level managers.
A random sample of size n is to be selected
for the purpose of testing the claim that the
mean salary is over $50,000.
a. How would you proceed if n = 10?
b. How would you proceed if n = 50?
E98. Boosting your SATs? Several commercial
companies offer special courses designed
to help you get a higher score on the SAT.
Suppose market research shows that students
are willing to take such a course only if,
on average, it raises SAT scores by at least
30 points. Heres a way to test a claim that a
course is able to do that. Youll need some
volunteers, the more representative the
better. Randomly divide them into two
groups. Those in the first group take the
special course; those in the second group
serve as controls. When the course is over,
both groups take the SAT.
Let 1 and 2 be the sample means for the
two groups, let s1 and s2 be the sample
standard deviations, and let n1 and n2 be the
sample sizes.
a. Define suitable notation, and state the
null and alternative hypotheses twice, first
in words and then in symbols.
22-03-2009 21:21
Lesson
5 de 9
b. If you construct a 95% confidence

interval for
, under what
circumstances do you reject H0?
c. Carry out the test using the t-statistic for
1 = 1100, 2 = 1060, s1 = 100, s2 = 80,
and n1 = n2 = 16. Check your conclusion
by constructing a confidence interval.
E99. If you consider all regular players (not
counting pitchers) on Major League Baseball
teams over recent years, the season batting
averages are approximately normally
distributed, with a mean around .250 and a
standard deviation around .050. Carefully
explain each component of a fixed-level test in
the context of these questions. Use = 0.05.
a. The New York Yankees had a team batting
average of .276 in 2005, with 16 regular
players contributing. Is this a higher
average than would be expected for a
random sample of 16 players?
b. The Seattle Mariners had a team batting
average of .256 in 2005, also with about
16 regular players contributing. Can this
team be considered above average in
hitting? [Source: www.mlb.com, October 2005.]
E100. Carbon monoxide. The data in Display 9.108
give the yield of carbon monoxide (CO)
for sidestream smoke (what the smoker
doesnt inhale) and mainstream smoke
(what the smoker inhales) from eight
brands of Canadian cigarettes.
a. Do you have paired data or independent

samples? How can you tell?
b. What is your impression from looking at
the data: Do you need formal inference
to decide whether carbon monoxide
levels are higher in sidestream smoke?
Predict whether the P-value for a
two-sided test will be closest to 0.25,
0.10, 0.05, 0.01, or 0.001.
c. To what extent is a probability model
for the data justified? What is the basis
of the justification? What kinds of
inferences are justified: From a sample
to a population? About cause and effect?
Something else?
d. Make a scatterplot of the data and
describe the pattern in words. Which
brands, if any, stand out? Does the shape
of the plot suggest that your decision in
part a will have a big influence on the
P-value?
e. Compute the differences
(COside COmain ) for the eight brands
and show them in a stemplot. Describe
the shape of the distribution, and tell
whether the shape raises concerns about
the accuracy of P-values based on the
f. Carry out both the paired and the
two-sample t-tests. Are your results
consistent with your prediction in part
d based on the shape of the scatterplot?
How close was your prediction about
the P-value in part b?
g. Write two or three sentences giving your
conclusion about carbon monoxide levels
in mainstream and sidestream smoke.
Display 9.108 Carbon monoxide yields. [ Source: Jay

Devore and Roxy Peck, Statistics: The Exploration
and Analysis of Data (Minneapolis, Minn.: West
Publishing Co., 1986). Original source: American
Journal of Public Health (1984): 22831.]
Chapter Summary
667
22-03-2009 21:21
Lesson
6 de 9
>
AP1. In measuring the angle formed by two

intersecting laser beams in a physics lab,
a student uses chalk dust to illuminate
the beams and then uses a protractor to
measure the angle between them. She takes
ten measurements and then produces a
mean, but is unhappy with the large margin
of error that she calculates. Which of the
following is the worst plan for reducing
the margin of error the next time she takes
similar measurements?
Increase the number of measurements.
Use a more precise method of measuring
angles.
Decrease the confidence level.
Check to be sure any outliers arent
mistakes.
Combine her results with those from
other students.
AP2. You plan to take a random sample of 20 cars
from your neighborhood to see whether
its reasonable to assume that the mean age
of all cars in your neighborhood is 7 years.
You know that the population of all of the
ages of cars in your neighborhood is likely
to be strongly skewed to the right. Which of
the following is not a good reason for doing
a log transformation on your data?
because the population is likely to be
strongly skewed to the right
to make the distribution of the values
more symmetric
to decrease the probability of a Type I
error
to decrease the value of the mean
to help ensure that the probability that a
95% confidence interval will capture the
population mean is 0.95
AP3. Which of the following describes a difference

between the sampling distribution of
z=
and the sampling distribution of
t=
in the case where n = 5?
The distribution of t is more symmetric.
The distribution of t is more skewed.
The distribution of t has a larger spread.
The distributions have different means.
The distribution of t is known, while the
distribution of z is unknown because
is unknown.
AP4. With a sample of size 15 and t = 2.76,
should the null hypothesis be rejected in a
two-sided significance test for a mean?
yes if = 0.05, and yes if = 0.01
yes if = 0.05, but no if = 0.01
no if = 0.05, but yes if = 0.01
no if = 0.05, and no if = 0.01
never, because t is negative
AP5. Researchers wish to estimate, for the
population of married couples, the average
of the differences in the heart rate of each
wife and her husband. They get a random
sample of 100 married couples and measure
the heart rate of each person. What is the
best way to proceed?
Get a random sample of unmarried
couples for comparison.
The independence assumption has
been violated, so the researchers must
start over and get a random sample of
married women and, independently, get
a random sample of married men.
Compute the mean heart rate of the
women and the mean heart rate of the
men. Compute a confidence interval for
the difference between the mean heart
rate of the women and the mean heart
rate of the men.
22-03-2009 21:21
Lesson
7 de 9
the ban went into effect, he scored the

respiratory health of a random sample
of Scottish bar employees. Two months
after the ban, he obtained an independent
random sample of bar employees and
scored their respiratory health. The
increase in the mean scores was statistically
significant (P = 0.049). Which of the
following is the best interpretation of this
result?
An observed difference in sample means
as large or larger than that in this sample
is unlikely to occur if the mean score for
all bar employees before the ban is equal
to the mean after the ban.
The probability is only 0.049 that
the mean score for all bar employees
increased from before the ban to after
the ban.
The mean score for all bar employees
increased by more than 4.9%.
There is a 4.9% chance that the mean
score of all bar employees after the ban
is actually lower than before the ban,
despite the increase observed in the
samples.
Only 4.9% of bar employees had their
scores drop while the other 95.1% had
their scores increase.
AP8. Several statistically trained knights seek
Remove the outlier from the data set.
to estimate the mean tail length of adult
This student is not at all typical.
fire-breathing dragons. They collect a
Because it is so easy to get this
random sample of 10 dragons, observe that
information, get a larger random sample.
the distribution of tail lengths is symmetric
with no outliers, and then compute a
Do a log transformation.
confidence interval. They publish a paper
Do a reciprocal transformation.
stating, We are 95% confident that the
Do a square root transformation.
mean tail length of adult fire-breathing
AP7. Scotland recently imposed a ban on
dragons is between 8.2 feet and 11.3 feet.
smoking in bars. Before the ban, a
The knights later discover that in the
researcher thought that the respiratory
population of all dragons, there are some
health of bar employees should improve
with exceptionally short tails due to
afterer working in smoke-free air. Before
injuries, though none of these dragons were
Compute the mean heart rate of the
women and the mean heart rate of the
men. Compute separate confidence
intervals for the mean heart rate of the
women and for the mean heart rate of
the men.
Subtract each husbands heart rate
from his wifes heart rate, nd the mean
of these differences, and compute a
single confidence interval for the mean
difference.
AP6. To estimate the mean number of hours
that students at a particular college sleep
on a school night, a dean selects a random
sample of 16 students and asks each how
many hours they sleep on a school night.
The number of hours are given in the stemand-leaf plot below. The dean checks, and
the outlier isnt a mistake. What is the best
way for the dean to proceed?
AP Sample Test
669
22-03-2009 21:21
Lesson
8 de 9
included in their sample. Which of the

following is the best description of how the
knights should revise the interpretation of
their confidence interval?
They should lower the center.
They should decrease the width.
They should increase the width.
They should decrease the confidence
level.
They should replace adult with
uninjured adult.
Investigative Tasks
AP9. The stereogram in Display 9.109 contains
the embedded image of a diamond. [Source:
W. S. Cleveland, Visualizing Data (1993). Original source:
J. P. Frisby and J. L. Clatworthy, Learning to See Complex
Random-Dot Stereograms, Perception 4 (1975): pp. 17378,
lib.stat.cmu.edu.] Look at a point between the
two diagrams and unfocus your eyes until

the two images merge into one and the
diamond pops out.
Display 9.109 Stereogram with embedded diamond.
In an experiment, one group (NV) of

subjects was either told nothing or told that
a diamond was embedded. A second group
(VV) of subjects was shown a drawing of
the diamond. The times (in seconds) that
it took the subjects to see the diamond are
summarized in Display 9.110.
Display 9.110 Boxplot and summary statistics of

time, in seconds, to see the diamond.
a. Check the conditions for doing a twosample t-test of the difference in means.
b. The researchers did a two-sample
t-test on the original data, with pooled
variances ( = 0.05 ). Do you agree with
that decision? Why or why not?
c. Replicate the test that the researchers did,
and find the P-value. What conclusion
do you come to if you take the results at
face value?
d. What would the test decision have been
if the variances werent pooled?
e. When the standard deviations arent
comparable, what is the effect of using
the pooled procedure rather than the
unpooled procedure?
22-03-2009 21:21
Lesson
9 de 9
AP10. Refer to the experiment described in AP9.

The results of a log transformation of the
times are given in Display 9.111.
a. Check the conditions for doing a

two-sample t-test on the log-transformed
data of the difference in means.
b. Do a two-sample t-test on the
transformed data, with pooled variances.
Record the P-value.
c. Do a two-sample t-test on the
transformed data, with unpooled
variances. Record the P-value.
d. When the standard deviations are
comparable, what is the effect of using
the pooled procedure rather than the
unpooled procedure?
Display 9.111 Boxplot and summary statistics for the

log transformation of the data.
AP Sample Test
671
22-03-2009 21:21
Lesson
1 de 2
10
Chi-Square Tests
Is it necessary to
have a child at
some point in your
life in order to feel
fulfilled? A poll
asked this question
of samples of
adults in various
countries. You can
use a chi-square
test to decide if
the distribution of
answers (yes, no,
and undecided)
reasonably could
be the same for
the adults in each
country.
22-03-2009 21:22
Lesson
2 de 2
You have now completed two chapters on statistical inference. Chapter 8

considered inference for proportions constructed from categorical variables,
and Chapter 9 considered inference for means constructed from measurement
(quantitative) variables. Chapter 10 returns to categorical variables. You can think
of it as a generalization of the results of Chapter 8.
In Chapter 8, there were only two categoriessuccess or failure, heads or
tails, and so onand you used significance tests to answer questions such as
these:
Fair coin? You spin a coin 100 times and get tails 64 times. Do you have
evidence that spinning the coin is unfair?
Who watches the Super Bowl? In a random sample of 100 college graduates
and 100 high school graduates, 40% of the college graduates and 49% of
the high school graduates watched the last Super Bowl. Is this convincing
evidence that the percentage of college graduates and the percentage of
high school graduates who watch the Super Bowl are not the same?
In this chapter, you will use chi-square tests to answer questions that are similar to
these but usually involve more than two categories:
Fair die? You roll a die 60 times. You get 12 ones, 9 twos, 10 threes, 6 fours,
11 fives, and 12 sixes. Do you have evidence that the die is unfair?
Super Bowl again. In addition to the random samples of college and high
school graduates, you have a random sample of 100 people who didnt
graduate from high school. Of this sample, 37% watched the last Super Bowl.
Do the percentages of people watching the Super Bowl vary with educational
level?
You will use the chi-square technique to answer a new type of problem as well:
Independence of Favorites? Each student in a random sample is asked to name
his or her favorite sport and favorite type of music. Do you have evidence of
an association between favorite sport and favorite type of music, or do these
two variables appear to be independent?
In this chapter, you will learn about three chi-square tests:

Goodness of fit. Are the proportions of the different outcomes in this
population equal to the hypothesized proportions?
Homogeneity of proportions. Are the proportions of the different outcomes in
one population equal to those in another population?
Independence. Are two different variables independent in this population?
The procedure in each case is almost the same. The difference is in the type of
question asked and the kinds of conclusions that you can draw.
22-03-2009 21:22
Lesson
1 de 18
10.1
Testing a Probability Model: The Chi-Square

Goodness-of-Fit Test
The chi-square test was developed by the English statistician Karl Pearson in 1900.
This was more than a small contribution to statistical methodologyit transformed
the way scientists view the world of data. Science 84 (a publication of the American
Association for the Advancement of Science) included the chi-square test as one
of the top 20 discoveries that have changed our lives, along with the invention of
the computer, the laser, plastic, and nuclear energy. The chi-square test was a tiny
event in itself, but it was the signal for a sweeping transformation in the ways we
interpret our numerical world. Now there could be a standard way in which ideas
were presented to policy makers and to the general public. . . . For better or worse,
statistical inference has provided an entirely new style of reasoning. [Source: Science 84
(November 1984): 6970.]
Is the sample consistent

with these proportions?
A goodness-of-fit test determines whether it is reasonable to assume that your

sample came from a population in which, for each category, the proportion of the
population that falls into the category is equal to some hypothesized proportion. If
the result from the sample is very different from the expected results, then you have
to conclude that the hypothesized proportions are wrong. To determine how different
very different is, you will need a test statistic and its probability distribution.
A Test Statistic
To begin a chi-square goodness-of-fit test, make a table with the first column
listing the possible outcomes. In the second column, give the frequency (or count)
that each outcome was observed (O ). In the third column, give the frequencies
you would expect (E ) if the hypothesized proportions are correct.
Example: Fair Die?

Suppose you roll a die 60 times and get 12 ones, 9 twos, 10 threes, 6 fours, 11 fives,
and 12 sixes. Set up a table that you can use to see if you have evidence that the
die is unfair.
Solution
If a fair die is rolled 60 times, you expect to get each face of the die on
60 rolls, or 10 times each. Display 10.1 summarizes this information.
of the
Display 10.1 Observed and expected frequencies for 60 rolls of a fair die.
674 Chapter 10 Chi-Square Tests
22-03-2009 21:22
Lesson
2 de 18
A test statistic is a
measure of the distance
between observation
and model.
You will conclude that the die is unfair if the observed frequencies are far
from the expected frequencies. Are they? Heres where you need a test statistic;
that is, you need to condense the information in the table into a single number
that acts as an index of how far away the observed frequencies are from the
expected frequencies.
Both the squared difference (O E)2 and the relative or proportional
difference (O E)/E are important in determining how far the observed
frequency is from the expected frequency. The test statistic involves both.
The Test Statistic
The test statistic for chi-square tests is
The symbol
is
read chi-square.
Here O is the observed frequency in each category, and E is the corresponding

expected frequency.
Example: The Test Statistic for the Die

Compute the value of for the fair die problem in the previous example using
a table.
Solution
Make a table as shown in Display 10.2.
Display 10.2 Table to compute
The value of
= 2.6.
for the fair die problem.
is the sum of the values in the last column, (O E)2/E, so
A Test Statistic
D1. What is the null hypothesis for the fair die example? What is the alternative
hypothesis?
675
22-03-2009 21:22
Lesson
3 de 18
D2. One test statistic that might be constructed for the data in the fair die
example is
where O represents an observed frequency and E represents the

corresponding expected frequency.
a. Compute the value of this test statistic for the fair die example.
b. Will your result in part a always occur? Prove your answer.
c. Is this a good test statistic?
d. You have seen two other situations in which the sum of the differences
always turned out to be 0. What were those situations? What did you do
in those situations?
D3. Another test statistic that might be constructed is
a. Compute the value of this test statistic for the fair die example.
b. Display 10.3 shows the results from the rolls of two different dice. For
which die does the table give stronger evidence that the die is unfair?
c. Compute and compare the values of
for Die A and Die B.
Does this appear to be a reasonable test statistic? Explain.
Display 10.3 Results from the rolls of two different dice.
D4. Will you reject the hypothesis that the die is fair if is relatively large, if
it is relatively small, or both? How might you determine whether a value
of 2.6 is relatively large?
D5. What is for Die A and for Die B in Display 10.3? Did the larger value of
correspond to the die that you thought seemed more unfair? What purpose
does dividing by E serve in the formula for the test statistic?
D6. If you are given the observed frequencies and the hypothesized proportions,
how do you find the expected frequencies?
D7. The statistic involves a sum of squared differences. What other statistics
have you seen that involve a sum of squared differences?
22-03-2009 21:22
Lesson
4 de 18
The Distribution of Chi-Square

In the fair die example, you saw that turned out to be 2.6. If the observed and
expected frequencies had been exactly equal, would have been 0 and you would
have had no reason to doubt that the die is fair. If the observed and expected
frequencies had been much farther apart than they were, would have been
much larger than 2.6 and you would have had evidence that the die is not fair.
How can you assess whether a value of of 2.6 is large enough to reject the
hypothesis that the die is fair? You need to see how much variation there is in the
value of when a die is fair. In Activity 10.1a, you will do this.
Generating a Chi-Square Distribution

What youll need: dice
1. Form five groups in your class. Your group should roll fair six-sided dice
until your group has a total of 60 rolls. Compute for your groups results.
Get the values of from the other four groups. From just these five results,
does it appear that a value of 2.6 is a reasonably likely outcome when a
fair die is rolled?
2. Describe how to use your calculator to simulate 60 rolls of a fair die. [See
Calculator Note 10A.]
3. Use your calculator to simulate 60 rolls of a fair die. Compute

for your
results. [See Calculator Note 10B.]
4. Continue until your class has 200 values of , where each value is
computed from 60 rolls of a fair die. Display the values in a histogram and
describe the shape of the histogram.
5. Should a one-sided test or a two-sided test be used to test if a die is fair?
Using the results from your simulation, estimate the P-value for a value
of 2.6. Is a value of 2.6 a reasonably likely outcome if the die is fair?
The distributions in Display 10.4 each show 2000 values of
rolls of a fair die.
computed from
Display 10.4 Two histograms of 2000 values of , one with

n = 1000 (left) and one with n = 60 (right).
677
22-03-2009 21:22
Lesson
5 de 18
The effect of the sample

size was accounted for
when you divided by E.
In the histogram on the left in Display 10.4, each value of was computed
from the results of 1000 rolls of a fair die. The histogram on the right was
constructed from the results of only 60 rolls, so it should look very much like
the one your class generated in Activity 10.1a.
As you can see, the distribution does not change much with a change in
sample size, as long as the sample size is large. Is there only one distribution
(like the z-statistic), or are there many (like the t-statistic)? Sample size doesnt
change the shape, center, or spread, but there is one more variable to checkthe
number of categories.
Each histogram in Display 10.5 shows a distribution of 5000 values of .
Each of the 5000 values was computed from the results of 60 rolls of a fair die.
However, each histogram was made using a die with a different number of
sides. Notice how the distribution changes as the number of categories (sides)
changes.
Display 10.5 Five histograms of 5000 values of , each value

computed from 60 rolls of a fair die.
You can use a calculator to compare different chi-square probability density

functions. In these calculator screens, the numbers used in the commands are
number of categories 1 (the degrees of freedom), so the graphs show chisquare density functions for 4, 8, and 20 categories. [See Calculator Note 10C to
learn how to graph a chi-square probability density function, which are continuous
approximations to distributions like those in Display 10.5.]
22-03-2009 21:22
Lesson
6 de 18

a. Describe how the distribution of changes as the number of categories
increases. Discuss these changes in terms of shape, center, and spread.
b. For which die are you most likely to get a value of of 15 or more? For
which die are you least likely to do so?
c. For each die, make a rough approximation of the smallest value of
that would be a rare event (that is, that falls in the upper 5% of the
distribution).
Using the Chi-Square Table and Your Calculator
df = number of
categories 1
To find the critical values of that cut off the upper 5% of the distribution
(or another percentage), you can use the table (Table C) in the appendix on
page 827. A partial table is shown in Display 10.6. In this table, df (degrees of
freedom), is equal to the number of categories minus 1.
The degrees of freedom concept is the same as it was in measuring variation
from the mean. With n data values, the deviations from the mean sum to 0,
so only n 1 of the deviations are free to vary. Similarly, with k categories the
deviations between observed and expected frequencies sum to 0, so only k 1 of
the deviations are free to vary.
Display 10.6 A partial table of critical values of
from Table C.
679
22-03-2009 21:22
Lesson
7 de 18
Example: Critical Values and Approximate P-Values

In the fair die example, the observed value of the statistic was 2.6. Would you
reject the null hypothesis that the die is fair at the 5% level of significance? What
is the approximate P-value?
Solution
To find the value of that cuts off the upper 5% of the distribution when is
computed for rolling a fair six-sided die, go to the row df = 6 1, or 5, in Display
10.6. Then go over to the column headed .05. The critical value, , is 11.07.
The computed value of from the fair die example is only 2.6, so you have no
evidence at the = 0.05 level that the die is unfair. As illustrated in Display 10.7,
it is quite likely that you would get a value of 2.6 or greater with a fair die.
Display 10.7 Chi-square distribution with 5 degrees of freedom

and critical value 11.07.
Looking at the row of values for 5 degrees of freedom in Display 10.6, you
see that the smallest value is 9.24, which cuts off an upper tail area of 0.10. The
observed value, 2.6, is well to the left of 9.24 under the distribution (see
Display 10.7). Therefore, all you can say about the P-value is that it is larger than
0.10. (Table C on page 827 tells you that it is also larger than 0.25.) Statistical
software or a calculator can give you an exact P-value. [To learn how to find
the P-value with your calculator, see Calculator Note 10D .] Here, the P-value is
approximately 0.76.
The Chi-Square Goodness-of-Fit Test

So far in this section, you have used the fair die example to develop the ideas of
a chi-square goodness-of-fit test. However, you can apply this test in important,
real-life situations in which you wish to assess how well a given probability model
fits your data. To perform a chi-square goodness-of-fit test, you should go through
the same steps as for any test of significance.
22-03-2009 21:22
Lesson
8 de 18
Chi-Square Goodness-of-Fit Test

1. Name the test and check conditions.
Each outcome in your population falls into exactly one of a fixed number
of categories.
You have a model that gives the hypothesized proportion of outcomes in
the population that fall into each category.
You have a random sample from your population.
The expected frequency in each category is 5 or greater.
2. State the hypotheses.
H0: The proportions in the population are equal to the proportions in
your model.
Ha: At least one proportion in the population is not equal to the
corresponding proportion in your model.
3. Compute the value of the test statistic, approximate the P-value, and
draw a sketch. The test statistic is
where O is the observed frequency in each category and E is the

corresponding expected frequency.
as extreme as or even
The P-value is the probability of getting a value of
more extreme than the one in the sample, assuming the null hypothesis
is true. Get a P-value from your calculator, or approximate the P-value by
to the appropriate critical value of
in Table C,
comparing the value of
where df = number of categories 1.
of the problem. If the P-value is smaller than (or, equivalently, if
is larger than the critical value for the given ), then reject the null
hypothesis. If not, then you dont have statistically significant evidence that
the null hypothesis is false and so you do not reject it. (Remember that you
dont say you accept the null hypothesis. This is because you dont know
that it is true; you simply dont have any evidence that the hypothesized
proportions are exactly the right ones, only evidence that your data are
consistent with those proportions. Your data will be consistent with other
proportions, too.)
681
22-03-2009 21:22
Lesson
9 de 18
Example: Milk Chocolate Versus Peanut Butter

Milk chocolate M&Ms are 13% red, 14% yellow, 16% green, 20% orange,
13% brown, and 24% blue. A random sample of 200 peanut butter M&Ms
yielded the distribution of colors shown in Display 10.8. Do you have evidence
that the distribution of peanut butter M&Ms is different from the distribution
of milk chocolate M&Ms?
Display 10.8 Distribution of colors in a sample of peanut butter

M&Ms.
Solution
If the distribution of colors for peanut butter M&Ms is the same as that for milk
chocolate M&Ms, you would expect the numbers of each color to be as shown
in Display 10.9. Note that in a chi-square test it is not necessary that all of the
expected frequencies be the same.
Display 10.9 Expected number of M&Ms of different colors.

[Source: us.mms.com, March 2006.]
Check conditions.
State the hypotheses.
The conditions for a chi-square goodness-of-fit test are met in this situation.
You have a random sample of 200 peanut butter M&Ms. Each M&M was only
one color. You can compute the expected number of each color because you know
the distribution of colors in milk chocolate M&Ms. All of the expected counts are
at least 5.
H0: The distribution of colors in peanut butter M&Ms is the same as the
distribution of colors in milk chocolate M&Ms; that is, there are 13% red,
14% yellow, 16% green, 20% orange, 13% brown, and 24% blue.
22-03-2009 21:22
Lesson
10 de 18
Compute the test

statistic, approximate
the P-value, and draw a
sketch.
The alternative hypothesis is

Ha: The distribution of colors in peanut butter M&Ms is not the same as
the distribution of colors in milk chocolate M&Ms; that is, at least one
proportion is different.
=
= 12.33
The value of from the sample, 12.33, is quite far out in the tail of the chi-square
distribution with 6 1, or 5, degrees of freedom. In fact, it is in the upper 0.031
of the tail.
Write the conclusion

situation.
Reject the null hypothesis. You cannot attribute the difference in the expected
and observed frequencies to variation in sampling alone. A value of this large
is very unlikely to occur in random samples of this size if peanut butter M&Ms
have the same distribution of colors as milk chocolate M&Ms. Conclude that
the distribution of colors in peanut butter M&Ms is different from that in milk
chocolate M&Ms. Display 10.10 shows a Fathom printout for this test.
Display 10.10 Fathom printout of the milk chocolate versus

peanut butter M&Ms example.
683
22-03-2009 21:22
Lesson
11 de 18

D9. In the example of milk chocolate versus peanut butter M&Ms, does the
final value of the statistic tell you where the lack of fit occurs (i.e., which
colors might be more prevalent or less prevalent in the peanut butter
candies)? If not, where can you look for such information?
D10. Which of these statements describe properties of the chi-square
goodness-of-fit test?
A. If you switch the order of the categories, the value of the statistic does
not change.
B. The observed frequencies are always whole numbers.
C. The expected frequencies are always whole numbers.
D. The number of degrees of freedom is 1 less than the sample size.
E. A high value of indicates a high level of agreement between the
observed frequencies and the expected frequencies.
D11. Why is a test typically one-sided toward the large values?
Why Must Each Expected Frequency Be 5 or More?

The theoretical distribution used in the table is a continuous distribution. For
example, Display 10.11 shows a (continuous) distribution for ten categories
(df = 9) and another for six categories (df = 5).
Display 10.11 Chi-square distributions for df = 5 and df = 9.
The distribution of computed from repeated sampling is discrete, however,

because only a limited number of distinct values of can be calculated
for a given number of categories and a given sample size. Like the normal
approximation to the binomial distribution, the distribution is a continuous
distribution that can be used to approximate a discrete distribution. The larger
the expected frequencies, the closer the distribution of possible values of is
to a continuous distribution. In order to have a reasonable approximation, the
expected frequency in each category should be 5 or greater. ( This is a conservative
rule, but it works well in most cases.)
22-03-2009 21:22
Lesson
12 de 18
When the Chi-Square Test Is Equivalent to the z-Test

The chi-square test can be thought of as an extension of the z-test to more than
two categories. Suppose you were checking a coin, rather than a die, to see if it
was fair. If you spin the coin n times, you would expect to see heads under the
hypothesis that the spinning coin is fair. Letting x1 denote the number of heads
in the sample of n spins and x2 the number of tails, the table for constructing
a statistic for the data would look like Display 10.12.
Display 10.12 Table for computing
when testing the fairness of a coin.
The test statistic for testing the null hypothesis that spinning the coin is fair
becomes
The sum of the number of heads and tails must be n, so x2 = n x1 and you can
write the test statistic as
But you already know another way of testing the hypothesis that spinning a coin is
fair. For large n, the test statistic for this hypothesis could be the familiar z-statistic
from Chapter 8, given by
You can now see that = z2. The square of a z-statistic has a distribution with
1 degree of freedom. This equality holds in general, even if p isnt .
In summary, if there are only two types of outcomes (success and failure),
you are back in the binomial situation and the z-test from Chapter 8 is equivalent
to the chi-square test of this chapter. This implies, among other things, that
the assumptions for the chi-square test are the same as those for the z-test: The
sample must be random and be large enough so that the sample proportion has an
approximately normal sampling distribution.
685
22-03-2009 21:22
Lesson
13 de 18
When the Chi-Square Test Is Equivalent to the z-Test

D12. Discuss the similarities and differences between a z-test using proportions
and a chi-square test using frequencies.
D13. It is hypothesized that among the students at your school who will be seniors
next year, 20% will buy their lunch on campus, 30% will carry their own
lunch, and 50% will eat off campus. A random sample of these students will
be surveyed to check the claim that only half will eat off campus. How would
you construct this test? Defend your choice.
Summary 10.1: Testing a Probability Model:

The chi-square goodness-of-fit test is used when
each outcome in your population falls into exactly one of a fixed number of
categories
you have a random sample from your population
you want to know if its plausible that the proportion of outcomes in
the population that fall into each category is equal to the corresponding
proportion in some hypothesized model
To perform a chi-square goodness-of-fit test, you go through the same steps
as for any test of significance (see page 681). The expected frequencies are
calculated from the probabilities specified in the null hypothesis, and
df = number of categories 1.
If the results from your sample are very different from the results expected
from the model, then will be large and you will reject the hypothesis that the
model is the correct one for your population. If the observed frequencies from
your sample are close to the expected frequencies, then will be relatively small
and you have no evidence that the model is incorrect.
In the exercises, you will see examples of how the chi-square goodness-of-fit
test is used in the real world.
Practice
A Test Statistic
P3. Refer to Display 10.5 on page 678. Suppose
P1. Suppose you want to test whether a
you roll a 12-sided die 300 times to see if it
tetrahedral (four-sided) die is fair. You roll
it 50 times and observe a one 14 times, a
is fair. You compute and get 21.3. Use the
two 17 times, a three 9 times, and a four
appropriate histogram to approximate a
10 times.
P-value for this test. What is your conclusion
if you are using = 0.05 ?
a. How many do you expect in each
category?
P4. Refer to Display 10.5 on page 678. Suppose
you roll an eight-sided die 100 times to
b. Compute the value of .
see if it is fair, and get = 21.3. Use the
P2. If each observed frequency equals the
appropriate
histogram to approximate a
expected frequency, what is the value of ?
P-value for this test. What is your conclusion
if = 0.05 ?
22-03-2009 21:22
Lesson
14 de 18
Using the Chi-Square Table and Your

Calculator
P5. Give df for each situation, and then use
Table C on page 827 to determine if the
result is statistically significant ( = 0.05).
a. You roll a tetrahedral die 100 times and
calculate a value of 8.24.
b. You roll a 20-sided die 500 times and
calculate a value of 8.24.
P6. Learn to use your calculator to find P-values,
and then find the P-value for the case of
a. rolling an 8-sided die, and = 2.6
b. rolling a 12-sided die, and = 21.3
b. What else would you like to know about

the time frames in which these data were
collected?
P9. Display 10.14, repeated from Chapter 7,
gives the distribution of the number of
children per family in the United States.

P7. For 1000 shoppers donating blood at a local
mall, the frequencies of blood types were as
Display 10.13 Frequency of blood types of

1000 blood donors at a mall.
Theory says these blood types should be in

the ratio 9:8:2:1. For example, the fraction
of people having type O blood should be
Do you have statistically significant evidence
that the data do not support this model?
Be sure to include all four steps of a test of
significance. [Source: P. V. Rao, Statistical Methods in the
Life Sciences (Pacific Grove, Calif.: Duxbury, 1998), p. 221.]
P8. Many people think that the full moon

prompts strange behavior and occurrences.
One commonly held belief is that epileptic
seizures are more frequent during a full
moon. To check this claim, a researcher
observed that, among 470 epileptic seizures
monitored at Tampa General Hospital, 103
occurred during the new moon, 121 during
the first quarter, 94 during the full moon,
and 152 during the last quarter. [Source: Los
Angeles Times, June 7, 2004, p. F2.]
a. Are these data consistent with a model

of equal likelihood of seizures during the
four periods? If not, where are the largest
deviations from the model?
Display 10.14 The number of children per family

in the United States. [Source: U.S. Census
20042005; www.census.gov.]
Suppose you want to determine whether the

distribution of the number of children in the
families of statistics students is consistent
with the distribution for families nationwide.
Record the number of children in the family
of each student in your statistics class. (If
you are unable to do this, use the table in
Display 10.15, on the next page.) Assume
that your class can be considered a random
sample of all statistics students.
a. Do these data satisfy the conditions for a
chi-square goodness-of-fit test?
b. When expected frequencies arent large
enough, investigators sometimes collapse
687
22-03-2009 21:22
Lesson
15 de 18
several categories into one category. Do

that with these data and finish the test.
Display 10.15 Observed frequencies of the number

of children for 40 families.
When the Chi-Square Test Is Equivalent to

the z-Test
P10. Answer the question in the introduction to
this chapter: You spin a coin 100 times and
get tails 64 times. Do you have evidence that
spinning a coin is unfair?
a. Use inference for a proportion, as in
Chapter 8.
b. Use a chi-square goodness-of-fit test.

c. Compare the two results in parts a
and b.
P11. To demonstrate the differences between
flipping and spinning, Professors A. Gelman
and D. Nolan modified a checker by adding
putty to the crown side, calling it heads.
In 100 flips of the checker, they observed
54 heads. In 100 spins of the checker, they
observed 23 heads. The results for the flips are
consistent with the model that the probability
of heads is 0.5, even with the putty. Are the
results for the spins of checkers with putty
consistent with the model that the probability
of heads is 0.5? [Source: stat-www.berkeley.edu.]
a. Answer the question by using inference
for a proportion, as in Chapter 8.
b. Answer the question by using a
chi-square goodness-of-fit test.
c. Compare the two results in parts a
and b.
Exercises
E1. In 1882, R. Wolf rolled a die 20,000 times.
The results are recorded in Display 10.16.
Is this evidence that the die was unfair, or
is it approximately what you would expect
from a fair die?
Display 10.17 Frequencies of outcomes of

120 rolls of a shaved die. [Source:
stat-www.berkeley.edu.]
Clearly, there are more 1s than you would

expect and fewer 6s, the outcome on the
side opposite the 1. Is there evidence that
Display 10.16 Results from 20,000 rolls of a die.
the shaved die was unfair with regard to the
[Source: D. J. Hand et al., A Handbook of Small
Data Sets (London: Chapman & Hall, 1994),
other four outcomes?
p. 29.]
E3. A study attempted to find a relationship
E2. Gelman and Nolan had students roll a die
between peoples birthdays and dates of
that had the corners on the 1 side slightly
admission for treatment of alcoholism. In
rounded. The results of 120 rolls are shown
a sample of 200 admissions, 11 were within
in Display 10.17.
7 days of the persons birthday; 24 were
between 8 and 30 days, inclusive; 69 were
22-03-2009 21:22
Lesson
16 de 18
between 31 and 90 days, inclusive; and

96 were more than 90 days from the persons
birthday. Find appropriate expected numbers
of admissions, and perform a chi-square
goodness-of-fit test to test the hypothesis
that admission is unrelated to a persons
birthday. [Source: Jay Devore and Roxy Peck, Statistics:
The Exploration and Analysis of Data, 3d ed. (Belmont,
Calif.: Duxbury Press, 1997), pp. 56566. Original source:
Psychological Reports, 1992, pp. 94446.]
in 315 smooth yellow peas, 101 wrinkled

yellow peas, 108 smooth green peas, and
32 wrinkled green peas. Are the observed
counts suspiciously close to the counts
predicted by Mendels theory?
E6. The characteristics of the 529 off spring of
the plants discussed in E5 are given in
Display 10.19.
E4. Display 10.18 gives the number of births

for each month in a hospital in Switzerland.
Only the 700 women who were having their
First baby were included. Is there evidence
that first births are not spread evenly
throughout the year?
Display 10.19 Frequencies of the genotypes of 529

pea plants. [Source: www.mendelweb.org.]
Display 10.18 The number of births to first-time

mothers per month in a hospital
in Switzerland. [Source: D. J. Hand et
al., A Handbook of Small Data Sets (London:
Chapman & Hall, 1994), p. 77. Original source:
P. Walser, Untersuchung ber die Verteilung der
Geburtstermine bei der mehrgebarenden Frau,
Helvetica Paediatrica Acta, Suppl. XX ad 24:
3, 130.]
E5. Gregor Mendel (Austrian, 18221884)

performed experiments to try to validate the
predictions of his genetic theory. Mendels
experimental results match his predictions
quite closely. In fact, statisticians think his
results match too closely. In one experiment,
Mendel predicted that he would get a 9:3:3:1
ratio between smooth yellow peas, wrinkled
yellow peas, smooth green peas, and
wrinkled green peas. His experiment resulted
According to Mendels theory, these

genotypes, starting from the top, should
be in the ratio 1:1:1:1:2:2:2:2:4. Do the data
support the theory? Do they support it better
than you would expect?
E7. One of the most famous U.S. elections of all
time (prior to 2000) was the TrumanDewey
election of 1948. It became famous because
Dewey was predicted to win by almost
everyone and by almost every poll. Display
10.20 gives the final predictions (in percentage
of votes) of the three most popular national
polls of the day, along with the actual result.
Display 10.20 Poll predictions and actual results

(percentage of votes) of the 1948 U.S.
presidential election. [Source: F. Mosteller,
The Pre-election Polls of 1948 (New York: Social
Sciences Research Council, 1949).]
689
22-03-2009 21:22
Lesson
17 de 18
Suppose each poll sampled about 3000

voters. Do any of the three poll results
demonstrate a reasonably good fit to the
actual results? Which of the three fits best?
E8. The 1948 presidential election polls of
Crossley, Gallup, and Roper (see E7) were
based on a method called quota sampling,
in which quotas such as so many males, so
many females, so many retired, so many
working, so many unemployed, and so on
were to be filled by field-workers for the
polling company. The state of Washington,
however, tried out a fairly new method called
probability sampling, which made use
of the random sampling ideas used today.
Display 10.21 shows how that poll, based
on about 1000 voters, came out, in terms of
percentages.
Sundays. [Source: D. A. Redelmeier and C. L. Stewart, Do

Fatal Crashes Increase Following a Super Bowl Telecast?
Chance 18, no. 1 (2005): 1924.]
a. There were 662 fatalities in the 4 hours

following Super Bowl telecasts and 936
in the corresponding hours on control
Sundays. Are these data consistent with
the model that the telecast has no effect?
What can you conclude?
b. Within the home state of the winning
Super Bowl team, the number involved in
traffic accidents (fatalities plus survivors)
totaled 141 for Super Bowl Sundays
and 265 for control Sundays. Do these
data fit the model given? What can you
conclude?
E10. It is sometimes said that older people are
overrepresented on juries. Display 10.22
gives information about people on grand
juries in Alameda County, California. Does it
appear that these grand jurors were selected
at random from the adult population of
Alameda County?
Display 10.21 Poll predictions and actual results

(percentage of votes) of the 1948
U.S. presidential election for the state
of Washington. [Source: F. Mosteller, The
Pre-election Polls of 1948 (New York: Social
Sciences Research Council, 1949).]
Can you reject the hypothesis that this

sample result is a good fit to the truth ?
E9. Does the number of people involved in
traffic accidents increase after a Super
Bowl telecast? To answer this question,
investigators looked at the number of
fatalities on public roadways in the United
States for the first 27 Super Bowl Sundays.
They compared the number of fatalities
during the 4 hours after the telecast with
the number during the same time period on
the Sundays the week before and the week
after the Super Bowl (54 control Sundays).
If the telecast had no effect, there should be
roughly twice the number of fatalities on the
54 control Sundays as on the 27 Super Bowl
Display 10.22 Distribution of ages for jurors in

Alameda County, California. [Source:
David Freedman et al., Statistics, 2d ed. (New
York: Norton, 1991), p. 484. Original source:
UCLA Law Review.]
E11. Write a research hypothesis about data

that can be analyzed using a chi-square
goodness-of-fit test. For example, People
are less likely to be born on some days of
the week than on others. Design and carry
out a survey about your hypothesis using a
random sample of a specified population.
E12. A sign on a barrel of nuts in a supermarket
says that it contains 30% cashews, 30%
hazelnuts, and 40% peanuts by weight. You
mix up the nuts and scoop out 20 lb. When
22-03-2009 21:22
Lesson
18 de 18
you weigh the nuts, you find that you have

6 lb of cashews, 5 lb of hazelnuts, and 9 lb of
peanuts.
a. Do you have evidence to doubt the
supermarkets claim?
b. If you have worked this problem using
the chi-square test, convert everything to
ounces and recalculate.
c. What is your conclusion now?
d. Do you see any problem with using a
chi-square test on data like these?
E13. One of the basic principles you have learned
in this textbook is that, all else being equal,
it is better to have a larger sample than a
smaller one.
a. Explain why a larger sample is better.
b. As long as each expected frequency is at
least 5, is there any advantage to having a
larger sample in a chi-square goodnessof-fit test?
E14. You can use statistical software to simulate
the distribution of values for any given
number of degrees of freedom. Display 10.23
shows values of a statistic for 2 degrees of
freedom. Each dot represents 9 points.
Display 10.23 Simulated
a. What distribution of the final digits of

the lengths of the 104 major rivers would
you expect if the U.S. Geological Survey
measures length to the nearest mile?
b. The distribution of the final digits
is shown in Display 10.24. Test the
hypothesis that the digits are equally
likely. If they do not appear to be equally
likely, what possible explanation can you
offer?
distribution for df = 2.
Use software to construct a distribution of

values for various numbers of degrees of
freedom. Describe how these distributions
change as the number of degrees of freedom
increases.
E15. The World Almanac and Book of Facts lists
104 Major Rivers in North America with
their lengths in miles. These data come from
the U.S. Geological Survey. For example,
the length of the Hudson River is given as
306 mi, the Columbia as 1243 mi, and the
Missouri as 2315 mi. Final digits of the

lengths of these three rivers are 6, 3, and 5,
respectively.
Display 10.24 Frequencies of the final digits of the

lengths of the 104 major rivers of
North America.
E16. If you look at the leading digits in a table of

data, you might expect each digit to occur
about of the time. ( There are only nine
possibilities because the digit 0 is never a
leading digit.) However, smaller digits tend
to occur more frequently than larger digits.
691
22-03-2009 21:22
Lesson
1 de 19
Benfords Law says that the digit k occurs

with relative frequency
a. Make a chart that shows the relative

frequencies expected for the digits
1 through 9.
b. Prove algebraically that the sum of the
relative frequencies is 1.
10.2
Are proportions
equal across several
populations?
c. Display 10.25 shows the percentages

of first digits seen in a sample of 100
numbers appearing on the first page of
various newspapers, collected by Dr. Frank
Benford himself. Do they obey his law?
Display 10.25 [Source: mathworld.wolfram.com.]
The Chi-Square Test of Homogeneity

A chi-square test of homogeneity tests whether it is reasonable to believe that when
several different populations are sorted into the same categories, they will have
the same proportion of members in each category. Such populations are called
homogeneous. Before learning the formal procedure, you will work through some
data from a student project.
Categorical Data with Two Variables

For her project, Justine decided to test the dry strength of three brands of paper
towels. Getting a random sample of towels from each brand is pretty much
impossible on a students budget, but she thought it was reasonable to assume
that all towels of the same brand are pretty much identical. So Justine bought one
roll of each of the three brands and used the first 25 towels from each. With the
help of two friends, she stretched each towel tightly, dropped a golf ball on it from
a height of 12 in., and recorded whether the golf ball went through the towel.
Justine tested the 75 towels in random order because she realized that it would
be impossible to hold her testing procedure completely constant. Display 10.26
shows the results from Justines tests.
Display 10.26 Two-way table of observed frequencies for Justines

paper towels data.
The table in Display 10.26 is called a two-way table because each outcome
is classified in two ways: according to the brand of paper towel (the population
it comes from) and whether or not it breaks. Disregarding the labels and totals,
it has two rows, three columns, and six cells. The data are called categorical
because the only information recorded about each response is which category
22-03-2009 21:23
Lesson
2 de 19
A segmented bar chart

stacks the categorical
frequencies on top of
each other.
(or class) it falls into. A cell of the table consists of a count (or frequency). For
example, the 20 in the second row and third column is the number of paper
towels in Justines sample of the 25 Wipe-Outs that did not break.
The column totals, indicating 25 towels of each brand, and the row totals,
indicating 30 towels that broke and 45 that did not break, are called the marginal
frequencies. The summary that or 0.40, or 40%, of the towels broke is called
a marginal relative frequency. A total of 18 of the Wipe-Up towels broke; this
is called a joint frequency (joint between the column variable, brand, and the
row variable, break). Among the Wipe-Ups, or 0.72, or 72%, of the towels
broke; among the Wipe-Outs, only or 0.20, or 20%, broke. These are called
conditional relative frequencies for these columns. The conditional relative
frequency for the first row tells you that, among the towels that broke, or 0.60,
or 60%, were Wipe-Ups.
Are the results from Justines three samples consistent with the null hypothesis
that the percentage of towels that break is the same for all three brands?
As in any analysis, you should look first at a graphical display of the data. One
possible plot is given in Display 10.27. This plot is called a stacked or segmented
bar chart . For each population, the observed frequencies in each category are
stacked on top of each other. From this plot alone, its clear that Wipe-Ups are
more likely to break than are the other two brands.
Display 10.27 A stacked bar chart for Justines paper towels data.

D14. Suppose the probability that a paper towel breaks is the same for all three
brands.
a. Using Justines data in Display 10.26, what is your best estimate of the
probability that a paper towel will break?
b. Use your answer to part a to construct a two-way table for the expected
results from this experiment under the hypothesis that all three brands
have the same probability of breaking.
693
22-03-2009 21:23
Lesson
3 de 19
c. What would the stacked bar chart look like if it were true that the
probability that a paper towel will break is the same for all three brands
and if the results in the sample happened to come out exactly as
expected?
Computing Expected Frequencies

Brad tested 32 Wipe-Ups, 18 Wipe-Its, and 25 Wipe-Outs, and his results are
Display 10.28 Two-way table of observed frequencies for Brads

paper towels data.
If the null hypothesis that the proportion of towels that break is the same for all
three brands of paper towels is true, what is the expected frequency in each cell?
The best estimate of the overall proportion that will break is the marginal
relative frequency
The expected frequency of the 32 Wipe-Ups that will break is

(overall proportion that break) (total number of Wipe-Ups) = (0.4)(32) = 12.8
Following this pattern results in a general formula for expected frequencies.
Formula for the Expected Frequency, E
Brads completed table of expected frequencies appears in Display 10.29.
Display 10.29
Expected frequencies for Brads paper towels data.
22-03-2009 21:23
Lesson
4 de 19
After you finish computing a table of expected frequencies, always check

that the row and column totals are the same as in your table of observed
frequencies.

D15. A chi-square test of homogeneity looks for evidence against the hypothesis
that, for each category, the proportion of the population that falls into the
category is the same for all populations. Is the evidence against this claim for
the three brands of paper towels stronger for Justines data in Display 10.26
or for Brads data in Display 10.28? Does the evidence against the hypothesis
depend more on marginal relative frequencies or on conditional relative
frequencies? Explain your reasoning.
Computing the Chi-Square Statistic

Homogeneous
populations have the
same proportion of
members in any given
category.
A chi-square test of homogeneity can be used to test whether the three brands
of towels are equally likely to break. This test is similar to a chi-square test for
goodness of fit. The value of is computed in the same way, using the observed
and expected values in the six cells of the table, but in this test the expected
frequencies are estimated from the sample data. In the goodness-of-fit test, the
probabilities were specified in the null hypothesis.
Example: Testing Paper Towels

Compute for Justines data in Display 10.26 (page 692).
Solution
The work is easily organized in a table, as shown in Display 10.30.
Display 10.30 Calculating
for Justines paper towels data.
695
22-03-2009 21:23
Lesson
5 de 19
As before,
which is the sum of the last column, so
is approximately 16.34.
Before you can find the P-value for this test, you need to determine the
number of degrees of freedom, df.
Degrees of Freedom for a Test of Homogeneity
For a chi-square test of homogeneity, the number of degrees of freedom is
df = ( r 1)(c 1)
where r is the number of rows in the table of observed values and c is the
number of columns (not counting the headings or totals in either case).
Example: Paper Towels, Degrees of Freedom

What is the number of degrees of freedom for Justines data in Display 10.26
(page 692)? Find and interpret the P-value.
Solution
There are two rows (whether the towel breaks or not) and three columns (the
three brands of paper towels). The number of degrees of freedom is
df = (r 1)(c 1) = (2 1)(3 1) = 2
Using the calculator to determine the P-value for this test, where
16.34,
you get approximately 0.00028. Alternatively, if you look at values in Table C
on page 827, you will find that this value of is significant at the 0.001 level, as
shown in Display 10.31. The P-value, 0.00028, means that if the null hypothesis
that the three brands are equally likely to break is true, then there is almost no
chance of getting observed frequencies as far from the expected frequencies as
Justine did. Consequently, you would reject the hypothesis that the probability of
breaking is the same for all three brands of paper towels.
Display 10.31
distribution with df = 2,
= 0.001.
22-03-2009 21:23
Lesson
6 de 19
Procedure for a Chi-Square Test of Homogeneity

The steps in a chi-square test of homogeneity are much the same as the steps in
the chi-square goodness-of-fit test.
Chi-Square Test of Homogeneity
Independent simple random samples of fixed (but not necessarily equal)
sizes are taken from two or more large populations (or two or more
treatments are randomly assigned to subjects who give a categorical
response).
Each outcome falls into exactly one of several categories, with the
categories being the same in all populations.
The expected frequency in each cell is 5 or greater.
H0: The proportion that falls into each category is the same for every
population.
Ha: For at least one category, it is not the case that each population has
the same proportion in that category; that is, in some category, the
proportion for at least one population is different from that for another
population.
draw a sketch. The test statistic is
For each cell in a two-way frequency table, O is the observed frequency

(count) and E is the expected frequency where:
The P-value is the probability of getting a value of as extreme as or more

extreme than the one in the sample, assuming the null hypothesis is true.
Get the P-value from a calculator, or approximate the P-value by comparing
your value of to the appropriate value of in Table C on page 827 with
df = (r 1)(c 1)
where r is the number of categories and c is the number of populations.
(continued)
697
22-03-2009 21:23
Lesson
7 de 19
(continued)
of the problem. If the P-value is smaller than (or, equivalently, if
is larger than the critical value for the given ), then reject the null
hypothesis. If not, there is no evidence that the null hypothesis is false,
so you do not reject it.
Example: Family Values

The Gallup Organization took a poll on family values in different countries.
One question asked was For you personally, do you think it is necessary or not
necessary to have a child at some point in your life in order to feel fulfilled?
Display 10.32 shows the results from fi ve countries, for samples of 1000 adults.
Display 10.32 Two-way table of observed frequencies and

segmented bar chart for the Gallup poll results.
[Source: www.gallup.com, May 2002.]
From the plot, you can see that the percentages do not appear to be the same
for each country. The difference between India and the United States seems too
great. Test the hypothesis that the proportion of adults who would give each
answer is the same for each country.
22-03-2009 21:23
Lesson
8 de 19
Solution
Under this hypothesis, the expected frequencies are given in Display 10.33.
Display 10.33 Two-way table of expected frequencies for the

Gallup poll results.
Check conditions.
Compute the test

the P-value, and draw
a sketch.
The conditions are met in this situation for a chi-square test of homogeneity.
There are five large populations, and a random sample of size 1000 was taken
from each population. Each answer falls into exactly one of three categories. All
the expected frequencies are at least 5.
H0: If you could ask all adults, the distribution of answers would be the same
for each country.
Ha: The distribution of answers is not the same in each of the five countries.
That is, in at least one country, the proportion of all adults who would give
one of the answers is different from the proportion in another country.
Comparing the test statistic to the distribution with df = (3 1)(5 1), or

8, you can see that the value of from the sample, 628.075, is extremely far out
in the tail and is certainly greater than
= 26.12 ( = 0.001). The P-value is
approximately 0.
Your calculator will perform a chi-square test of homogeneity and draw
a corresponding sketch. [See Calculator Note 10E .]
699
22-03-2009 21:23
Lesson
9 de 19
Display 10.34 shows a Minitab printout of this test.
Display 10.34 Minitab printout for the family values example.

situation.
Reject the null hypothesis. You cannot attribute the differences to the fact
that you have only a random sample of adults from each country and not the
entire adult population. A value of this large is extremely unlikely to occur
in fi ve random samples of this size if the distribution of answers is the same in
each country. Conclude that if you questioned all people in these countries, the
distributions of answers would differ among some countries.
Example: Premature Infants

Chronic lung disease is one of the primary long-term pulmonary complications
among premature infants. This condition generally impairs growth and can result
in poor long-term cardiopulmonary function, an increased susceptibility to
infection, and increased risk of abnormal neurological development.
A research hypothesis suggested that the use of inhaled nitric oxide would
decrease the incidence of chronic lung disease and death in premature infants.
To test this hypothesis, a randomized, double-blind, placebo-controlled study of
inhaled nitric oxide versus a placebo in premature infants undergoing mechanical
ventilation was designed and conducted. A total of 105 infants were randomly
assigned to the nitric oxide treatment group, and 102 infants were assigned to the
control group. Demographic and baseline clinical characteristics did not differ
signifi cantly between the control group and the group given inhaled nitric oxide.
The results for the primary goalsreducing death and chronic lung diseaseare
22-03-2009 21:23
Lesson
10 de 19
Do these data provide evidence that the inhaled nitric oxide and the placebo
result in different distributions of outcomes?
Display 10.35 Table of observed frequencies for the premature

infant study. [Source: Michael D. Schreiber et al., Inhaled
Nitric Oxide in Premature Infants with the Respiratory Distress
Syndrome, New England Journal of Medicine 349, no. 22
(November 27, 2003).]
Check conditions.
Compute the test

the P-value, and draw a
sketch.
Solution
Treatments were randomly assigned to subjects in large enough numbers so
that all expected cell frequencies are at least 5 (see Display 10.36). The conditions
are met.
H0: If both treatments could be assigned to all subjects, the resulting
distributions of outcomes would be the same.
Ha: If both treatments could have been assigned to all subjects, the resulting
distributions of outcomes would differ. (The chi-square test itself does
not tell how they differ.)
Display 10.36 shows the expected frequencies.
Display 10.36 Table of expected frequencies for the premature

infant study.
These frequencies result in a
value of
701
22-03-2009 21:23
Lesson
11 de 19
Comparing this value to the tabled values of the distribution with 2 degrees of
freedom shows that the observed value of the test statistic is a little smaller than
the 5% critical value, 5.99. Thus, the P-value here is a little larger than 0.05 but
smaller than 0.10. A calculator gives a P-value of 0.08.

to the computations
problem.
This is a borderline case. If you strictly interpret this as a fi xed 5% level test, the
decision is not to reject the null hypothesis and to say you dont have statistically
significant evidence that treating with nitric oxide is an improvement over
the placebo. However, the P-value is fairly small and gives some evidence that
something other than random behavior is going on here. As you will see in P20,
this point of view is strengthened if you collapse the death and survival with
chronic lung disease categories into one category; both are considered undesirable
outcomes by the research physicians. The original article reporting this research
ends with a positive recommendation for the nitric oxide treatment under certain
conditions. See P21 for more information on this study.
Procedure for a Chi-Square Test of Homogeneity

D16. Discuss the differences between a chi-square test of homogeneity and a
chi-square goodness-of-fit test.
D17. Discuss the similarities and differences between a z-test of the equality of
two proportions and the chi-square test of homogeneity.
Multiple z-Tests Versus One Chi-Square Test

The decision that the brands of paper towels were not of equal strength in
Justines project was based on frequencies and a chi-square test. Suppose,
however, that the proportion of towels that broke was given for each brand, as
in Display 10.37. It appears that a Wipe-Ups paper towel has a much higher
probability of breaking than one from the other two brands.
Display 10.37 Table of sample proportions from Justines paper

towels data.
22-03-2009 21:23
Lesson
12 de 19
Never use more tests

than you need.
If there were only two brands of paper towels, you could use a z-test to test
whether they are equally likely to break. For Justines data, testing the null
hypothesis p1 = p2 will lead to the conclusion that these two proportions differ.
This two-sample z-test is equivalent to a chi-square test of homogeneity using
only the first two columns of data (a test with 1 degree of freedom). As in
goodness-of-fit tests, = z2. But there are still two more comparisons to be
made: p2 versus p3 and p1 versus p3.
Why not use multiple z-tests rather than one chi-square test to determine
whether the population proportions are homogeneous? This is a deep question,
and the answer relies on complex statistical theory. The simple answer is that you
have a greater chance of coming to an erroneous conclusion with multiple tests
than you do with only one. A guiding statistical principle for hypothesis testing
is to never use more tests than you absolutely need. The chi-square test is a clever
way of combining many z-tests into one overall test. However, it is appropriate
to use z-tests to compare proportions within a table once overall significance has
been established.
Degrees of Freedom: Information About Error

Jodain: Ive just been letting this degrees of freedom thing slide, but in
this section theyve gone too far.
Ms. C: Whats the trouble?
Jodain: Well, just for a starter, back when we did the t-test, df was sample
size minus 1. I was clueless why we subtracted 1, but never
mind that now. Then, in Section 10.1, we had almost the same
formula, number of categories minus 1. But why was it number of
categories, not sample size?
Rest of class: Now we get this hugely different formula, df = (r 1)(c 1)!
(grumble, grumble)
Ms. C: Come on; its a good question. And theres going to be yet another
formula for df in the next chapter, on regression.
Jodain: You keep saying Im not supposed to just memorize stuff , but . . .
Ms. C: Okay, the entire statistical story is way out there, but Ill try to
explain. It has to do with the amount of information in the error
term, or SE. Do you remember SE from the one-sample t-test?
Jodain: Sure, I remember. Its
where
Rest of class: Figures! Jodain remembers everything!
Ms. C: Think about the deviations, (x ). There are n of them. Because
the sum of the deviations from the mean is 0, if you know all the
deviations but one, you can figure out the last one. So you need
only n 1 deviations to get all the information about the size of a
typical deviation from the mean.
703
22-03-2009 21:23
Lesson
13 de 19
Rest of class: Its kind of like if the probability itll rain tomorrow is 0.4, then you
dont learn anything new if they also tell you that the probability
it wont rain is 0.6. You could have already figured that out. So you
really have only one piece of information.
Ms. C: Well, thats sort of close.
Jodain: Then the reason we didnt have to bother with df in a z-test for a
proportion in Section 8.2 is that we knew what the standard error
should be because it depended only on p0.
Ms. C: Right. You werent estimating the error term from the data using a
sum of squared deviations. So you didnt need to worry about df.
Rest of class: We like z-tests!
Ms. C: What does this have to do with df equaling the number of
categories minus 1 for a chi-square goodness-of-fit test?
Jodain: Hmmm. The statistic itself looks like one big error term. If you
are testing whether a six-sided die is fair, there are six categories
and six deviations, O minus E. I suppose if you know all but one of
them, you can figure out that one.
Rest of class: Huh?
Ms. C: Jodains right. The last deviation is determined by the others
because the sum of the deviations from the center is always
equal to 0:
Because the last deviation doesnt give you any new information
about the size of the deviations from the center,
df = number of categories 1
Jodain: But what about the formula in this section, df = (r 1)(c 1)?
Rest of class: (groan) Jodain, why do you do this to us? (mumble, mumble)
Ms. C: Well, how many deviations, O minus E, are you using in the
formula?
Rest of class: We know! The number of rows times the number of columns!
Ms. C: Great answer! Now all you have to do is figure out how many of
these give you new information about the size of O minus E and
how many are redundant.
Jodain: Well, with Justines paper towels example, there were 2 times 3,
or 6, values of O minus E. They have to sum to 0 in each row and
each column. So, for example, if you know the deviations Ive put
in this table (Display 10.38), you can figure out the rest, meaning
that they dont tell you anything new.
22-03-2009 21:23
Lesson
14 de 19
Display 10.38 Table of deviations with one row and one column
missing.
Rest of class: We can be missing an entire row and an entire column of

deviations!
Ms. C: That means you have only (r 1) times (c 1) independent
deviations.
Jodain: Heres my rule:The concept of df applies when I need to use a sum
of squared deviations from a parameter or parameters in my test
statistic and I must estimate the parameter or parameters from the
data. I count the number of deviations that are free to vary. This is
the number of degrees of freedom.
Ms. C: That covers it for everything youll see in this class.
Rest of class: We have a different rulejust give us the formula!
Summary 10.2: The Chi-Square Test of Homogeneity

Use the chi-square test of homogeneity when
you have independent samples from two or more populations
you can classify the response from each member of the sample into exactly
one of several categories
you want to know if its plausible that the proportion that falls into each
category is the same for each population
This might sound familiar. In Section 8.4, you learned how to test whether the
difference between two proportions is statistically significant. This test is the
same as a chi-square test of homogeneity when there are two populations and
two categories. When there are more than two populations or more than two
categories, the chi-square test of homogeneity can test the differences among all
the populations in one test.
In a chi-square test of homogeneity, the expected frequencies are calculated
from the sample data, and
The number of degrees of freedom for a two-way table with r rows and
c columns is
df = (r 1)(c 1)
705
22-03-2009 21:23
Lesson
15 de 19
Practice
P12. Students in a statistics class surveyed
random samples of 50 female students
and 50 male students, asking each if they
preferred a bath or a shower. Display 10.39
gives the results.
Display 10.39 Table of observed preferences for

bath versus shower.
a. Construct a segmented bar chart to help

you see whether males or females are
more likely to prefer a bath.
b. What is your conclusion?
P13. Refer to the data in P12. Give the meaning
of each fraction and tell whether each is a
marginal relative frequency, a conditional
relative frequency, or neither.
a.
b.
c.
d.
P14. Refer to P12. Assuming gender doesnt
affect the probability of preferring a bath
or a shower, construct a table of expected
frequencies for 50 males and 50 females.
P15. Display 10.40 gives the number of
observations that fall into each of five
categories in random samples from three
populations.
a. Construct a segmented bar chart of the
information given in the table.
b. Suppose the populations are homogeneous.
What is your best estimate of the proportion
of each population that falls into Category
A? Category B? Category C? Category D?
Category E?
Display 10.40 Table of observed frequencies of

observations in five categories in
samples from three populations.
c. Compute the expected frequency for each

cell if the populations are homogeneous.
d. Does it appear that the three populations
have the same proportion of members
that fall into any given category?
Constructing the Chi-Square Statistic
P16. Compute for the bath/shower data in P12.
What is the number of degrees of freedom
for that table? Estimate and interpret the
P-value.
P17. Compute for the data in P15. What is the
number of degrees of freedom for the table?
Is significant at the 5% level?
Procedures for a Chi-Square Test
of Homogeneity
P18. Verify the table of expected frequencies in
Display 10.33 in the family values example
on pages 698699.
P19. Suppose the value of for the family values
example on pages 698699 had turned out to
be 11.35. Estimate the P-value for such a test.
Rewrite the conclusion.
P20. In the premature infants example on pages
700702, collapse the two undesirable
outcomes, death and survival with chronic
lung disease, into one category and analyze
the resulting 2 2 table. Is there now a
the two distributions?
22-03-2009 21:23
Lesson
16 de 19
P21. Another important aspect of the study

on premature infants was the effect of the
nitric oxide treatment on intraventricular
hemorrhage (bleeding within the heart).
Analyze the results reported in Display
10.41 to see if the two distributions differ
significantly. If they do, explain the nature of
the difference.
Display 10.41 Table of observed frequencies for
premature infants.
Exercises
Each time you are asked to perform a chi-square test,
include all four of the steps given on pages 697698.
E17. A recent Gallup poll asked the same question
the Gallup Organization has asked every
year for many years: What do you think
is the most important problem facing
this country today? Display 10.42 shows
the percentage responses for three major
concerns from 2003 to 2006.
Display 10.42 Results from a Gallup poll: responses

for three major concerns, 20036.
[Source: www.gallup.com.]
a. Assume that 1000 people were surveyed

each year, and convert the table to one
displaying frequencies. Construct and
interpret a segmented bar chart of these
frequencies.
b. What are the populations in this case?
c. Perform a chi-square test of homogeneity.
Use = 0.05.
E18. Overall, how satisfied are you with the

quality of education students receive in
kindergarten through grade twelve in
the U.S. todaywould you say you are
completely satisfied, somewhat satisfied,
somewhat dissatisfied or completely
dissatisfied?This question was posed to a
random sample of about 1000 adults in 2004
and another sample of the same size in 2005.
The results are shown in Display 10.43. Was
there a significant change in the distribution
of results from 2004 to 2005? If so, what was
the direction of the shift?
E19. Joseph Lister (18271912), a surgeon at
the Glasgow Royal Infirmary, was one of
the first to believe in Pasteurs germ theory
of infection. He experimented with using
carbolic acid to disinfect operating rooms
during amputations. Of 40 patients operated
on using carbolic acid, 34 lived. Of 35
patients operated on not using carbolic acid,
19 lived. [Source: Richard Larson and Donna Stroup,
Statistics in the Real World: A Book of Examples (New York:
Macmillan, 1976), pp. 2057. Original reference: Charles
Winslow, The Conquest of Epidemic Diseases (Princeton, N.J.:
Princeton University Press, 1943), p. 303.]
a. Display these data in a two-way table and

in a segmented bar chart.
Display 10.43 Results of a poll about satisfaction with the quality of education in 2004 and
2005. [Source: poll.gallup.com, 2005.]
707
22-03-2009 21:23
Lesson
17 de 19
b. Compute the value of the test statistic,

z, for a test of the difference of two
proportions. Find and interpret the
P-value for this test.
c. Compute the value of for a test of
homogeneity. Find and interpret the
d. Compare the two P-values from parts b
and c. Compare the values of z2 and .
E20. Obesity is associated with many long-term
illnesses, such as diabetes and hypertension.
In a Swedish study, a group of subjects
who had undergone surgery for obesity
were followed over a 10-year period and
compared, over that time, to a matched
sample of conventionally treated control
subjects. The matching was done based
on 18 variables, including age, gender,
weight, smoking habit, and many other
health-related variables. Among 517 subjects
who had surgery to reduce obesity,
7 developed diabetes. Among 539 subjects
in the control group, 24 developed diabetes.
[Source: Lars Sjstrm et al., Lifestyle, Diabetes, and
Cardiovascular Risk Factors 10 Years after Bariatric Surgery,
New England Journal of Medicine 351, no. 26 (December 23,
2004).]
a. Display the data in a two-way table and in

a segmented bar chart.
b. Compute the value of the test statistic,
z, for a test of the difference of two
proportions. Find and interpret the
c. Compute the value of for a test of
homogeneity. Find and interpret the
d. Compare the two P-values in parts b and
c. Compare the values of z2 and .
E21. Suppose an experiment is designed to
determine the effects of a bandage on the
amount of pain a child perceives in a skinned
knee. Of the first 60 children who came to
a school nurse for a skinned knee, 20 were
selected at random to receive no bandage,
20 to receive a skin-colored bandage
(matched to their skin color), and 20 to
receive a brightly colored bandage. After
15 minutes, each child was asked, Is the
pain gone, almost gone, or still there? Of

the 20 children who got no bandage, 0 said
the pain was gone, 5 said it was almost
gone, and 15 said it was still there. Of the
20 children who got a flesh-colored bandage,
9 said the pain was gone, 9 said it was almost
gone, and 2 said it was still there. Of the
20 children who got a brightly colored
bandage, 15 said the pain was gone, 2 said it
was almost gone, and 3 said it was still there.
Organize these data into a table, display
them in a plot, and perform a chi-square test
of homogeneity.
E22. How does the amount of crime this year

compare with the amount last year? This
question was asked of a random sample
of 1010 persons in Great Britain and of an
independent random sample of 1012 persons
in the United States. In Great Britain,
43% said more, 25% said less, 21% said the
same, and the rest had no opinion. In the
United States, 47% said more, 33% said
less, 18% said the same, and the rest had no
opinion. Is there evidence of a significant
difference in the response pattern between
the two countries? Organize these data in a
table, display them in a plot, and perform a
chi-square test of homogeneity. [Source: poll.
gallup.com, 2005.]
E23. Gallup Poll Monthly reported on a survey

built around the question What is your
opinion regarding smoking in public
places? On the workplace part of the survey,
respondents were asked to choose which of
three policies on smoking in the workplace
they favored. The percentages of the sample
making each choice for four different years
22-03-2009 21:23
Lesson
18 de 19
are shown in Display 10.44. Display these

data in a bar chart. Perform a chi-square test
of homogeneity to test if the proportion of
people who chose each response was the same
across the four years. Assume the sample size
was approximately 1000 each year.
E25. How does the United States compare to

Canada and Great Britain with regard
to crime within households? Recent
polls regarding victimization within the
households in which the respondents lived
were conducted in the three countries; the
results are shown in Display 10.46. Do the
data show significantly differing crime rates
for the three countries surveyed? If so, where
do the differences occur?
Display 10.44 Results from Gallup polls on smoking

in public places. [Source: poll.gallup.com,
2005.]
E24. Gallup polls in 1997 and 2006 asked a

randomly selected national sample of
approximately 1000 adults, 18 years and
older, this question: Which of the following
statements reflects your view of when the
effects of global warming will begin to
happen? (1) They have already begun to
happen, (2) they will start happening within
a few years, (3) they will start happening
within your lifetime, (4) they will not happen
within your lifetime, but they will affect
future generations, or (5) they will never
happen.The responses are shown in Display
10.45. Organize the data, display them in a
segmented bar chart, write an appropriate
null hypothesis, and perform a chi-square
test of homogeneity.
Display 10.45 Results from a Gallup poll on global

warming. [Source: www.pollingreport.com,
September 2006.]
Display 10.46 Results of polls regarding crime

within households. [Source: poll.gallup
.com, 2005.]
E26. How does the United States compare to

Canada and Great Britain with regard to
crime against individuals? Recent polls
regarding victimization of those responding
to the survey were conducted in the three
countries; the results are shown in Display
10.47. Do the data show significantly
differing crime victimization rates for the
three countries surveyed? If so, where do the
largest differences occur?
Display 10.47 Results of polls regarding crime

against individuals. [Source: poll.gallup
.com, 2005.]
709
22-03-2009 21:23
Lesson
19 de 19
E27. Display 10.48 (at the bottom of the page)

contains the response percentages from
recent surveys that asked this question of
randomly selected adults in three countries:
How often, if ever, do you drink alcoholic
beverages such as liquor, wine, or beer
every day, a few times a week, about once a
week, less than once a week, only on special
occasions such as New Years and holidays,
or never? Do the three countries differ with
respect to drinking patterns? If so, explain
where the main differences occur.
E28. It is generally assumed that men are heavier
drinkers of alcoholic beverages than are
women. Display 10.49 gives the percentages
of adults in independent surveys of about
500 men and 500 women from each of three
countries who responded yes to the question
of whether they had at least several alcoholic
drinks per week.
Display 10.49 Percentages of people who drink

regularly, out of 500 men and
500 women. [Source: poll.gallup.com, 2006.]
Describe the test you would use to answer

each question. Show the test statistic (with
all numbers substituted in) that you would
compute.
a. Is the proportion of men who drink
regularly significantly greater than
the proportion of women who drink
regularly in the United States?
b. Are there significant differences among

the proportions of men who drink
regularly across the three countries?
c. Is there a significant difference between
the proportions of women who drink
regularly in Canada and in the United
States?
E29. Gallup conducted a poll of about 750
Internet users in 2003 and a poll with a
different sample of about 750 Internet
users in 2005. Display 10.50 gives some
of the findings. For example, 68% of
those interviewed used the Internet for
checking news and weather in 2003, and the
percentage jumped to 72% in 2005.
Display 10.50 Results of polls regarding Internet

users habits, for samples of
750 people in 2003 and 750 different
people in 2005. [Source: poll.gallup.com,
2006.]
a. Can this data table be used to conduct a

chi-square test of homogeneity on these
proportions across the two years? If so,
find and interpret the P-value for this test.
If not, explain why not.
b. Could you test for equality of proportions
of those sending and reading e-mail
across the two years? If so find and
interpret the P-value for this test. If not,
explain why not.
Display 10.48 Results of surveys on drinking patterns. [Source: poll.gallup.com, 2006.]
22-03-2009 21:23
Lesson
1 de 17
E30. Two questions were asked in random order

of a random sample of about 1000 adults
in a recent poll: At the school your oldest
child attends, do you think there is too
much emphasis, the right amount, or too
little emphasis on sports? On preparing for
standardized tests?The results are shown
in Display 10.51. Can these data be used to
test for homogeneity of proportions across
the two areas by the methods in this section?
If so, find and interpret the P-value for this

test. If not, explain why not.
E31. Look up the definition of homogeneous.
Explain what two homogeneous populations
would be.
E32. Describe how you can tell, for a given set
of data, whether to use a chi-square
goodness-of-fit test or a chi-square test of
homogeneity.
Display 10.51 Results of a poll of parents opinions of the amount of emphasis schools
place on sports and standardized test preparation, for a sample of
1000 adults. [Source: poll.gallup.com, 2005.]
10.3
Are these variables
independent in this
population?
If categorical variables
arent independent, we
call them dependent or
associated.
The Chi-Square Test of Independence

Suppose you take a random sample of people from your community. From this
sample alone, you could probably tell that being over 40 and having some gray
hair are not independent characteristics in the population of all people. If your
community is typical, people over 40 are more likely to have some gray hair than
are people 40 or younger. The two categorical variables for this single sample
are age (with categories over 40 and 40 or younger) and hair (with categories has
some gray hair and doesnt have gray hair). For each of these two variables, each
member of the population falls into exactly one category.
When all you can observe is a single sample from a population, such as the
sample from your community, you can use a chi-square test of independence
to decide whether it is reasonable to believe that two different variables are
independent in that population. In Activity 10.3a, you will collect some data that
will be used to illustrate the ideas in this section.
Independent or Not?
1. In the next step, you will count the number of females and males in your
class. You will also ask if the last digit of their phone number is even or
odd. Do you think that the categorical variables gender and even/odd phone
number are independent? Explain.
2. Collect the data described in step 1 and organize the data in a two-way
table.
(continued)
711
22-03-2009 21:23
Lesson
2 de 17
(continued)
3. Using the definition of independence in Chapter 5, determine if gender
and even/odd phone number are independent variables in the selection of a
student at random from your class.
4. Based on your answer to step 3, what are you willing to conclude about the
independence of the two variables in the population of all students? Explain.
5. Even if gender and even/odd phone number are independent variables in the
population of all students, why are you likely to find that this is not the case
in your class?
6. In Activity 5.4a, you determined whether you are right-eye dominant or
left-eye dominant. Retrieve the table you made then, or make another
two-way table for your class in which one categorical variable is eye
dominance (left or right) and one categorical variable is hand dominance
(left or right).
7. For a randomly selected student from your class, are the events right-eye
dominant and right-hand dominant independent according to the
definition of independence in Chapter 5?

D18. In Chapter 5, you showed that two events A and B are independent by
verifying that one of these statements is true (and if one statement is true, all
three are true):
P(A) = P(A | B)
P(B) = P(B | A)
Why do you need a test of independence here when you already have one
from Chapter 5?
D19. Why do you think a chi-square test of independence is sometimes called a
test of association?
Tabular and Graphical Displays of the Data

Now you will see what two-way tables and graphical displays of categorical data
look like when two variables are dependent (or associated). In Chapter 5 you saw
the data on the Titanic disaster, repeated in Display 10.52.
Display 10.52 Titanic survival data. [Source: Journal of Statistics Education 3,

no. 3 (1995).]
22-03-2009 21:23
Lesson
3 de 17
In testing for independence, there is only one population with each subject or
unit categorized according to two variables. For the Titanic data, the categorical
variables are Class of Travel and Survival Status. Recall from Section 10.2 that the
number of first-class travelers who survived, 203, is a joint frequency, the number
of survivors, 499, is a row marginal frequency and the number of second-class
travelers, 285, is a column marginal frequency. Of great importance to the
notion of independence are the conditional relative frequencies. The column
conditional relative frequencies, shown in Display 10.53, give the proportions of
those who survived or not for each class of traveler. Notice that these are quite
different from column to column, indicating that the proportion of survivors
changes dramatically from class to class. In other words, the survival rate depends
on (or is associated with) the class of travel.
Display 10.53 Column conditional relative frequencies for the

Titanic data.
Will the row conditional relative frequencies tell much the same story?
Display 10.54 gives these data, which show what proportion of the survivors fall
into each class of travel as well as what proportion of the non-survivors fall into
each class of travel. Again, these conditional distributions are quite different,
indicating that there is evidence of an association between survival status and
class of travel.
Display 10.54 Row conditional relative frequencies for the Titanic

data.
The segmented bar chart was a natural graphic to use for studying
homogeneity because, in that situation, you had different populations that could
be compared with side-by-side bars. In studying independence you have one
population and one sample categorized by two variables, so you can construct a
segmented bar chart on either variable, as shown in Display 10.55 (on the next
page). You can tell that the two variables are not independent by observing that
the bars arent divided into segments according to the same proportions. For
example, in the left-hand chart, the first bar is more than two-thirds Survived,
whereas the third bar is only about one-third Survived.
713
22-03-2009 21:23
Lesson
4 de 17
Display 10.55 Segmented bar charts for the Titanic survival data.
The column chart in Display 10.56 treats the two variables symmetrically.
You can see that the two variables do not appear to be independent because the
columns in the back row dont follow the same pattern as the columns in the front
row. From left to right, the heights of the columns in the back row go shortest,
middle, tallest; in the front row, the heights go tallest, shortest, middle.
Display 10.56 A column chart for the Titanic survival data.
Tabular and Graphical Displays of the Data

D20. Refer to the three plots in Displays 10.55 and 10.56.
a. For each plot, give one fact that the plot shows more clearly than the
other two plots.
b. How can you tell from the right-hand segmented bar chart that class of
travel and survival status do not appear to be independent?
D21. Construct two different segmented bar charts and one column chart to
display the data your class collected in Activity 10.3a on the variables
gender and even/odd phone number. How do your plots show whether these
variables appear to be independent?
D22. What does a column chart look like if two variables are independent?
22-03-2009 21:23
Lesson
5 de 17
Expected Frequencies in a Chi-Square Test

of Independence
In a chi-square test of independence, the null hypothesis is that the variables are
independent. As you learned in Chapter 5, events A and B are independent if
and only if P(A and B) equals P(A) P(B). With this definition, you can fill in
the two-way table of expected frequencies in Display 10.57 if you assume that
handedness and eye color are independent variables.
Display 10.57 A possible two-way table for the variables

handedness and eye color for n = 30.
Assuming independence, the probability that a randomly selected person is

right-handed and blue-eyed is
P(right-handed and blue-eyed) = P(right-handed) P(blue-eyed)
The expected frequency of people who are right-handed and blue-eyed is then
In general, if two variables are independent, the probability that a randomly

selected observation falls into the cell in column C and row R is
Thus, the expected number of observations that fall into this cell is
This is the same formula as in Section 10.2.
715
22-03-2009 21:23
Lesson
6 de 17
Frequencies in a Chi-Square Test of Independence

D23. Refer to your data from Activity 10.3a.
a. Complete an expected frequencies table for your class assuming gender
and even/odd phone number are independent variables.
b. Complete an expected frequencies table for your class assuming
handedness and eyedness are independent variables.
c. For which of the two tables in parts a and b are the expected and
observed values farther apart?
Procedure for a Chi-Square Test of Independence

Now that you have reviewed the idea of independence, its time to organize all the
steps needed when you examine a sample from a population in order to decide if
two variables are independent in that population.
The steps in a chi-square test of independence are much the same as the steps
in the chi-square test of homogeneity.
Chi-Square Test of Independence
A simple random sample is taken from one large population.
Each outcome can be classified into one cell according to its category on
one variable and its category on a second variable.
The expected frequency in each cell is 5 or greater.
H0 : The two variables are independent. That is, suppose one member
is selected at random from the population. Then, for each cell, the
probability that the member falls into both category A and category
B, where A is the category from the first variable and B is the category
from the second variable, is equal to P(A) P(B).
Ha: The two variables are not independent.
draw a sketch.
(continued)
22-03-2009 21:23
Lesson
7 de 17
For each cell in the table, O is the observed frequency and E is the expected
frequency where:
E=
(row total) (column total)

______________________
grand total
The P-value is the probability of getting a value of as extreme as or

more extreme than the one in the sample, assuming the null hypothesis is
true. Get the P-value from your calculator or approximate the P-value by
comparing the value of to the appropriate value of in Table C on page
827 with df = (r 1)(c 1), where r is the number of categories for one
variable and c is the number of categories for the other variable.
4. Write your conclusion linked to your computations and in the context of
the problem. If the P-value is smaller than (or, equivalently, if is larger
than the critical value for the given ), then reject the null hypothesis. If not,
there is no evidence that the null hypothesis is false, so you do not reject it.
Example: Scottish Children

The information in Display 10.58 shows the eye color and hair color of 5387
Scottish children. You might suspect that these two categorical variables are not
independent because people with darker hair colors tend to have darker eye
colors, whereas people with lighter hair colors tend to have lighter eye colors. Test
to see if there is an association between hair color and eye color.
Display 10.58 Eye color and hair color data from a sample of
Scottish children. [Source: D. J. Hand et al., A Handbook
of Small Data Sets (London: Chapman & Hall, 1994), p. 146.
Original source: L. A. Goodman, Association Models and
Canonical Correlation in the Analysis of Cross-Classifi cations
Having Ordered Categories, Journal of the American Statistical
Association 76 (1981): 32034.]
Solution
From the column chart in Display 10.59, you can see that it does indeed appear
to be the case that children with darker hair colors tend to have darker eye colors,
whereas children with lighter hair colors tend to have lighter eye colors. Compare,
especially, the rows for fair hair and dark hair.
717
22-03-2009 21:23
Lesson
8 de 17
Display 10.59 Column chart for hair color and eye color.
Check conditions.
Compute the test

statistic, calculate the
P-value, and draw a
sketch.
This situation satisfies the conditions for a chi-square test of independence

if the children can be considered a simple random sample taken from one large
population. Each child in the sample falls into one hair color category and one
eye color category. The Data Desk printout in Display 10.60 shows the expected
frequencies under the assumption of independence. The expected frequency in
each cell is 5 or greater.
H0 : Eye color and hair color are independent.
Ha: Eye color and hair color are not independent.
Your calculator gives a value of 1240, extremely far out in the tail, and a
corresponding P-value of approximately 0. [See Calculator Note 10E to learn how to
use your calculator to find these statistics.]
Display 10.60 also gives the computation of the test statistic, , and the P-value.
Notice that the expected frequencies in each cell are computed from the marginal
frequencies. Display 10.60 also includes the conditional relative frequencies for
the columns (hair color): Among those with fair hair, 22.4% have blue eyes;
among those with black hair, only 2.54% have blue eyes.
22-03-2009 21:23
Lesson
9 de 17
Display 10.60 Data Desk printout of a chi-square test of

independence for the Scottish children example.
A sketch of the distribution is shown in Display 10.61.
Display 10.61 Chi-square distribution with 12 degrees of

freedom.
situation.
Reject the null hypothesis. These are not results you would expect for a
sample from a population in which there is no association between eye color and
hair color. As you can see from Display 10.61, a value of 1240 is much larger
than the value of 32.91 that cuts off an upper tail of 0.001. Thus, a value of of
1240 or larger is extremely unlikely to occur in a sample of this size if hair color
and eye color are independent. Examining the table and the column chart, you
might conclude that darker eye colors tend to go with darker hair colors and
lighter eye colors tend to go with lighter hair colors.
719
22-03-2009 21:23
Lesson
10 de 17
Procedures for a Chi-Square Test of Independence

D24. Refer to your data from Activity 10.3a.
a. Assuming your class is a random sample of students, test the hypothesis
that gender and even or odd phone number are independent
characteristics among students.
b. Assuming your class is a random sample of students, test the hypothesis that
handedness and eyedness are independent characteristics among students.
Homogeneity Versus Independence

Ms. C: Did you notice that a chi-square test of homogeneity and a
chi-square test of independence are performed in exactly the same
way, except for the wording of the hypotheses and the conclusion?
The two null hypotheses say exactly the same thing about the
table: The columns are proportional.
Class: So how do we know which test to use?
Ms. C: You cant tell from the table itself. You have to find out how the
data were collected. Did they take one sample of a fixed size from
one population and then classify each person according to two
categorical variables? If so, its a test of independence. Or did they
sample separately from two or more populations and then classify
each person according to one categorical variable? If so, its a test
of homogeneity.
Class: Huh?
Ms. C: Its simple. Suppose you want to find out whether there is an
association between age and wearing blue jeans. You take a
random sample of 50 people under age 40. Then you take a
random sample of 50 people age 40 or older. Your table might look
like Display 10.62.
Display 10.62 Sample table for blue jeans data, collected from
two populations.
Class: We get it! We have to do a test of homogeneity because we sampled

separately from two populations: people under 40 and older
people. There are two populations and one variable about jeans.
Jodain: Now I see what you mean by the columns being proportional.
For people under 40, 42% are wearing jeans now, 46% are not
22-03-2009 21:23
Lesson
11 de 17
wearing them now, and 12% never wear them. Thats about the
same distribution as for people 40 or older, so we wont be able to
reject the hypothesis that these populations are homogeneous with
respect to wearing jeans. Jeans-wearing behavior is about the same
for both age groups.
Ms. C: Right. In the homogeneity case, you have one variable but several
populations. The columnor sometimes the rowproportions
represent the separate distributions for those populations. Now
suppose you go out and take a random sample of 100 people,
classify them, and get the data in Display 10.63. This time you
sampled from just one population.
Display 10.63 Sample table for blue jeans data, collected from
one population.
Class: We use a test of independence because thats the only other

possibility.
Jodain: We use a test of independence because there is only one sample,
from the population of all people. We categorize each person in
the sample on the two variables, age and jeans-wearing behavior.
The columns are roughly proportional, so well conclude that we
cant reject the hypothesis that the two variables are independent.
Jeans-wearing behavior and age arent associated. But this time we
didnt predetermine how many people would be in each age group.
Ms. C: Right again, Jodain.
Class: Jodain is always right!
Ms. C: Also, it now makes sense to talk about conditional distributions for
the rows. You can estimate that , or about 58%, of people now
wearing jeans are under age 40 and about 42% are age 40 or older.
That statement wouldnt make sense in the homogeneity case.
Jodain: Then why wouldnt we always design our study as a test of
independence? Then we can look at the table both ways, and
as a bonus we get an estimate of the percentage of people in each
age group!
Ms. C: It depends on what you want to find out. Suppose your research
hypothesis is that the proportion of people who wear jeans differs
for the two age groups. You would design the study as a test of
homogeneity, taking an equal sample size from each age group.
721
22-03-2009 21:23
Lesson
12 de 17
That gives you the best chance of rejecting a false null hypothesis. In
short, if you want to compare populations, take independent random
samples from each and conduct a test of homogeneity. If you want
to describe association between two variables in a single population,
take a single random sample and conduct a test of independence.
Class: How old are you, Ms. C?

D25. Suppose you want to study whether people with straight hair and people
with curly hair are equally likely to use blow-dryers. You are considering two
designs for your survey:
I. You take a random sample of 50 people with straight hair and ask
them whether they use a blow-dryer. You take a random sample of
50 people with curly hair and ask them whether they use a blow-dryer.
II. You take a random sample of 100 people, note if they have curly hair
or straight hair, and ask them whether they use a blow-dryer.
a. Which of the two designs would result in a test of homogeneity, and
which would result in a test of independence?
b. How are the data from the samples likely to be different?
c. Is there any reason why one design would be better than the other?
d. Which of the two designs could also be correctly analyzed by using a
z-test for the difference between two proportions?
Strength of Association and Sample Size

associated = not
independent
Evidence of association
is not the same as a
measure of strength.
A chi-square test of independence tells you if there is evidence of an association

between two categorical variables. As is typical of a test of significance, it does
not tell you about the size of any difference in the proportions from column to
column (the strength of the association) or about whether these differences are
of practical importance. As an illustration, Display 10.64 shows three separate
samples of responses on the same two variables.
Display 10.64 Tables showing how sample size can affect the test
statistic.
22-03-2009 21:23
Lesson
13 de 17
Statistical significance is
not the same as practical
significance.
In each sample, the columns are almost proportionalthe difference between

the proportions in the first and second columns is only 0.04. This is not a very strong
association; in most applications, a difference of 0.04 between two proportions
would be of little practical significance. However, if you do a chi-square test of
independence on each of the three samples, you will see that as the sample size
increases, the test statistic gets larger and the P-value gets smaller. At a sample size
of 4000, the association between the variables becomes highly significant statistically,
but the strength of the association is no larger than for the smaller samples. For
fixed conditional distributions, the numerical value of the statistic is directly
proportional to the sample size.
Two bits of advice are in order: Be wary when using this test (or any test) with
very large data sets, and always compute and compare proportions to see if the
statistical significance has any practical value.

D26. As in all hypothesis tests, the P-value measures the strength of the evidence
against the null hypothesis (or, in favor of the alternative hypothesis)
provided by the data. If the P-value is very small, there is much evidence
in favor of the alternative hypothesis.
a. In the context of a chi-square test of independence, discuss the difference
between the statements There is much evidence in favor of association
and There is evidence of a strong association.
b. Suppose you have three 2 2 tables of frequencies for two categorical
variables and a chi-square test of independence gives these results:
Table A: very small P-value and a small difference between conditional
proportions
Table B: very small P-value and a large difference between conditional
proportions
Table C: very large P-value and a small difference between conditional
proportions
For Table A, is there much or little evidence of association? Is there
evidence of strong association or evidence of weak association? Answer
these questions for Table B and Table C.
Summary 10.3: The Chi-Square Test of Independence

Suppose there is no reason to suspect anything other than complete independence
between two variables in a population. It would be a near-miracle to get a random
sample in which the variables satisfy the definition of independence in Chapter 5.
Requiring P(A and B) to be exactly equal to P(A) P(B) for every cell of the table
is too rigid a position to take when dealing with random samples. A chi-square test
of independence is much more useful: Is it reasonable to assume that the random
723
22-03-2009 21:23
Lesson
14 de 17
sample came from a population in which the two variables are independent? Use
the chi-square test of independence when
a simple random sample of a fixed size is taken from one large population
each outcome can be classified into one cell according to its category on one
variable and its category on a second variable
you want to know if its plausible that this sample came from a population in
which these two categorical variables are independent
In a chi-square test of independence, the expected frequencies are calculated from
the sample data:
E=
(row total) (column total)

______________________
grand total
Each value of E should be 5 or greater.

The number of degrees of freedom for a two-way table with r rows and
c columns is df = (r 1) (c 1).
If the result from your sample is very different from the result expected under
independence, then will be large and you will reject the hypothesis that the
variables are independent. However, even if the test tells you there is evidence of
an association (dependence), it does not tell you anything about the strength of
the association or whether it is of practical importance.
Practice
P22. Which pairs of variables do you believe
are independent in the population of U.S.
students? Explain.
A. hair color and eye color
B. type of music preferred and ethnicity
C. gender and color of shirt
A.
B.
D. type of movie preferred and gender

E. eye color and class year
F. class year and whether taking statistics
Tabular and Graphical Display of Data
P23. Which column charts in Display 10.65
display variables that are independent?
C.
D.
Display 10.65 Four column charts.
22-03-2009 21:23
Lesson
15 de 17
Expected Frequencies in a Chi-Square Test

of Independence
P24. Display 10.66 shows the marginal totals for
a sample of 2237 residents of the United
States classified according to gender and
handedness. On a copy of the table, fill in
the expected frequencies for each cell under
the hypothesis that gender and handedness
are independent variables. Do you think
these variables really are independent?

P26. A Time magazine poll asked a random
sample of people from across America this
question: How much effort are you making
to eat a healthy and nutritionally balanced
diet? The results are given in Display 10.68.
Display 10.68 Results of a poll on eating habits.

[Source: Time, January 30, 1995.]
Display 10.66 Table of expected frequencies for
a test of independence between
gender and handedness.
Procedure for a Chi-Square Test of

Independence
P25. Is there a difference between handedness
patterns in men and women? A good set of
data to help you answer this question comes
from the governments 5-year Health and
Nutrition Survey (HANES) of 197680,
which recorded the gender and handedness
of a random sample of 2237 individuals from
across the country. The observed frequencies
for men and women in each of three
handedness categories are shown in Display
10.67. Use an appropriate statistical test to
answer the question posed. If the patterns
differ significantly, explain where the main
differences occur.
Display 10.67 Table of observed frequencies of

gender and handedness. [Source: D.
Freedman, R. Pisani, and R. Purves, Statistics,
3d ed. (New York: W. W. Norton, 1998), p. 537.]
a. Explain what the percentages measure.

Given the way the data were collected,
would you do a test of homogeneity or a
test of independence?
b. Suppose the sample size was 1000 and it
just happened to be equally split between
men and women. Is there evidence of
dependence between gender and an effort
toward a healthy diet?
c. Suppose the sample size was only
500 and it just happened to be equally
split between men and women. Is there
evidence of dependence between gender
and an effort to eat a healthy diet?
P27. A health science teacher had his class do a
sample survey of the students at a nearby
middle school. Two of the questions were Do
you eat breakfast at least three times a week?
and Do you think your diet is healthy? The
responses are shown in Display 10.69.
Do the answers to these two questions
appear to be independent, based on these
data? Conduct a statistical test.
Display 10.69 Table of responses from middle school

students regarding eating habits.
725
22-03-2009 21:23
Lesson
16 de 17

P28. The American Community Survey is a
nationwide survey that collects and analyzes
data at the community level. Eventually
it will replace the long form used in the
census conducted every 10 years. In the
2004 survey, about 322,400 interviews were
conducted with adults over age 25.
a. In that 2004 survey, 80.1% of the men
had at least a high school education, and
80.7% of the women had at least a high
school education. A two-way table for
the results is given in Display 10.70. Is
there a significant association between
gender of the respondent and attainment
of a high school education? If so, is this
evidence of a strong association, or strong
evidence of a weak association?
Display 10.70 Results of a survey regarding the

education level of 322,400 adults in
2004. [Source: www.census.gov.]
b. In the same survey, 26.1% of the men had

a bachelors degree or higher, while 22.8%
of the women had a bachelors degree or
higher. Is the association between college
degree and gender stronger than that
between high school degree and gender?
c. In large sample surveys like the American
Community Survey, almost any small
difference can show up as being
statistically significant but not necessarily
of practical importance. Would you call
either of the differences seen in parts a and
b practically important? Why or why not?
Exercises
on the age of a mother at the birth of her first
Always include all steps when doing a statistical test.
child and whether the mother eventually
E33. According to a National Center for
developed breast cancer. Of the 6168 mothers
Education Statistics report, 48,574,000
in the sample, 26.39% had their first child at
children were projected to be enrolled in the
age 25 or older. Of the 6168 mothers in the
nations K12 public schools in 2006. About
sample, 98.44% had not developed breast
69% were projected to be enrolled in grades
cancer. [Source: Jessica Utts, Seeing Through Statistics,
K8 and 31% in grades 912. It was also
2d ed. (Pacific Grove, Calif.: Duxbury, 1999), p. 209.]
projected that about 16.8% of the students
a. For what reason might the variables
would be in the Northeast, 22.1% in the
age at birth of first child and whether
Midwest, 36.5% in the South, and 24.6% in
developed breast cancer be associated?
the West. [Source: nces.ed.gov.]
b.
Construct a two-way table showing
a. For what reason might the variables
the
percentage of mothers who fall
region of country and grade level be
into each cell under the assumption of
associated?
independence.
b. Construct a two-way table showing
c.
Construct a two-way table showing the
the proportion of students who fall
number
of mothers who fall into each cell
into each cell under the assumption of
under the assumption of independence.
independence.
E35. A student surveyed a random sample of 300
c. Construct a two-way table showing the
students in her large college and collected
number of students who fall into each cell
the data in Display 10.71 on the variables
under the assumption of independence.
class year and favorite team sport.
E34. The first U.S. National Health and Nutrition
Examination Survey in the 1980s reported
22-03-2009 21:23
Lesson
17 de 17
Display 10.71 Results of a survey of 300 college students.
a. Perform a chi-square test to determine if

the variables class year and favorite team
sport are independent.
b. Could you have made a Type I error?
c. Describe how the survey should have
been designed in order to test for
homogeneous proportions among the
populations of freshmen, sophomores,
juniors, and seniors with respect to
favorite team sports.
E36. In combined polls of 3026 adults taken over
a 3-year period, the Gallup Organization got
the results in Display 10.72 regarding the
level of exercise attained regularly by adult
men and women. (Assume the respondents
are equally split between men and women.)
Display 10.72 Results of a survey regarding the

exercise level of 3026 adults, taken
over a 3-year period. [Source: poll.gallup
.com, 2005.]
a. Change the percentages into observed

frequencies and perform a chi-square test
to determine if exercise and gender are
independent.
b. Could you have made a Type I error?
c. Describe how the survey should have been
designed in order to test for homogeneity
of male and female populations with
regard to level of exercise.
E37. Display 10.73 is an example of a single

population, Titanic passengers, that has been
sorted by two variables: gender and survival
status. These data cannot reasonably be
considered a random sample from any
well-defined population.
Display 10.73 Titanic passengers sorted by gender

and survival status.
a. Construct a plot that displays these data

to see whether the variables gender and
survival status appear to be independent.
b. Test to see if the association between
the variables gender and survival status
can reasonably be attributed to chance
or if you should look for some other
explanation.
E38. The fate of the members of the Donner party,
who were trapped in the Sierra Nevada
mountain range over the winter of 184647,
is shown in Display 10.74. These data cannot
reasonably be considered a random sample
from any well-defined population.
Display 10.74 The Donner party survival data.

[Source: www.utahcrossroads.org, June 2002.]
727
22-03-2009 21:23
Lesson
1 de 8
a. Construct a plot that displays these data

to see whether the variables gender and
survival status appear to be independent.
b. Test to see if the association between the
variables gender and survival status can
reasonably be attributed to chance or if you
should look for some other explanation.
E39. Are people involved in alcohol-related
accidents on public roadways in the United
States just as likely to survive on Super Bowl
Sunday as on other Sundays? To help answer
this question, investigators looked at data for
alcohol-related crashes during the 4 hours
after the telecast of the first 27 Super Bowls.
They compared these data with the data
for alcohol-related crashes during the same
time period on the Sundays the week before
and immediately after the Super Bowl
Sundays. Display 10.75 gives the number of
people who were killed and who survived
alcohol-related crashes for the 27 Super Bowl
Sundays and the 54 control Sundays.
Display 10.75 Alcohol-related trafic fatalities on

Super Bowl Sundays and control
Sundays. [Source: D. A. Redelmeier and C. L.
Stewart, Do Fatal Crashes Increase Following a
Super Bowl Telecast? Chance 18, no. 1 (2005):
1924.]

b. Does the comparison of the survival
rates on the two types of Sundays call
for a test of homogeneity or a test of
independence? Defend your choice.
c. Is there a statistically significant
association between the type of Sunday
and whether a person survived an
alcohol-related accident? State your
conclusion carefully, in light of parts a
and b.
E40. Display 10.76 gives the smoking behavior by
occupation type for a sample of white males
in 1976. Assuming this can be considered
a random sample of all white males in the
United States in 1976, perform a chi-square
test of independence to see if it is reasonable
to assume that the variables occupation type
and smoking behavior are independent.
Display 10.76 Smoking behavior sorted by type

of occupation. [Source: D. J. Hand et al.,
A Handbook of Small Data Sets (London:
Chapman & Hall, 1994), p. 284. Original source:
T. D. Sterling and J. J. Weinkam, Smoking
Characteristics by Type of Employment, Journal
of Occupational Medicine 18 (1976): 74353.]
Chapter Summary
In this chapter, you have learned about three chi-square tests: a test of goodness
of fit, a test of homogeneity, and a test of independence. Although each test is
conducted in exactly the same way, the questions they answer are different.
In a chi-square goodness-of-fit test, you ask Does this look like a random
sample from a population in which the proportions that fall into these categories
are the same as the proportions hypothesized? This test is an extension of the test
of a single proportion developed for the binomial case.
In a chi-square test of homogeneity, you ask Do these samples from different
populations look like random samples from populations in which the proportions
22-03-2009 21:24
Lesson
2 de 8
that fall into these categories are equal? This test is an extension of the test for the
equality of two binomial proportions.
In a chi-square test of independence, you ask Does this sample look like a
random sample from a population in which these two categorical variables are
independent (not associated)? This test is not equivalent to any test developed
earlier in this book. It is, however, related to the concept of independent events
and the computation of P(A and B).
Review Exercises
E41. This question was asked of random samples
of about 1000 residents of the United States
in each of a succession of years: Which of
the following statements reflects your view of
when the effects of global warming will begin
to happen? The percentages selecting each
choice, by year, are given in Display 10.77 (at
the bottom of the page). Is there evidence of
change in the pattern in these percentages
across the years? If so, describe the pattern of
change you see.
E42. Women are more critical of environmental
conditions than are men. So says a Gallup
Poll report on a survey of 1004 adults taken
in 2005. Results are shown in Display 10.78.
The percentages in the first row of data show
that 31% of men reported a positive view of
current environmental conditions whereas
only 18% of women reported such a view.
Assume that there were approximately equal
numbers of men and women.
a. Construct a plot that displays these data.
b. What test should you use to decide
whether the data support the quoted
claim?
c. Perform the test you think is best.
Display 10.78 Results of a poll of 1004 adults

regarding how critical they are of
environmental conditions. [Source:
poll.gallup.com, 2005.]
E43. Many years ago, Smith College, a residential

college, switched to an unusual academic
schedule that made it fairly easy for students
to take most or all of their classes on the
first three days of the week. (Smith has long
since abandoned this experiment.) At the
time, the infirmary staff wanted to know
about after-hours use of the infirmary under
this schedule. They gathered data for an
entire academic year, recording the time
and day of the week of each after-hours
visit, along with the nature of the problem,
Display 10.77 Result of a poll of 1000 U.S. residents each year, asking when they expect to
be affected by global warming. [Source: poll.gallup.com, 2005.]
Chapter Summary
729
22-03-2009 21:24
Lesson
3 de 8
which they later classified as belonging

to one of four categories: A visit to the
infirmary was unnecessary for the problem,
the problem required a nurses attention,
the problem required a doctors attention, the
problem required admission to the infirmary
or to the local hospital. The results appear in
Display 10.79.
to your children or the next generation of

Americans in their lifetime? The results are
Display 10.80 Results of Gallup poll on the threat

of global warming. [Source: www
.americans-world.org, June 2002.]
Display 10.79 After-hours use of the infirmary.

[Source: George W. Cobb, Design and Analysis
of Experiments (New York: Springer, 1998),
pp. 23940.]
a. Construct a plot to display these data.

b. Are there any interesting trends in the
table? What might explain them?
c. Could you use a test of independence on
these data? A test of homogeneity? Does
the design of the study fit the conditions
of these tests?
d. Perform the test you think is best, giving
any necessary cautions about your
conclusion.
e. How might you group the data to see
whether the data for the weekdays when
students attend class differ from the data
for the weekdays when students dont
attend class?
f. Perform a chi-square test on your
regrouped data.
E44. A 2002 Gallup poll asked these questions
of a randomly selected national sample of
approximately 1000 adults, 18 years and
older: Do you think that global warming
will pose a serious threat to you or your way
of life in your lifetime? Do you think that
global warming will pose a serious threat
a. Does this situation call for a chi-square

test of homogeneity, a chi-square test of
independence, or neither?
b. Perform the test you selected, or explain
why neither test is appropriate.
E45. Vehicles can turn right, turn left, or continue
straight ahead at a given intersection. It
is hypothesized that vehicles entering the
intersection from the south will continue
straight ahead 50% of the time and that
a vehicle not continuing straight ahead is
just as likely to turn left as to turn right. A
sample of 50 vehicles gave the counts shown
in Display 10.81. Are the data consistent
with the hypothesis?
Display 10.81 Vehicle turning data.
E46. In testing a random-digit generator, one

criterion is to be sure that each digit
occurs of the time.
a. Generate 100 random digits on your
calculator, and use a chi-square test to
see if its reasonable to assume that this
criterion is met.
b. This criterion isnt sufficient. Give an
example of a sequence of digits that
clearly is not random but in which each
digit occurs of the time. How might
you test against this possibility?
E47. In a random sample of 100 college graduates,
100 high school graduates, and 100 people
22-03-2009 21:24
Lesson
4 de 8
who didnt graduate from high school, 40%

b. Enter a set of possible observed
of the college graduates watched the last
frequencies into a copy of the table. Pick
Super Bowl, as did 49% of the high school
observed values so that the variables
graduates and 37% of the people who
clearly arent independent.
didnt graduate from high school. Is there
c. Can you enter a set of possible observed
a relationship between watching the Super
frequencies, different from the expected
Bowl and a persons educational level?
frequencies, into a copy of the table
a. Does the design of this study suggest a
such that the variables clearly are
test of goodness of fit, homogeneity, or
independent?
independence?
E50. Foresters often are interested in the patterns
b. Perform the test you selected in part a.
of trees in a forest. For example, they gauge
the integration or segregation of two species
E48. The data in Display 10.82 show how a random
of trees (say, I and II) by looking at a sample
sample of people describe their political
of one species and then observing the
ideology and whether they usually think of
species of its nearest neighbor. Data from
themselves as Republicans, Independents, or
such a study can be arrayed in a two-way
Democrats. Does political ideology appear
table such as this one.
to be independent of party affiliation? If
not, explain something of the nature of the
dependency.
Display 10.82 Results from the General Social

Survey of the United States for the
year 2000. [Source: www.icpsr.umich.edu,
June 2002.]
E49. Display 10.83 shows a two-way table for a

sample of size 180.
Display 10.83 Two-way data table for sample of

size 180.
a. Assuming that the two variables are

independent, find the expected frequency
for each cell.
a. Which cell counts, among A, B, C, and D,

would be large if segregation of species
is high? Would this lead to a large or
small value of the statistic in a test of
independence?
b. Which cell counts, among A, B, C, and D,
would be large if segregation of species
is low? Would this lead to a large or
small value of the statistic in a test of
independence?
E51. Display 10.84 (on the next page) gives
information about the scores on the 1996
and 2001 AP Calculus AB Exams. The data
for 1996 were obtained from all students
who took the exam. The data for 2001 were
obtained from a random sample of 200
students.
a. Suppose you want to investigate whether
these data provide evidence of a change
in the distribution of grades from 1996 to
2001. Is a chi-square test appropriate? If
so, perform the test you selected, showing
all steps. If not, explain why not.
Chapter Summary 731
22-03-2009 21:24
Lesson
5 de 8
Display 10.84 Data and summary statistics for 1996 and 2001 AP Calculus AB Exams.
[Source: For the 1996 data: AP Calculus Course Description, May 1998,
May 1999, p. 76. For the 2001 data: apcentral.collegeboard.com, June 2002.]
b. Suppose you want to investigate whether

these data provide evidence of an increase
in the mean grade from 1996 to 2001.
What test is appropriate? Perform this
test, showing all steps.
E52. A common Gallup poll question often
asked of a sample of residents of the United
States is this: In general, are you satisfied
or dissatisfied with the way things are going
in the United States at this time? Display
10.85 shows the results of five such polls
conducted from 2001 to 2005.
c. Suppose each poll had sampled about

500 residents. Is there evidence that the
level of satisfaction changed significantly
across these years? If so, describe the
pattern of change.
E53. The data in Display 10.86 show the Titanic
passengers sorted by two variables: class of
travel and survival status. These data cannot
reasonably be considered a random sample.
They are the population itself, so you must
be careful in stating the hypotheses and the
conclusion for any test of significance. Test
to see if the apparent lack of independence
between the variables class of travel and
survival status can reasonably be attributed
to chance or if you should look for some
other explanation. If the latter, what is
that explanation?
Display 10.85 Results of five polls taken from 2001

to 2005. [Source: poll.gallup.com, 2005.]
a. Why dont the row percentages sum to 1?

b. Suppose each poll had sampled about
1000 residents (about the size of most
Gallup polls). Is there evidence that the
level of satisfaction changed significantly
across these years? If so, describe the
pattern of change.
Display 10.86 Titanic passengers sorted by class of

travel and survival status.
22-03-2009 21:24
Lesson
6 de 8
E54. In 1988, there were 540 spouse-murder

cases in the 75 largest counties in the United
States. Display 10.87 gives the outcomes.
Display 10.87 Outcomes for spouse-murder

defendants in large urban counties.
[Source: Bureau of Justice Statistics, Spouse
Murder Defendants in Large Urban Counties,
Executive Summary, NCJ-156831 (September
1995).]
a. What are the populations?

b. Construct a suitable plot to display the
data. Describe what the plot shows.
c. Write suitable hypotheses and perform a
chi-square test.
E55. In this chapter, you have seen three
chi-square tests. Match each test (ac)
with its description (AD) and its design
(IIV). (One description and one design
will be left over.)
a. chi-square test of goodness of fit
b. chi-square test of homogeneity
c. chi-square test of independence
Description:
A. You test that two populations have
equal proportions of members that fall
into each of a given set of categories.
B. You test that a population has the
same proportion that falls into each
of a given set of categories as some
hypothesized distribution.
C. You test that you can predict the result
for one categorical variable better if
you know the result for the other.
D. You test that two populations are
independent.
Design:
I. one sample from one population
sorted according to one categorical
variable
II. one sample from one population
sorted according to two categorical
variables
III. two samples from two populations
sorted according to one categorical
variable
IV. two samples from two populations
sorted according to two categorical
variables
Chapter Summary 733
22-03-2009 21:24
Lesson
7 de 8
AP1.This partially completed table of expected

frequencies is for a test of the independence
of a fathers handedness and his oldest
childs handedness. What is the expected
frequency in the cell marked ??
cannot be determined from the

information given
AP2. In a pre-election poll, 13,660 potential
voters were categorized based on their
annual income (into one of eight categories)
and their intended vote (two categories,
Republican or Democratic candidate). The
resulting chi-square test of independence
of these variables gave a test statistic of 270.
How many degrees of freedom are there?
2
7
16
269
13659
AP3. Researchers interested in determining
whether there is a relationship between
a mothers birthday and the birthday of
her oldest child took a random sample of
200 mothers. The mothers birthday was
categorized as within a week, more than a
week but less than a month, or more than
a month from her oldest childs birthday.
Assuming that there are 365 days in a year
and the same number of people are born on
each day, what is the expected frequency of
mothers in the within at most one week
category?
AP4. In a study of whether two acne medications,
are equally effective, researchers got
eight volunteers and randomly chose one
side of each persons face to receive each
medication. After one month of use, they
counted the number of pimples on each side

of each persons face. Which test is most
appropriate in this situation?
matched-pairs t-test
two-sample t-test
chi-square test for goodness of fit
chi-square test for homogeneity of
proportions
chi-square test of independence
AP5. A random sample of 1000 adults from each
of the 50 states is taken, and the people
are categorized as to whether they have
graduated from college or not, generating a
50-by-2 table of counts. Which is the most
appropriate test for determining whether
the graduation rates differ among the states?
Compare each count to the expected
frequency of 500 and use a chi-square
goodness of fit test.
Use a chi-square test of homogeneity of
proportions.
Use a chi-square test of independence.
Use a one-proportion z-test.
Use a t-test for the difference of means.
AP6. For a project, a student plans to roll a
six-sided die 1000 times. The teacher
becomes suspicious when the student
reports getting 168 ones, 165 twos, 170
threes, 167 fours, 164 fives, and 166 sixes.
The teacher performs a chi-square test with
the alternative hypothesis that the students
reported observed counts are closer to the
expected frequencies than can reasonably
be attributed to chance. What is the test
statistic and P-value for this test?
= 0.14; P-value 0.0004
= 0.14; P-value 0.9996
= 1000; P-value 0
= 23.33; P-value 0.0004
= 23.33; P-value 0.9996
AP7. A researcher wanted to determine whether
Barbarians and Vandals have similar food
22-03-2009 21:24
Lesson
8 de 8
preferences. Independent random samples

were taken from each tribe and each person
was categorized as to whether he or she
prefers to eat gruel, turnips, or raw meat.
The value of was statistically significant.
What conclusion can be drawn?
The proportion of all Barbarians
who prefer gruel is different from the
proportion of all Vandals who prefer
gruel.
The proportion of all Barbarians who
prefer gruel is equal to the proportion of
all Vandals who prefer gruel.
The population proportions are different
for at least one of the three choices.
The population proportions are different
for all three of the choices.
The population proportions are equal for
all three of the choices.
AP8. Random samples are taken from two
populations (Barbarians and Vandals) and
each person is categorized as a pillager or a
burner. Which of the following, when done
appropriately, is equivalent to a chi-square
test of homogeneity for this situation?
one-sample z-test
two-sample z-test
one-sample t-test
two-sample t-test
chi-square goodness-of-fit test
Investigative Tasks
AP9. To study the effectiveness of seat belts and
air bags, researchers took a careful look
at data that were collected on accidents
between 1997 and 2002. The data came
from a division of the National Highway
Traffic Safety Administration that collects
detailed data on a random sample of
accidents in which there is a harmful
event and from which at least one vehicle
is towed. A total of 22,804 front-seat

occupants were involved in the accidents
studied. [Source: Mary C. Meyer and Tremika Finney,
Who Wants Airbags? Chance 18, no. 2 (2005): 316. The
data here are approximations created from scaling down the
population projections in the article.]
a. Do the data in the table show a statistically

significant association between seat belt
use and surviving an accident?
b. Do the data in the table show a statistically

significant association between air bag use
and surviving an accident?
c. From the two analyses in parts a and b,

which safety feature appears to have the
stronger association with survival?
AP10. The analysis in AP9 does not account for
seat belts and air bags interacting with
each other. This table splits the air bag data
according to whether seat belts were also in
use during the accident.
a. What do the data suggest as to the effect

of air bag use when seat belts are also
used? Is there a significant association?
b. What do the data suggest about the effect
of air bag use when seat belts are not
used? Is there a significant association?
AP Sample Test 735
22-03-2009 21:24
Lesson
1 de 3
11
Inference for Regression
Do Mars rocks with

more sulfur tend
to be redder? Data
from the Pathfi nder
mission to the red
planet were used to
explore questions
like this.
22-03-2009 21:24
Lesson
2 de 3
Trivia Question 1: Who are Shark, Barnacle Bill, Half Dome, Wedge, and Yogi?
Answer: Mars rocks, found at the landing site of Mars Pathfinder in
July 1997.
Trivia Question 2: Mars is often called the red planet. What makes it red?
Answer: Sulfur?
Display 11.1 shows how the redness of the five Mars rocks is related to their sulfur
content. The response variable, redness, is the ratio of red to blue in a spectral
analysis; the higher the value, the redder the rock. The explanatory variable is the
percentage, by weight, of sulfate in the rock.
Display 11.1 Scatterplot and data table for the relationship

between the variables redness and sulfate percentage
for Mars rocks. [Source: Harry Y. McSween Jr. and Scott L. Murchie,
Rocks at the Mars Pathfinder Landing Site, American Scientist 87
(Jan.Feb. 1999): 3645. Data taken from a graph.]
22-03-2009 21:24
Lesson
3 de 3
Shark: Have you figured out how change in the sulfate content is
reflected in change in redness?
Yogi: According to my calculations, the slope of the regression line
is 0.525. Rocks that differ by 1 percentage point in their sulfate
content will differ by about 0.5, on average, in their measure of
redness. I learned how to do this in Chapter 3.
Shark: Thats the story if you take the numbers at face value.
Yogi: Why shouldnt I?
Shark: Well, after all, we arent the only rocks on the planet. The slope,
0.525, is an estimate based on a sampleus! If that cute little
rolling Sojourner robot had nuzzled up to different rocks, you
would have a different estimate. So the goal of this chapter is to
show how to construct confidence intervals and test hypotheses
for the slope of a regression line.
Yogi: You need a whole chapter just for that? I know from Chapter 3
how to estimate the slope. So, if you tell me how to find the
standard error for the slope, I can do the rest. To get a confidence
interval, I just do the usual: 0.525 t* SE. To test a null
hypothesis, I simply compute
and use the t-table to find the P-value.

Shark: It sounds familiar, all right. Maybe this can be a short chapter!
This chapter addresses how to make inferences about the unknown true
relationship between two quantitative variables. The methods and logic you will
learn in this chapter apply to a broad range of such questions.

that the slope of a regression line fitted from sample data will vary from
sample to sample, and what things a ect this variability
how to estimate the standard error of the slope
how to construct and interpret a confidence interval for the slope
how to test whether the slope is different from a hypothesized value
how to know when to trust confidence intervals and tests
how to transform variables to make inferences more trustworthy
22-03-2009 21:24
Lesson
1 de 16
11.1
Bivariate: two variables,
usually called x and y
Variation in the Slope from Sample to Sample

As you know by now, statistical thinking, though powerful, is never as easy or
automatic as simply plugging numbers into formulas. In order to use statistical
methods appropriately, you need to understand their logic, not just the
computing rules. The logic in this chapter is designed for bivariate populations
that you can think of as modeled by linear fit plus random deviation. That is,
all points dont lie exactly on the line of best fit but cluster around it, forming an
elliptical cloud.
Linear Models
In Chapter 3, you learned to summarize a linear relationship with a least squares
regression line,
That equation is a complete description if you have the entire population, but if
you have only a random sample, the values of b0 and b1 are estimates of the true
population parameters. That is, there is some underlying true linear relationship
that you are trying to estimate, just as you use as an estimate of . The notation
for a linear relationship is
response = prediction from true regression line + random deviation
is the observed
value minus the value
predicted by the true
regression line.
where and refer, respectively, to the intercept and slope of a line that you
dont ordinarily get to seethe true regression line you would get if you had
data for the whole population instead of only a sample. The letter e indicates
the size of the random deviationhow far a point falls above or below the
true regression line. The true regression line, sometimes called the line of means
or the line of averages, is written
Because such linear models are often used to predict unknown values of
y from known values of x, or to explain how x influences the variation in y,
y is called the response variable and x is called the predictor variable or the
explanatory variable.
Activity 11.1a is designed to help you understand the roles of these equations
and the relationship between them.
739
22-03-2009 21:24
Lesson
2 de 16
How Fast Do Kids Grow?

What youll need: a copy of Display 11.2
On average, kids from the ages of 8 to 13 grow taller at the rate of 2 in. per
year. Heights of 8-year-olds average about 51 in. At each age, the heights are
approximately normal, with a standard deviation of roughly 2.1 in. [Source:
National Health and Nutrition Examination Survey, May 30, 2000, www.cdc.gov.]
1. Use the information above to fill in the second column of your copy of the
table in Display 11.2.
Display 11.2 Partial data table for Activity 11.1a.
2. Find the equation of the true regression line,

that relates
average height, y, to age, x. Interpret and .
3. Now suppose you have a randomly selected child of each age. To find
how much the height of your 8-year-old child deviates from the average
height, use your calculator to randomly select a deviation, , from a normal
distribution with mean 0 and SD 2.1. [See Calculator Note 11A.] Record
and then add to the average height. Record the sum in the column
labeled Observed Height. Repeat, with a new deviation, for each age.
4. Fit a regression equation to the pairs (x, y) where y is the observed height.
Record your estimated slope, b1. Is it close to ?
5. Collect the values of b1 found by the members of your class and plot these
values. Discuss the shape, center, and spread of this plot of estimated
slopes.
6. The population of the deviations in step 3 had mean 0. Why?
7. Suppose that, instead of creating your data by simulation, you used actual
data by choosing one child from each age group at random and measuring
their heights. In what ways is the model in step 3 a reasonable model for
this situation? In what ways is it not so reasonable?
Yogi: You dont expect me to believe that real data get created the way
we did it in the activity, do you?
Shark: No. At best, this model is a good approximation, and its up to you
to judge how well it describes the process that created your data.
But precisely because the model is a simplified version of reality,
when its reasonable, its quite useful.
740 Chapter 11 Inference for Regression
22-03-2009 21:24
Lesson
3 de 16
This is one reason why

you can think of the
regression line as the
graph of the average
y-values.
The conditional distribution of y given x refers to all the values of y for

a fixed value of x. For example, think of the distribution of childrens heights
described in Activity 11.1a. If you were to make a scatterplot of (age, height)
for a sample of hundreds of children, the vertical column of points for all the
8-year-old children would approximate the conditional distribution of the
height, y, given that the age, x, is 8, as shown in Display 11.3. Each conditional
distribution of height for a given age has a mean, called for a population and
for a sample, and a measure of variability, called for a population and s for a
sample. A linear model is appropriate for a set of data if the conditional means fall
near a line and the variability is about the same for each conditional distribution.
Display 11.3 Scatterplot of height versus age. Conditional

distributions of childrens height given their age
are the vertical columns of dots.
Linear Models
D1. Refer to the information about childrens heights in Activity 11.1a. What
are the mean and standard deviation of the conditional distribution of the
heights given that the age is 10? Given that the age is 12?
D2. Explain the difference between the linear model
and the
fitted equation
. In particular, what is the difference between
and b1? What is the difference between a random deviation from the
model and an observed residual, y ?
The Variability of b1 from Sample to Sample

In Activity 11.1a, you actually knew the value of the theoretical slope, , so when
you found an estimate b1 based on the data from your sample, you could tell how
far that estimate was from the true value. In real situations, of course, you dont
know the true slope. You must use your data to estimate not only but also the
standard error of b1. Then, if a linear model is appropriate, you can rely on the
fact that 95% of the time the estimated slope, b1, will be within approximately two
standard errors of the true slope, .
To move toward finding a formula for the standard error of b1, think about the
situations when b1 would tend to be close to and when it would tend to be farther
away. As you might have guessed, with a larger sample size, b1 tends to be closer to
. Activity 11.1b is designed to help you see what other factors affect the variation
in b1 when the standard model (true regression line plus random variation) applies.
741
22-03-2009 21:24
Lesson
4 de 16
What Affects the Variation in b1?

In this activity, you will create four scatterplots. For each plot, there will be only
two values of the predictor, x, and you will generate four values of the response,
y, for each x. These response values will come from normal distributions with
the specified means and standard deviations.
1. For each value of x given in Cases 14, generate four values of y from a
normal distribution with the given mean and standard deviation. [See
Calculator Note 11A to review how to generate random numbers from a
normal distribution, and see Calculator Note 11B to learn how to augment
a list, which will speed up your work on this step.] Plot the resulting eight
ordered pairs for each case on a scatterplot. Fit a regression line and record
the slope.
Case 1
x = 0; conditional distribution of y has = 10 and = 3
Case 2
Case 3
Case 4
2. Collect the estimated slopes from your class. Construct four dot plots of
these estimated slopes, one plot for each case. What is the theoretical slope,
, in each case? For each case, estimate the mean of the distribution of
estimated slopes and compare it to .
3. How does the variation in the estimated slopes change with an increase
in the variation of the response variable, y? How does the variation in the
estimated slopes change with an increase in the spread of the explanatory
variable, x?
You learned in Activity 11.1b that the variation in the estimated slopes
depends not only on how much the values of y vary for each fixed value of x but
also on the spread of the x-values.
In the methods of inference in this chapter, the variability in y is assumed
to be the same for each conditional distribution. That is, if you picked a value
of x and computed the standard deviation of all the associated values of y in the
population, youd get the same number, , as you would if you picked any other
x, as shown in Display 11.4. This implies that also measures the variability of all
values of y about the true regression line. You can use this fact to estimate from
your data.
22-03-2009 21:24
Lesson
5 de 16
Display 11.4 The variability of all values of y about the true

regression line is the same as the variability of the
y-values at each fixed value of x.
Variability in x and y
Note that s
measures the
variability of the
residuals.
The common variability of y at each x is called . It is estimated by s, which can

be thought of as the standard deviation of the residuals. You compute s using
all n values of y in the sample:
The spread in the values of x is measured by

using all n values of x in the sample.
where
is computed
Display 11.5, which shows some results from Activity 11.1b, shows that the
variability in the slope depends on the sample size and the variability in x and y.
Plot I: x = 0,
=3
Plot III: x = 0,
=5
= 10; x = 1,
= 12; Plot II: x = 0,

=3
= 10; x = 1,
= 12; Plot IV: x = 0,

=5
= 10; x = 4,
= 10;
= 18;
= 18;
Display 11.5 Regression lines relating variability in b1 to variation

in y and to spread in x.
743
22-03-2009 21:24
Lesson
6 de 16
Each plot in Display 11.5 shows five regression lines for one of the four cases;
for example, Plot I shows five regression lines for Case 1. Each line was constructed
using the instructions in step 1 of the activity. Plots I and II (or III and IV) show
that a wider spread in x results in regression lines with less variability in their
slopeeven though the values of y vary equally for each x-value. By comparing
plot I with plot III (or plot II with plot IV), you can see something more expected:
More variability in each conditional distribution of y means more variability in the
slope, b1.

D3. In the scatterplots in Display 11.6, imagine taking a sample of responses
within each rectangle. That is, the rectangles define the regions in which the
responses lie for each value of x. Suppose you take a sample and calculate a
regression line. Then you repeat the process many times. Which of the three
plots should produce regression lines with the smallest variation in slope?
Which should produce regression lines with the largest variation in slope?
Display 11.6 Three scatterplots for D3.
D4. If samples were to be selected from the rectangles in Display 11.7, would
you have any concerns about using a straight line to model the relationship
between x and y? Why or why not?
Display 11.7 A scatterplot for D4.
22-03-2009 21:24
Lesson
7 de 16
The Standard Error of the Slope

You now know what affects how much the slope, b1, of the regression line varies
from sample to sample: the sample size, n; the variability, , of y at each fixed
value of x (estimated by s); and the spread in the values of x. This box gives the
formula for estimating the standard error, , of the sampling distribution of
the slope.
Formula for Estimating the Standard Error of the Slope
The formula does what you would expect: The slope varies less from sample to
sample when the sample size is larger, when the values of y tend to be closer to the
regression line, and when the values of x are more spread out.
Example: French Fries

A statistics class at the University of WisconsinStevens Point decided to study the
mass and number of fries in small and large bags of french fries at McDonalds.
They bought 32 bags of each size during two different time periods on two
consecutive days at the same McDonalds and weighed the fries. The data are
given in Display 11.8. (Note that data are missing for three of the large bags due to
various problems with the data collection.)
Study the issue of predicting the total mass of the bag of fries from the
number of fries by answering ac.
a. Plot the data and, if appropriate, fi t least squares regression lines for each
bag size. Interpret the slope for the small bags and then compare with the
slope for the large bags.
b. Study the residual plots. Do you see any pattern that suggests that a line plus
random error is a poor model?
c. Calculate the standard error of the slope for the small bags and interpret it.
Then calculate the standard error of the slope for the large bags and compare
with that for the small bags.
745
22-03-2009 21:24
Lesson
8 de 16
Display 11.8 Number and mass of french fries. [Source: Nathan Wetzel,
McDonalds French Fries. Would You Like Small or Large Fries?
STATS, 43 (Spring 2005): 1214.]
22-03-2009 21:24
Lesson
9 de 16
Solution
a. The plots, shown in Display 11.9, indicate a positive linear trend, even though
there is considerable variation around the regression lines. The slope for the
small bags means that if one bag has 1 more fry than another, you would
expect the bag of fries to weigh 0.741 gram more. In other words, a fry weighs
about 0.741 gram. The slope for the large bags is 0.321, so the estimate of the
weight of a fry is less than half that from the small bags.
Display 11.9 Regression of total mass versus number of fries

in bag.
b. The residual plots look much like randomly scattered points, although the
small fries have some potentially influential observations because of the
large residuals, while the large fries have a potentially influential observation
because of a large gap on the x-axis.
747
22-03-2009 21:24
Lesson
10 de 16
c. The slope of the least squares regression line for small bags has an estimated
standard error given by
So, if you were to take many random samples of 32 bags of small fries and
compute the slope of the regression line for predicting mass from number
of fries, the standard deviation of the distribution of these slopes would be
about 0.1656.
For the large bags, the estimated standard error is
The estimated standard error of the slope for the large bags is smaller than that
for the small bags, even though there is much more variation in the masses (y)
of the large bags. This happens because the larger variation in masses is offset
by the larger variation in the numbers of fries (x) for the large bags.

D5. You plan to collect data in order to predict the actual temperature of
your oven from the temperature shown on its thermostat. Suppose this
relationship is linear. How will you design this study? What have you learned
in this section that will help you?
D6. Use Display 11.5 on page 743 to explain why it makes sense for the
expression with the xs to be in the denominator of the formula for .
Summary 11.1: Variation in the Slope from Sample

to Sample
Suppose you want to estimate an underlying linear relationship,
where
is the equation of the true regression line. The intercept, b0,
and slope, b1, in the least squares regression equation
serve as your
estimates of the population parameters and . Typically, you will be most
interested in how closely b1 estimates . Use the values in your sample and this
formula to estimate the standard error of the slope:
22-03-2009 21:24
Lesson
11 de 16
The slope, b1, of the regression line varies less from sample to sample when
the sample size is larger
the residuals are smaller
the values of x are farther apart
In the following sections, you will learn to compute confidence intervals and
test statistics for the slope, but your conclusions wont be valid unless line plus
random deviations is the appropriate model for your data. So always plot your
data first to judge whether a line is a reasonable model.
Practice
Linear Models
P1. The scatterplot of data on pizzas in Display
3.31 on page 134 shows the number of
calories versus the number of grams of fat
in one serving of several kinds of pizza. Fat
contains 9 calories per gram.
a. What would be the theoretical slope of a
line representing such data?
b. What does the intercept of the line tell you?
c. What are some reasons why not all of the
points fall exactly on a line with the slope
in part a?
P2. According to Leonardo da Vinci, a persons
arm span and height are about equal.
Display 11.10 gives height and arm span
measurements for a sample of 15 high school
students.
Display 11.10 Height versus arm span for 15 high

school students.
a. What is the theoretical regression line,

, that Leonardo is
proposing for this situation? Use
arm span as the explanatory variable.
b. Find the least squares regression line,

= b0 + b1x, for these data. Interpret
the slope. Compare the estimated slope
and intercept to the theoretical slope and
intercept in part a. Are they close?
c. Calculate the random deviation, ,
for each student, using the theoretical
regression line. Plot these random
deviations against the arm spans and
comment on the pattern.
d. For each student, calculate the residual
from the estimated regression line. Plot
the residuals against the arm spans and
comment on the pattern. Is the pattern
similar to that for the random deviations?
P3. Every spring, visitors eagerly await the
opening of the spectacular Going-to-the-Sun
Road in Glacier National Park, Montana. A
typical range of yearly snowfall for the area is
from 30 to 70 inches. The amount of snow is
measured at Flattop Mountain, near the top
of the road, on the first Monday in April and
is given in swe (snow water equivalent: the
water content obtained from melting). From
analysis of past data, when the amount of
snow was 30 inches of swe, the road opened,
on average, on the 150th day of the year.
Every additional 0.57 in. of swe measured
at Flattop Mountain meant another day
on average until the road opened. [Source:
Spring Opening of the Going-to-the-Sun Road and Flattop
Mountain SNOTEL Data, Northern Rocky Mountain Science
Center, U.S. Geological Survey, December 2001, www.nrmsc
.usgs.gov.]
a. If you write an equation summarizing

the given information, should you use
749
22-03-2009 21:24
Lesson
12 de 16
the form
or
Explain.
b. Write an equation that predicts when the
road will be open, given the swe. In this
situation, what is the response variable, y?
The predictor variable, x?
c. In 2005, the Flattop Mountain station
recorded 31.0 inches of swe. What date
would you predict the road opened?
d. Do you think the random deviation, ,
in this situation tends to be relatively
large or small? Make a guess as to what it
might be on average.
P4. Refer to the table in Display 11.1 on page
737, which gives the sulfate percentage, x,
and redness, y, of the Mars rocks. Compute s,
the estimate of the common variability of y
at each x.
P5. Five quantities are listed here, along with two
possible values of each. For each quantity,
decide which value will give you the larger
variability in b1 (assuming all other things
stay the same), and give a reason why.
a. the standard deviation, , of the
individual response values of y at each
value of x: 3 or 5
b. the spread of the x-values: 3 or 10
c. the number of observations, n: 10 or 20
d. the true slope, : 1 or 3
e. the true intercept, : 1 or 7
P6. Look again at the Mars rocks data in
a. Compute the estimate of the standard
error of the slope, .
b. On the computer output for the
regression in Display 11.11, locate the
standard error of the slope. Then locate
your estimate from P4 of the variation in
y about the line. What is the equation of
the regression line?
Display 11.11 Regression analysis for the Mars

rocks data.
c. If the values in the Sulfate row were

missing in Display 11.11, explain how
you would compute Coefficient and
s.e. of Coeff.
P7. Using the same underlying distribution
as in Plot II of Display 11.5 on page 743,
ten observations of y were generated at
each value of x. The results are given in
Display 11.12.
Display 11.12 Data and scatterplot for P7.
a. Find both the true regression line,

and the least squares
and
regression line,
compare them.
22-03-2009 21:24
Lesson
13 de 16
b. Compute s and compare it to = 3.

c. Compute sb1, the estimate of the standard
error of b1, and compare it to the true
value of the standard error. (Note: The
true value of the SE for b1 is
d. Does the least squares regression line
go through the mean of the responses at
both x = 0 and x = 4?
P8. Each of the lists in IV gives the values of x
used to compute a regression line. Assuming
= 1 in all cases, order the lists from the
one with the largest standard error of the
slope to the one with the smallest. (You
should be able to rank these without doing
much, if any, computation.)
I. x = 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12
II. x = 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
III. x = 1, 1, 1, 1, 1, 1, 12, 12, 12, 12, 12, 12
IV. x = 1, 1, 1, 4, 4, 4, 9, 9, 9, 12, 12, 12
V. x = 1, 2, 3, 4, 5, 6, 7
P9. As you did in P8, order these lists from the
largest to the smallest value of the true SE of
the slope. (For these, you might need to use
the formula and do some computation.)
I. = 2; x = 1, 1, 1, 1, 3, 3, 3, 3
II. = 1; x = 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3
III. = 3; x = 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6
IV. = 1; x = 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6
Exercises
E1. Display 11.13 gives some information about
a random sample of motor vehicle models
commonly sold in the United States.
Display 11.13 Table and scatterplots of car models

data. [Source: Journal of Statistics Education
Data Archives, June 2002.]
The variables are:

price typical selling price, in thousands of
dollars
mpg typical highway mileage, in miles per
gallon
liter size of the engine, in liters
horsepower rating of the vehicle
hp
rpm maximum revolutions per minute
the engine is designed to produce
751
22-03-2009 21:24
Lesson
14 de 16
Consider these pairs of variables:

x = hp, y = price
x = hp, y = mpg
x = liter, y = mpg
x = rpm, y = mpg
a. Refer to the scatterplot for each pair of
variables and tell whether you think a
line gives a suitable summary of their
relationship.
b. Compute the slope of the regression line
for x = hp and y = mpg. Interpret this
slope in context. Compute the estimated
standard error of the slope.
c. Find your values from part c on the
computer printout in Display 11.14. Also
find and interpret s, the estimate of .
Bivariate Fit of Miles per Gallon by Horsepower
E2. How does energy consumption in a

residence relate to the outdoor temperature?
Some interesting patterns can be detected
through study of available monthly records
on natural gas and electricity usage for a
gas-heated single-family residence (with no
air conditioning) in the Boston area, along
with monthly data on outdoor temperature
and heating degree days. Heating degrees
for a day with mean temperature less than
65F are defined as (65 mean daily
temperature). The data are given in Display
11.15, with these variables:
Mean Temp mean monthly temperature in
Boston, in degrees Fahrenheit
Mean Gas mean natural gas usage per
day for the month, in therms
Mean KWH mean electricity usage per
day for the month, in
kilowatt-hours
total heating degree days for
Heat DD
the month
Display 11.14 Regression analysis for the car

models problem.
(continued)
22-03-2009 21:24
Lesson
15 de 16
(continued)
a. Refer to the scatterplots in Display 11.15.

For each pair of variables, tell whether
you think a line gives a suitable summary
of the relationship.
b. By looking at the scatterplots, estimate
which of the four pairs of variables has
the largest standard error of the slope and
which has the smallest.
c. Compute the slope of the least squares
regression line for the relationship
between y = Mean Gas and x = Mean
Temp. Interpret the slope in context.
Compute the estimated standard error
of the slope.
d. Find your values from part c on the
computer printout in Display 11.16. Also
find and interpret s, the estimate of .
Display 11.16
Display 11.15 Table and scatterplots of household
energy data. [Source: R. Carver, What
Does It Take to Heat a New Room? Estimating
Utility Demand in a Home, Journal of Statistics
Education 6, no.1 (1998).]
Regression analysis for Mean Gas

versus Mean Temp.
E3. Suppose you collect a dozen cups, glasses,

round bowls, and plates of various sizes
and then plot the distance around the rim
versus the distance across, both measured in
centimeters.
753
22-03-2009 21:24
Lesson
16 de 16
a. What is the theoretical slope of a line

fit to such data? Interpret this slope in
context.
b. What are some reasons why not all of the
points fall exactly on a line with the slope
in part a?
E4. Suppose you have a collection of equilateral
triangles of different sizes, as in the
structural steel of a bridge support. You have
both the measured areas and the measured
side lengths of these triangles, and you plot
the area inside the triangle against the square
of the length of a side.
Display 11.17 Sulfate percentage and redness of six

Mars soil samples.
a. What is the theoretical slope of a line fit

to such data? Interpret the slope in this
context.
b. What are some reasons why some of the
points may not fall exactly on a line with
the slope in part a?
E5. Display 11.17 gives the sulfate content and
redness of six soil samples from Mars.
a. Compare the plot in Display 11.17 to the
one for the Mars rocks in Display 11.1
on page 737. Which value of do you
expect to be largerthe one for the
five rocks, or the one for the six soil
samples? Why?
b. Compute for the soil samples, and
check your conjecture from part a. (You
computed for the rocks in P6.)
E6. In the Mars soil sample data of Display 11.17,

one of the points appears to be an outlier.
Suppose that point is removed.
a. Suppose a line is fit to the remaining five
points. Do you expect the estimated SE of
the slope to be larger for the original data
or for the revised data?
b. Compute using the remaining five
points, and check your conjecture from
part a. (You computed for all size
points in E5.)
E7. In Activity 11.1a about the heights of
children ages 8 to 13, the response variable
is height, the predictor variable is age, and
the average height of 8-year-olds is 51 in.
Begin by assuming that, on average children
grow 2 in. per year and that, at any one age,
the distribution of heights is approximately
normal. Sketch a scatterplot that shows the
sort of data you would expect to get in step 3
of Activity 11.1a if you used these standard
deviations instead of 2.1.
a. = 3
b. = 1
22-03-2009 21:24
Lesson
1 de 16
E9. Suppose you were to create a model for the

c. This time, use a different value of for
Mars rocks data in Display 11.1 on page
different ages: age 8, = 1 in.; age 9,
737 using the same approach as in Activity
= 2 in.; age 10, = 3 in.; . . . ; age 13,
11.1a. Write out directions telling how to use
= 6 in.
the model to find the redness of a rock that
E8. Refer to E7. Sketch a scatterplot that shows
has the same sulfur content as Half Dome.
the sort of data you would expect to get in
(To do this, you will need to make some
step 3 of Activity 11.1a if you used these new
assumptions about things you dont know.)
theoretical models.
In what ways do you consider your model a
a. With = 3, assume that children grow
reasonable one for the Mars rocks data? In
faster and faster as they get older so that
what ways is it not so reasonable?
the average height at age x is given by
E10. As in P8, order these lists according to the
50 + (x 8) 2.
size of the standard error of the slope. (No
b. Keep everything the same as in part a,
computation should be necessary.)
except this time use the values of from
I. = 5; x = 1, 1, 1, 1, 3, 3, 3, 3
E7, part c.
II. = 2; x = 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3
III. = 3; x = 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6
IV. = 2; x = 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6
V. = 7; x = 1, 2, 2, 3
11.2
Making Inferences About Slopes

The first part of this section presents and illustrates a significance test for a slope,
and the second part gives a confidence interval. The logic here is important, but
youve seen it before, and the computations follow a familiar pattern. Pay special
attention to checking how well the model fits the data. This is the part of inference
that separates humans from computers. Any old box of microchips can compute
a P-value, but it takes experience and judgment to figure out whether the P-value
really tells you anything.
The Test Statistic for a Slope

Often there is no positive or negative linear relationship between two variables.
This is true, for example, of the sum of the last four digits of a persons phone
number and the number of letters in his or her full name, as you will see in
Activity 11.2a.
Phone Numbers and Names

Let x be the sum of the last four digits of a persons phone number and y be
the number of letters in that persons full name. Record x and y for each
student in your class. Then make a scatterplot and compute the equation of
755
22-03-2009 21:25
Lesson
2 de 16
No association between
x and y.
Although the scatterplot from Activity 11.2a probably shows little or no

obvious association, chances are excellent that b1 isnt exactly equal to 0. Even
when the true slope, , is 0, the estimate, b1, will usually turn out to be different
from 0. In such cases, the estimated slope is not significant and differs from
0 simply because cases were picked at random. If another class did the activity,
the value of b1 probably would not be 0 either and probably would be different
from the value of b1 for your class as well. A significance test for the slope of a
regression line asks, Is that trend real, or could the numbers come out the way
they did by chance?
The test statistic is based on how far b1 is from 0 (or some other hypothesized
value of ) in terms of the standard error.
The test statistic for the slope is the difference between the slope, b1, estimated
from the sample, and the hypothesized slope,
, measured in standard errors:
If a linear model is correct and the null hypothesis is true, then the test statistic
has a t-distribution with n 2 degrees of freedom.
Yogi: What a relief! I was waiting for them to introduce yet another new
distribution: the z, the t, the , the blah-blah-cube.
Shark: But arent you worried about one little thing? Why does this
statistic have a t-distribution and not something else?
Yogi: Worried? I said I was relieved! I know where the t-distribution is
on my calculator. The df rule is easy. I am happy, and now you are
trying to raise problems. You have not had enough sulfur in your
diet.
Shark: Okay, well let it go until your next course in statistics. For now,
simply notice that the sample slope behaves something like a
mean.

D7. Using the printout for the Mars rocks
data in Display 11.11 on page 750,
verify the value of the t-statistic. Then
verify df, and use Table B on page 826 to
check that the P-value in the computer
printout is consistent with that in the
table.
D8. Display 11.18 shows the combined data
for the five Mars rocks and six soil
samples. In relation to the t-statistics
22-03-2009 21:25
Lesson
3 de 16
for the rocks and for the soil samples, where do you expect the value of the
t-statistic for the combined data to fall? Why?
Display 11.18 Redness versus sulfate percentage for the five

Mars rocks and six soil samples together.
Significance Test for a Slope

The box describes the steps for testing the significance of a slope. Generally, you
will use this test when you have bivariate data from a sample that appear to have
a positive (or negative) linear association and you want to establish that this
association is real. That is, you want to determine that the nonzero slope you see
didnt happen just by chancethat there actually is a true linear relationship with
a nonzero slope so knowing the value of x is helpful in predicting the value of y.
Components of a Significance Test for a Slope
randomness
linearity
uniform residuals
normality
1. Check conditions. For the test to work, the conditional distributions of y

for fixed values of x must be approximately normal, with means that lie on
a line and standard deviations that are constant across all values of x. Of
course, you cant check the population for this; you have to use the sample.
You need to do several things to check conditions:
Verify that you have one of these situations:
i. a single random sample from a bivariate population
ii. a set of independent random samples, one for each fixed value of the
explanatory variable, x
iii. an experiment with a random assignment of treatments to units
Make a scatterplot and check to see if the relationship looks linear.
Make a residual plot to check departures from linearity and that the
residuals are of uniform size across all values of x.
Make a univariate plot (dot plot, stemplot, or boxplot) of the residuals
to see if its reasonable to assume that they came from a normal
distribution.
[See Calculator Note 11C to learn how your calculator can help you check
conditions.]
(continued)
757
22-03-2009 21:25
Lesson
4 de 16
(continued)
2. State the hypotheses. The null and alternative hypotheses usually will be
H0: = 0 and Ha:
0, where is the slope of the true regression line.
However, the test may be one-sided, and the hypothesized value,
, may
be some constant other than 0.
3. Compute the value of the test statistic, find the P-value, and draw a
sketch. The test statistic is
Here b1 and sb1 are computed from your sample. To find the P-value, use
your calculators t-distribution with n 2 degrees of freedom, where n is
the number of ordered pairs in your sample. [See Calculator Note 11D.]
4. Write your conclusion linked to your computations and in context. The
smaller the P-value, the stronger the evidence against the null hypothesis.
Reject H0 if the P-value is less than the given value of , typically 0.05.
Alternatively, compare the value of t to the critical value, t*. Reject H0
if | t | t* , for a two-sided test.
Yogi: Four conditions! What happened to good old line plus random
variation?
Shark: Its still there. But variation takes in a lot of territory. The
variation about the line has to be both random and regular.
Regular here means that the vertical spread is the same as you go
from left to right across your scatterplot and that the distribution
of points in each vertical slice is roughly normal.
Yogi: This is starting to sound complicated.
Shark: Not really. Sometimes a violation of the conditions will be obvious
from the scatterplot. To be safeor if you happen to be taking
some important testyou should look at a residual plot as well as
a dot plot or boxplot of the residuals.
Yogi: That still leaves one more condition. Surely youre not going to tell
me I can check randomness by looking at a plot?
Shark: No. For that condition, you need to check how the data were
collected. The observations should have been selected randomly,
which means, partly, that they should have been selected
independently.
Yogi: (Loud sigh)
Shark: Just read the next example, and youll see how easy it is.
22-03-2009 21:25
Lesson
5 de 16
Example: Price Versus Horsepower

In E1 on page 751, you were asked about price versus horsepower for a random
sample of car models. Display 11.19 shows the data and a scatterplot. On the face
of it, the relationship looks strong enough for you to conclude that the pattern
is not simply the result of random sampling: In the population as a whole, there
really must be a relationship between price and horsepower. You would expect a
formal test to lead to the same conclusion, and it does. Carry out a significance
test for the slope of the true regression line for price versus horsepower.
Display 11.19 Price versus horsepower for a random sample of

15 car models.
Check conditions:
randomness and
linearity.
Check condition:
uniform residuals.
Solution
You have a random sample. The relationship looks reasonably linear in the
scatterplot. The equation of the least squares regression line through these points
is = 1.544 + 0.126x.
If you examine the residual plot in Display 11.20, you can see that while the
relationship appears to be generally linear, the variation from the regression line
tends to grow with x. That is, the values of y tend to fan out and get farther from
the regression line as the value of x increases. In the next section, you will see how
a transformation helps fix this violation of the conditions for inference.
759
22-03-2009 21:25
Lesson
6 de 16
Display 11.20 Residual plot from the regression analysis of price

versus horsepower.
Check condition:
normality.
If you plot the residuals themselves, as in Display 11.21, they look as if they
reasonably could have come from a normal distribution. There are gaps, but the
distribution is reasonably symmetric.
Display 11.21 Dot plot and histogram of residuals. (By hand, you might prefer
a dot plot, but your calculator will only graph a histogram.)
The hypotheses are H0: = 0 versus Ha:

0, where is the slope of the
true linear relationship between the price and the horsepower for all car models.
Do computations and
draw a sketch.
Give conclusion in
context.
The t-statistic is approximately 5.79, and the P-value is approximately 0.00006.

There is strong evidence to reject the null hypothesis and say that the slope
of the true regression line for price versus horsepower is different from 0. A linear
model with 0 for the true slope probably would not have produced these data.
The results in the example are summarized in the Data Desk printout of a
regression analysis in Display 11.22. Note that s is the standard deviation of the
residuals and is equal to the square root of the mean square for residuals. The
22-03-2009 21:25
Lesson
7 de 16
regression coefficients are given in the column labeled Coefficient. The estimated
standard error of b1 is shown in the Horsepower line in the column labeled s.e.
of Coeff . The t-ratio and its corresponding P-value finish out this line.
Display 11.22 Data Desk summary of the regression of price

versus horsepower for a random sample of cars.
Yogi: F-ratio! Aaargh! See, there is another distribution!

Shark: True, but you can ignore the F-ratio; here it is equivalent to the
t-test.
As for all statistical tests, there are three ways to get a tiny P-value like the one
in the example: The model could be unsuitable, the sample could be truly unusual,
or the null hypothesis could be false.
Unsuitable model? The relationship looks linear, and the sample is random.
But we are a bit worried about the fact that the variability of the y-values
seems to grow with the value of x.
Unusual sample? The P-value tells just how unusual the sample would be if
the null hypothesis were true. Here, less than one sample in 10,000 would give
such a large value of the test statistic.
False H0? By a process of elimination, this is the most reasonable explanation.
Whats the bottom line? In the population as a whole, theres a positive linear
relationship between price and horsepower.

D9. Consider the data on phone numbers and names that you collected in
Activity 11.2a as having come from a random sample of students. Test to
see if there is a linear relationship between the sum of the last four digits
of a students phone number and the number of letters in his or her full
name. Learn to use the t-test on your calculator to do the computations.
[See Calculator Note 11D .] Enter the data into statistical software, if you have
it available, and check the computations there as well.
D10. Suppose you have a standard computer output for a regression. What is
being tested in the line that includes a t-statistic for the coefficient of x?
What does the P-value for this test represent? What is the conclusion if the
P-value is very small, say, 0.001?
761
22-03-2009 21:25
Lesson
8 de 16
Confidence Interval Estimation
Be careful! Statistical
significance doesnt
always mean the results
are useful.
Now that you know that the price of a car increases with horsepower, your next
question might be How much? Whenever you reject a null hypothesis that a
slope is 0, it is good practice to construct a confidence interval. If the interval
is extremely wide, due to large variation in the residuals and small sample
size, that tells you that the estimate b1 is practically useless. For example, an
estimated increase in annual income of $50 to $10,000 for every additional
year of experience doesnt tell you much about what an individual employees
salary increase might be. It also might happen that the slope is so small as to be
practically meaningless, even though a large sample size makes it statistically
significant. For example, a special exercise that lowers your blood pressure by
1 to 1.5 points for every additional 5 hours per week you spend exercising will not
attract many takers, even if a huge study proves that the decrease is real. Always
temper statistical significance with practical significance.
As in past chapters, a confidence interval takes the form
statistic
Constructing a Confidence Interval for a Slope
randomness
linearity
uniform residuals
normality
1. Check conditions. To get a capture rate equal to the advertised rate, the
conditional distributions of y for fixed values of x must be approximately
normal, with means that lie on a line and standard deviations that are
constant across all values of x. Of course, you cant check the population for
these; you have to use the sample.
Verify that you have a random sample from a bivariate population
(or a set of independent random samples, one for each fixed value of
x), or verify that treatments were assigned randomly to subjects in an
experiment.
Make a scatterplot and check to see if the relationship looks linear.
Make a residual plot to check departures from linearity and that the
residuals are of uniform size across all values of x.
Make a univariate plot of the residuals to see if its reasonable to assume
that they came from a normal distribution.
2. Do computations. The confidence interval is
The value of t* depends on the confidence level and the number of degrees
of freedom, df, which is n 2. [See Calculator Note 11E.]
3. Give interpretation in context. For a 95% confidence interval, you would
say that you are 95% confident that the slope of the underlying linear
relationship lies in the interval. By 95% confidence, you mean that out of
every 100 such confidence intervals you construct from random samples,
you expect the true value, , to be in 95 of them.
22-03-2009 21:25
Lesson
9 de 16
Some calculators do not give you the value of . However, when testing that
= 0, they will give you t and b1, so you can compute from
Example: Price Versus Horsepower

Construct a 95% confidence interval for the slope of the line that relates price to
horsepower.
Check conditions.
Do computations.
context.
Solution
The conditions are the same as for a test of signifi cance and were checked in the
previous example.
You have 15 2, or 13, degrees of freedom, so the value of t* from Table B
on page 826 is 2.160. Thus, a 95% confidence interval estimate of the true slope is
given by
or (0.079, 0.173).
You can also use a calculator to find this interval. [See Calculator Note 11E.]
You are 95% confident that the slope of the true linear relationship between
price and horsepower is between 0.079 and 0.173. Converting from thousands of
dollars to dollars, the increase in the cost of a car per unit increase in horsepower is
somewhere between $79 and $173. In other words, if one model has 1 horsepower
more than another model, its price tends to be between $79 and $173 more. This
result means that any true slope between 0.079 and 0.173 could have produced
such data as a reasonably likely outcome. A value of outside the confidence
interval could not have produced numbers like the actual data as a reasonably
likely outcome.

D11. Construct 95% confidence intervals for the slopes of these regression lines.
State which interval is narrowest and which is widest. Explain why the
interval widths differ.
a. for predicting redness from sulfate percentage for the Mars rocks (use the
information in Display 11.11 on page 750)
b. for predicting redness from sulfate percentage for the Mars soil samples
(use the information in Display 11.17 on page 754)
c. for predicting redness from sulfate percentage for the rocks and soil
samples taken together (the fitted slope is 0.6268 and the estimated
standard error is 0.11)
D12. Suppose a test of the null hypothesis that the slope of the true regression
line is 0 produces a P-value of 0.02. What do you know about the 95%
confidence interval estimate of the slope?
763
22-03-2009 21:25
Lesson
10 de 16
More About the Degrees of Freedom

Yogi: I was wondering just a teeny little bit about why there are n 2
degrees of freedom.
Shark: Did you read the discussion about df in Section 10.2?
Yogi: That Jodain is pretty sharp! How much sulfur does Jodain have?
Shark: Never mind that. Can you use Jodains idea to see why there are
n 2 degrees of freedom in the test of significance for a slope?
Yogi: The concept of df should apply here because I am using a sum of
squared deviations from the regression line to estimate :
Shark: Right. The center is your regression line. How many deviations,
or residuals, are there?
Yogi: There are n values of (yi ), where n is the number of pairs
(x, y) in the sample. I suppose two of them must be redundant. If I
know n 2 of the residuals, I can figure out the other two?
Shark: Right. Try it with this example, where I wont tell you either the
values of y or the equation of the regression line. The two missing
residuals are R and S. Can you figure out what they are?
Yogi: Well, I know that (y ) = 0 because thats always the case. So

1 + (2.5) + R + S equals 0. But thats not enough to get R and S.
Shark: Correct. You need a second condition on the residuals. That
condition is that x and the residuals are uncorrelatedthe residuals
dont grow or shrink as x increases. If you check the formula for the
correlation, you will see that for it to be zero means that
Yogi: Okay. That gives me two linear equations in two unknowns:
The missing residuals are R = 2 and S = 0.5.
22-03-2009 21:25
Lesson
11 de 16
Shark: Right! So do you see why there are n 2 degrees of freedom?

Yogi: Anytime you give me all but two of the residuals, I can find the
two missing onesso they dont give me any new information
about the spread of the points about the regression line. Ill bet
even Jodain doesnt know that!
Summary 11.2: Making Inferences About Slopes
randomness
linearity
uniform residuals
normality
For both a significance test for a slope and a confidence interval for a slope,
there are four conditions you should check, which you can remember as brief
statements:
You have a random sample (or a random assignment of treatments to
subjects).
The relationship between the variables looks linear.
Residuals have equal standard deviations across all values of x.
Residuals are normal at each fixed x.
You check the first condition by finding out how the observations were
collected. Check the second and third conditions by making a scatterplot and
residual plot. Check the last condition by making a dot plot or boxplot of the
residuals.
As youll see in the next section, all is not lost if the last three conditions are
not met. But if the observations werent selected at random from the population
(or, in an experiment, if treatments werent randomly assigned), then if you
proceed with inference you must state your conclusions very, very cautiously.
To test H0: = 0, compute the test statistic
To find the P-value, compare the value of the test statistic with a t-distribution
with n 2 degrees of freedom. If you cannot reject the null hypotheses, then there
is no statistically significant evidence of a linear relationship between x and y.
For a confidence interval for the slope of the true regression line, compute
b1 t* . Again, use n 2 degrees of freedom.
As a rule, first do a test to answer the question Is there an effect? Then, if
you reject H0, construct a confidence interval to answer the question How big is
the effect?
Dont confuse statistical significance with practical importance. Significant
means big enough to be detected with the data available; important means big
enough to care about.
You should not use the techniques of this section for time-series data, that
is, for cases that correspond to consecutive points in time. In these situations,
the individual observations typically arent selected at random and so are highly
dependent. Todays temperature depends on yesterdays. The unemployment rate
next quarter is unlikely to be very far from the rate this quarter. If your cases have
a natural order in time, chances are good that you should use special inference
methods for analysis of time series rather than the methods of this chapter.
765
22-03-2009 21:25
Lesson
12 de 16
Practice
P10. The regression analysis in Display 11.23
is for the six soil samples from Mars.
The value of the test statistic and the
corresponding P-value for Sulfate are
missing. Use only the information in the
rest of the printout and Table B to find these
values. What is your conclusion?
Display 11.23 Regression analysis for the Mars soil

samples.
Display 11.24 Sum of phone number digits and

number of letters in name.
P11. Display 11.24 shows data like those

generated in Activity 11.2a on name length
and the sum of the last four digits of the
phone number for a sample of 25 people,
along with the scatterplot and regression
line. Comment on the shape of the cloud of
points in the scatterplot. Will the t-statistic
for testing that the slope is not significantly
different from zero be large or small? Explain
your reasoning.

P12. For the Mars rocks and soil samples taken
together, the fitted slope is 0.6268, with an
estimated standard error of 0.11. Use your
calculator to compute the t-statistic and find
the P-value for testing the hypothesis that
the true slope is 0. State your conclusion.
P13. Chirping. Display 11.25 shows the number of
times a cricket chirped per second and the
air temperature at the time of the chirping.
Some people claim that they can tell the
temperature by counting cricket chirps.
22-03-2009 21:25
Lesson
13 de 16
Display 11.26 Regression analysis for the chirping

problem.
P14. Display 11.27 shows IQ and head

circumference (in centimeters) for a
sample of 20 people. (These data were
explored in D21 in Chapter 3.) With head
circumference as the predictor and IQ as the
response, is there statistically significant
evidence of a linear trend?
Display 11.25 Temperature and number of chirps

per second by a Nemobius fasciatus
fasciatus cricket, as measured
electronically.[Source: George W. Pierce,
The Songs of Insects (Cambridge, Mass.: Harvard
University Press, 1949), pp. 1221.]
a. Fit a straight line to the data that could be

used to predict the temperature from the
b. Make a residual plot of (chirp rate,
residual) and a dot plot of the residuals to
check conditions.
c. Use the Minitab printout in Display
11.26 to test whether there is a linear
relationship between the chirp rate and
the temperature.
Display 11.27 IQ and head circumference. [Source:
M. J. Tramo et al., Brain Size, Head Size, and
IQ in Monozygotic Twins, Neurology 50 (1998):
124652.]
767
22-03-2009 21:25
Lesson
14 de 16

P15. Mars rocks contain a high proportion of
silicon dioxide (the predominant compound
in sand and glass) and much smaller
amounts of titanium dioxide (similar to
silicon dioxide but with titanium replacing
the silicon). Display 11.28 shows a computer
printout of the regression of titanium dioxide
percentage versus silicon dioxide percentage
for the five Mars rocks.
a. Construct a 90% confidence interval for

the slope.
b. What does the value s = 0.0257 tell you?
P16. Soil samples. Display 11.29 shows
information for the six soil samples that is
similar to that in P15 for the Mars rocks.
Construct a 90% confidence interval for the
slope. Based on your confidence intervals for
the rocks and the soil samples, would you
say the relationship between silicon dioxide
content and titanium dioxide content is the
same or dfferent for rocks and soil?
Display 11.29 Excerpt from regression of titanium

dioxide percentage versus silicon
dioxide percentage for the six soil
samples.
Display 11.28 Regression of titanium dioxide
percentage versus silicon dioxide
percentage for the five Mars rocks.
Exercises
E11. Display 11.30 shows the gas mileage (mpg)
and horsepower ratings (hp) for the random
sample of car models in E1. The scatterplot
and printout for the regression of mpg versus
hp are shown in Display 11.31.
c. Find a 90% confidence interval estimate

of the slope of the true regression line.
Interpret the result in the context of the
variables.
Display 11.30 Gas mileage and horsepower ratings.
a. Locate the estimated standard error of b1.

b. Is there evidence to say that the slope of
the true regression line is different from
0? Use = 0.05.
Display 11.31 Regression analysis for gas mileage
versus horsepower rating.
22-03-2009 21:25
Lesson
15 de 16
E12. The scatterplot and printout in Display

11.32 show the regression of mean monthly
gas usage (in therms) versus mean monthly
temperature for a single-family residence over
a sample of months. (See Display 11.15 on
pages 752753 for the complete set of data.)
E13. Display 11.33 shows the gas mileage (mpg)

and maximum revolutions per minute the
engine is designed to produce (rpm) for
the random sample of car models in E1. Is
there evidence that the mean mpg is a linear
function of the maximum rpm? Do all four
steps of the test of significance.
Display 11.33 Gas mileage and maximum

revolutions per minute for the car
models problem.
Display 11.32 Scatterplot and computer printout

for mean monthly gas usage versus
mean monthly temperature.
a. Locate the estimated standard error of the

slope.
b. Is there suffcient evidence to say that
the slope of the true regression line is
different from 0? Use = 0.05.
c. Find a 90% confidence interval estimate
of the slope of the true regression line.
Interpret the result in the context of the
variables.
E14. Refer to the data on mean monthly

electricity usage (in KWH) and mean
monthly temperature (in degrees Fahrenheit)
for a single-family residence over a sample of
months in Display 11.15 on pages 752753.
Is there evidence that the mean monthly
electricity usage is a linear function of the
mean monthly temperature?
a. Do all four steps of the test of
significance.
b. Would you say that this is strong evidence
of a linear relationship or evidence of a
strong linear relationship? Why?
E15. Refer to the data on chirp rate in P13.
a. Construct (and interpret, as always) a
95% confidence interval for the slope of
the true regression line
i. for predicting the temperature from
the chirp rate
ii. for predicting the chirp rate from the
temperature
b. Explain why the interval widths in part a
are not the same.
769
22-03-2009 21:25
Lesson
16 de 16
E16. Display 11.34 shows the arsenic

concentrations in the toenails of a sample
of 21 people who used water from their
private wells and the arsenic concentration
in their well water. Both measurements
are in parts per million.
c. Choose the data point that you think is

exerting the strongst influence on the
results. What happens to the answers in
part a if this data point is removed?
E17. Refer to the height versus arm span data in
Display 11.10 on page 749. Is there evidence
in the data to refute Leonardos claim that,
on average, height is equal to arm span?
E18. When traveling by air, would you rather
arrive on time or be relatively sure that
your bags would arrive at the same time
you do? Or perhaps there is no relationship
between these two variables. Use the data
in Display 11.35 to test whether there is
evidence of a linear relationship that
cannot be explained by chance between
the percentage of on-time arrivals and the
rate of mishandled bags for these major
airlines. ( These data were explored in
Chapter 3.)
Display 11.34 Arsenic in drinking water and

toenails (in parts per million). [Source:
M. R. Karagas et al., Toenail Samples as an
Indicator of Drinking Water Arsenic Exposure,
Display 11.35 Number of mishandled bags and
Cancer Epidemiology, Biomarkers and Prevention
percentage of on-time arrivals. [Source:
5 (1996): 84952.]
U.S. Department of Transportation, Air Travel
a. Find (and interpret, as always) a 95%
Consumer Report, October 2005.]
confidence interval estimate of the slope

of the true regression line.
i. for predicting arsenic in toenails
from arsenic in well water
ii. for predicting arsenic in well water
from arsenic in toenails
b. Explain why the interval widths in part a
are not the same.
E19. To test the potential effectiveness of laetisaric

acid in controlling fungal diseases in crop
plants, various concentrations of laetisaric
acid were applied to petri dishes containing
the fungus Pythium ultimum. After 24 hours,
the average radius of the fungus colonies
at each concentration was calculated. The
data are shown in Display 11.36. Find the
22-03-2009 21:25
Lesson
1 de 19
linear regression equation and interpret the

slope. Then determine whether the slope is
the slope of the line. Do you see any

weaknesses in this analysis?
E22. Is it appropriate to model price as a linear
function of number of bathrooms? Regardless
of your answer, fit a regression line and test
the significance of the slope. Next, analyze
the relationship between these two variables
using the techniques of Chapter 9 and
compare your two analyses.
Display 11.36 Acid concentration and average

radius of fungus colonies. [Source: Myra
L. Samuels and Jeffrey A. Witmer, Statistics for
the Life Sciences, 3d ed. (Upper Saddle River, N.J.:
Prentice Hall, 2003), p. 538. Original source:
W. S. Bowers, H. C. Hoch, P. H. Evans, and M.
Katayama, Thallophytic Allelopathy: Isolation
and Identification of Laetisaric Acid, Science 232
(1986): 1056.]
For E20E22: Display 11.37 shows the selling

price of houses, the area of the houses, the number
of bedrooms, and the number of bathrooms for
a sample of previously owned houses resold in
Gainesville, Florida.
E20. Is the slope of the least squares line for
predicting price from area statistically
significant? If so, interpret the slope.
E21. Fit a least squares regression line to price as
a function of number of bedrooms. Interpret
11.3
Display 11.37 House prices, square footage, and

number of bedrooms and bathrooms.
[Source: Board of Realtors, Gainesville, Fla.,
January 1996.]
Transforming for a Better Fit

Yogi: What are we doing in another section? I already know everything
about inference for regression.
Shark: Remember how the last section gave four conditions that are part
of the standard regression model?
771
22-03-2009 21:26
Lesson
2 de 19
Yogi: I learned how to check thoseverify that it was a random sample

plus look at a scatterplot, residual plot, and a dot plot of the
residuals. Lets move on.
Shark: But what do you do if your data dont meet the conditions?
Yogi: Thats easy. I say the data dont meet the conditions of inference for
regression, and I go on to the next exercise.
Shark: Not so fast. What if you were trying to find a cure for tree blight
and those were the only data you had?
Yogi: Rocks dont get tree blight.
Shark: (Silence)
Yogi: (Mumble/grumble) Okay, what do I do?
Shark: Sometimes, if one or more of the conditions fail to fit your data,
you can rescue the situation using the logarithm button on your
calculator.
Yogi: Great. I just love that button!
Checking the Fit of the Model

More than any other part of inference, model checking is what makes the
difference between statistical thinking and mindless number-crunching. Model
checking is also the hardest part of inference because often the answers are not
clear-cut: A plot shows a hint of curvature at one end; a point is only somewhat
apart from its neighbors; a sample is not strictly random, but there are other
reasons to think that it is representative. If you are the sort of person who likes
definite answers, you might have to push yourself to do justice to model
checking.
For inference about the slope of a regression line, there are four conditions
to check:
1. You have a random sample or a random assignment of treatments to subjects.
(Find out how the data were collected.)
2. The relationship is linear. (Check the scatterplot and residual plot. Sometimes
nonlinearity is obvious from the original scatterplot and it is unnecessary to
make a residual plot.)
3. Residuals have equal standard deviations across values of x. (Check the
residual plot. Again, sometimes it is unnecessary to make a residual plot
because its obvious from the original scatterplot that the residuals tend to
grow or shrink as x increases.)
4. Residuals are approximately normal at each fixed x. (Make a dot plot or
boxplot of only the residuals.)
In some instances, a condition will be clearly violated or clearly satisfied. In
others, patterns in the data might merely raise questions without providing clear
answers. In still others, there might be so little evidence that theres not much you
22-03-2009 21:26
Lesson
3 de 19
can say. In general, you can be more confident about your judgment when you
have a larger number of points than when you have fewer points, but there is no
simple rule.
Checking the Fit of the Model

D13. Four scatterplots IIV, along with their regression lines, residual plots, and
dot plots and boxplots of the residuals are shown in Display 11.38.
a. Which plots raise questions about whether the relationship is linear?
b. What features of a plot suggest that the variation in the response is not
constant but depends on x? Which plots show this?
c. What features of a plot suggest that the conditional distributions of y
given x arent normal? Which plots show this?
d. Which plot(s) contains an influential point? Imagine fitting a least
squares regression line to this plot using all the data and then removing
the influential point and refitting the line. How will the two lines differ?
I.
II.
773
22-03-2009 21:26
Lesson
4 de 19
(continued)
Display 11.38 Four scatterplots with their regression line, residual

plot, and dot plot and boxplot of the residuals.
Transformations to Improve Linearity

In Display 11.38, you saw both curvature and heteroscedasticitythe points
tended to fan out at one end of the scatterplot. You might remember that
Chapter 3 gave two basic transformations to straighten curved data. If you have
reason to believe the underlying relationship is
Log transformations
can decrease
heteroscedasticity.
a power function, y = axb, try replacing (x, y) with (log x, log y)

an exponential function (either growth or decay), y = abx, try replacing (x, y)
with (x, log y)
But what do you do about heteroscedasticity, a common situation with
bivariate data? For example, think about the weights of humans plotted against
their heights. Almost all babies who are around 20 in. tall will be within a few
pounds of each other. However, adults who are 6 ft tall can be hundreds of
pounds apart in weight. If you wanted to make an inference about the slope of the
regression line in a situation like this, the data would not meet the conditions. As
you will see in the example on pages 777779, a transformation using logarithms
often helps with this problem as well as with curvature.
22-03-2009 21:26
Lesson
5 de 19
Example: Largemouth Bass

The length of a captured fish is easy for a wildlife biologist to measure with a
minimum of handling, but weighing the fish is a bit more difficult. A model to
help the biologist predict weight from length would save both time and the health
of the fish. Use the data in Display 11.39 to explore a model for the relationship
between length and weight of largemouth bass.
Solution
Display 11.39 shows the length and weight measurements for a sample of 11
largemouth bass, along with a scatterplot and a residual plot of these data.
Display 11.39 Table of lengths and weights of a sample of

largemouth bass, along with a scatterplot and a
residual plot. [Source: Mathematics Teacher 90 (November
1997): 666.]
Replacing (x, y)
with (log x, log y) or
(ln x, ln y) is called a
log-log transformation.
You can see from the plot that the relationship between mean weight and length
should not be modeled by the straight line. You can also figure this out by
thinking about the physical situation. Length is a linear measure, and weight is
more closely connected to volume, a cubic measure. As you learned in Chapter 3,
you can linearize power functions by taking the logarithm of each value of x and
of each value of y. (You can use either base 10 logarithms or natural logarithms
for your change of scale.) As you can see from the scatterplot, residual plot, and
plot of the residuals in Display 11.40, ln(weight) versus ln(length) is linear, and the
residuals are small and scattered randomly about the line.
775
22-03-2009 21:26
Lesson
6 de 19
Display 11.40 Scatterplot, residual plot, dot plot of residuals,

and printout for the least squares regression of
ln(weight) versus ln(length) for largemouth bass.
The slope of the
regression line for
(ln(length), ln(weight))
is the exponent in the
original power function.
Display 11.40 also shows a printout summarizing the regression analysis. The
slope, b1, is 3.38051, with an estimated standard error of 0.0353. This is the slope
of the regression line for ln(weight) versus ln(length):
ln(weight)
8.50305 + 3.38051 ln(length)
Solving for weight

weight constant length3.38051
It appears that the mean weight of largemouth bass is proportional to the length
raised to a power that is just a little higher than 3.
Assuming the sample was randomly selected, can you reject the hypothesis
that the relationship between length and weight is cubic? With 9 degrees of
freedom, the 95% confidence interval for the true slope is given by
b1
t* sb1 = 3.38051
(2.262)(0.0353)
or (3.301, 3.460). So a slope of 3 is not a plausible value. The relationship is a little

bit stronger than cubic.
22-03-2009 21:26
Lesson
7 de 19
Example: The Worlds Women

This example is based on the data set in Display 11.41, which provides
information on population and economic variables for women in a random
sample of 30 countries from around the world. The table lists these variables:
ps percentage of parliamentary seats in single or lower chamber occupied
by women
fr total fertility rate (average number of births per woman)
se girls share of secondary schoolmiddle school and high
schoolenrollment
lew life expectancy at birth for women
lem life expectancy at birth for men
img infant mortality rate for girls (per 1000 live births)
imb infant mortality rate for boys (per 1000 live births)
What would be the appropriate model for predicting the variable infant
mortality rate for girls from fertility rate?
Display 11.41 Data on population and economic variables for women in a random sample
of countries from around the world. [Source: United Nations Department of Economic
and Social Affairs, The Worlds Women 2000: Trends and Statistics, unstats.un.org.]
Solution
The plots in Display 11.42 show the regression of the infant mortality rate for
girls (img) versus the fertility rate ( fr). The association is positive and looks as
if it follows a linear trend, but the plot is heteroscedasticcountries with larger
fertility rates tend to have larger variation in the infant mortality rate for girls.
Further, the boxplot of the residuals is skewed left and has four outliers.
777
22-03-2009 21:26
Lesson
8 de 19
Display 11.42 Scatterplot, residual plot, and boxplot of residuals

for img versus fr.
Display 11.43 Scatterplots, residual plots, and boxplots of

residuals for log(img) versus fr and log(img) versus
log(fr).
Compare Display 11.42 with Display 11.43, which shows the same
information for the regressions of log(img) versus fr and log(img) versus log( fr).
In each case, the transformation has eliminated the heteroscedasticity and the
outliers in the residuals by bringing in the larger values and spreading out the
smaller ones. The log transformation gives a beautifully symmetrical boxplot of
the residuals, but the log-log transformation gives a more randomly scattered
residual plot. Either transformation would be acceptable, so the decision of which
22-03-2009 21:26
Lesson
9 de 19
model to use would have to be made on the basis of whether a power model or
an exponential model makes more sense in the context. The sample of n = 30 is a
little more than 10% of the total number of countries reporting (about 200), so the
formal inference procedures used here would give SEs that are a little too large.
Logistic functions
model the spread of an
epidemic, for example.
They have the form
Yogi: I have got a great idea! Forget all of this transformation stuff. I
just go to the Stat Calc menu on my calculator, fit every function
in the listlinear, quadratic, cubic, quartic, log, exponential,
power, logistic (whatever that is), and sineand see which gives
me the largest value of r. Thats a whole lot easier than log this
and log that.
Shark: Chapter 3 was a long time ago, but . . .
Yogi: Oh, right. They wouldnt let me do that there either. Remind me
again why not. After all, r does tell me how closely the points
cluster about my function.
Shark: You can get a very high value of r even though the equation
you used isnt a good fit to the data. Look at Display 11.39 on
largemouth bass. It gives a value for r2 of 0.95. That tells you that
the points cluster closely to the line. But, as you saw, a line isnt a
good model for these data. They clearly follow a curve, not a line.
You can see that best from the residual plotnot from the value
of r. Also, you can get a very low value of r even though the points
form a fat elliptical cloud and a line is a perfectly appropriate
model.
Yogi: Well, okay. I see why I shouldnt pay much attention to r. But if I
think the points follow an exponential curve, why cant I just use
my calculator to fit an exponential equation instead of converting
all the ys to log ys and fitting a straight line? I promise to check
the residual plot.
Shark: Good statisticians transform for linearity because linear functions
are simpler to deal with than curvesand simpler to understand.
You know pretty well what the slope and y-intercept mean for a
line, but its much harder to interpret the parameters for other
types of equations.
Yogi: Hmmm. You must be right, because my calculator only tests for
the significance of a slope for a line! And the statistical software
we are using only fits lines, not other types of functions.
Shark: Good point.

D14. Two analyses are shown in Display 11.44. The First shows percentage of
renters (pr) versus total population (pop) for the 77 largest U.S. cities. The
five largest cities were then removed, and the results are shown in the
second analysis.
779
22-03-2009 21:26
Lesson
10 de 19
Analysis with All Cities
Analysis with the Five Largest Cities Removed
Display 11.44 Analyses of percentage of renters versus total

population. [Source: T. Erikson, Data in Depth, CD-ROM
(Emeryville, Calif.: Key Curriculum Press, 2001).]
a. Compare the two sets of results. Did eliminating the five largest cities
improve the conditions for inference? How influential did the five cities
turn out to be?
22-03-2009 21:26
Lesson
11 de 19
b. The analysis in Display 11.45 shows a log-log transformation of the data

with the five outlying cities removed. Did the log-log transformation
further improve the conditions for inference? Did it change the analysis
much? What is your conclusion as to the relationship between the
percentage of renters and the size of a city?
Log-Log Analysis with the Five Largest Cities Removed
Display 11.45 Log-log analysis for percentage of renters versus

total population.
c. Because these data are for the entire population of the largest U.S.
cities, in what sense is the regression line meaningful? In what sense is
inference for the slope meaningful?
Summary 11.3: Transforming for a Better Fit

If a scatterplot or residual plot indicates that your data do not fit the conditions
for inference very wellthe relationship is not linear, the residuals look far
from normal, or the variability of y isnt the same at each fixed xtransforming
to a new scale can often put your data in a form that gives a better fit to these
conditions.
If a power model, y = axb, fits your data, then the plot of log y versus log x will
be linear. If transforming to logs improves the fit of the linearity assumption,
the transformation often will also give residuals that are more nearly normal
and variability in y that is more constant across values of x.
If an exponential model, y = abx, fits your data, then the plot of log y versus x
will be linear. There is also a good chance that the transformed data will better
fit the other two conditionsthat residuals are roughly normal and standard
deviations are equal.
781
22-03-2009 21:26
Lesson
12 de 19
Practice
P17. Using the data on largemouth bass from
Display 11.39 on page 775 and the regression
analysis in Display 11.40, conduct a test of
the null hypothesis that the slope of the true
regression line is 3 versus the alternative
hypothesis that it is not 3.
P18. Suppose a wildlife biologist is working with
black crappies instead of largemouth bass,
getting the measurements in Display 11.46.
Display 11.46 Lengths and weights of black

crappies. [Source: Mathematics Teacher 90
(November 1997): 671.]
a. Examine the scatterplots, residual

plots, and dot plots of the residuals in
Display 11.47 for three different models:
(length, weight), (length, ln(weight)),
and (ln(length), ln(weight)). Give any
strengths and weaknesses of each model.
Is there any reason to prefer one model
over the others?
b. Display 11.48 shows the printout for
a significance test that the slope of the
regression line for (ln(length), ln(weight))
is significantly different from 0. Use the
information to perform a test that the
slope is significantly different from 3.
c. Use the printout in Display 11.48 to find
a 95% confidence interval for the slope of
the true regression line of the transformed
data. How does your model differ from the
model for largemouth bass?
Display 11.48 Printout for (In(length ), In(weight )).
Display 11.47 Scatterplots, residual plots, and dot plots of the residuals
for three models for lengths and weights of black crappies.
22-03-2009 21:26
Lesson
13 de 19
P19. The Worlds Women (continued). The

regression analysis for the logarithm of
infant mortality for girls, log(img), versus
fertility rate, fr, appears in Display 11.49.
Display 11.49 Regression analysis for log(img)

versus fr.
a. What would you predict for log(img) for

a country that had a fertility rate of five
children per woman? Based on these
data, describe the distribution of log(img)
for countries with fr = 5. What would
you predict for img for such a country?
b. Construct a 95% confidence interval for
the slope of the regression line. What does
this slope indicate about the relationship
between the variables fr and img?
P20. Some chimpanzees hunt alone or in small
groups, while others hunt in large groups.
Not surprisingly, the success of the hunt
depends in part on the size of the hunting
party. Display 11.50 shows some data on the
number of chimps in a hunting party versus
the success rate of parties of that size.
Display 11.50 Hunting party size and percentage

successful. [Source: Mathematics Teacher
(August 2005): 13; C. B. Stanford, Chimpanzee
Hunting Behavior and Human Evolution,
American Scientist 83 (1995).]
a. Plot the data with size of the hunting

party as the explanatory variable.
Describe the pattern you see. Will a linear
model work well here? Why or why not?
b. Find a transformation that results in a
linear regression equation with better
predicting ability than a linear equation
fit to the original data. What is the slope
of your equation?
c. Estimate the slope of the true regression
line for the transformed data of part b
using a 95% confidence interval. Interpret
the result in the context of this problem.
Exercises
E23. The scatterplot in Display 11.38, part IV,
on page 774 shows the crime rate (cr)
plotted against the population (pop) for the
76 largest U.S. cities for which data were
available. A test of the significance of the
slope of the regression line has a P-value
of 0.69 with df = 74 and t = 0.4053.
The five largest cities were removed, and

Display 11.51 (on the next page) shows
analyses with the original scale and with log
transformations of both variables.
a. Compare these plots with the plots from
the complete set of cities in Display
11.38, part IV. Did eliminating the five
783
22-03-2009 21:26
Lesson
14 de 19
largest cities improve the conditions for

inference? Did it change the analysis
much? How influential did the five
cities turn out to be? Did making a
log-log transformation further improve
the conditions for inference? Did it
change the analysis much? What is
your conclusion about the relationship
between the variables crime rate and

total population?
b. Because these data are for the entire
population of largest U.S. cities, in what
sense is the regression line meaningful?
In what sense is inference for the slope
meaningful?
Display 11.51 Scatterplot with regression line, residual plot, and boxplot of residuals for
(population, crime rate) and (log(population), log(crime rate)) for U.S. cities
with the four largest cities removed.
22-03-2009 21:26
Lesson
15 de 19
E24. Are graduation rates for public high schools

in the United States related to the ratio of
the number of students to the number of
teachers? Display 11.52 gives graduation
rates and student-teacher ratios for all
50 states. Graduation rate is defined as the
percentage of the freshman class from four
years earlier who graduated from high
school by the year in question.
Suppose you want a model that allows you to
predict graduation rate from student-teacher
ratio. The plots and tables in Display 11.53 (on
the next page) show the analyses for both the
original data and the log-transformed data.
a. What conditions for inference are
violated by the original data? Did the
logarithmic transformation improve

those conditions? Did the transformation
change the results of the analysis much?
What is your conclusion as to the
relationship between graduation rates
and student-teacher ratios?
b. Do you think any transformation will
bring these data into a homogeneous,
linear relationship?
c. Because these data comprise the entire
population of states, in what sense is the
regression line meaningful? In what sense
is inference for the slope meaningful?
Display 11.52 Graduation rates and student-teacher ratios. [Sources: U.S. Census
Bureau, Statistical Abstract of the United States, 20042005; and
NCES Report 11247516.]
785
22-03-2009 21:26
Lesson
16 de 19
Display 11.53 Scatterplots and analyses of (student-teacher ratio, graduation rate) and
(ln(student-teacher ratio), ln(graduation rate)).
E25. How does the number of police offcers in a

state relate to the rate of violent crime?
a. For the sample of states shown in
Display 11.54 (on the next page), find a
good-fitting model relating the number
of police officers to the violent crime rate.
b. Construct a 95% confidence interval for
the slope and interpret it. (Note that the
number of states sampled is greater than
10% of the population, which consists of
the 50 states. The only problem caused by
that fact is that the estimate of will be a
little larger than it should be.)
E26. How do violent crime rates relate to property

crime rates in U.S. cities? Display 11.55 (on
the next page) gives these rates (in terms of
crimes per 100,000 population) for a recent
year.
a. For the sample of cities shown in Display
11.55, find a good-fitting model relating
the violent crime rate to the property
crime rate.
interval estimate of the slope of the true
regression line.
22-03-2009 21:26
Lesson
17 de 19
E27. To measure the effect of certain toxicants

found in water, concentrations that kill 50%
of the fish in a tank over a fixed period of
time (LC50s) are determined in laboratories.
There are two methods for conducting
these experiments. One method has water
continuously flowing through the tanks, and
the other has static water conditions. The
Environmental Protection Agency (EPA)
wants to adjust all results to the flow-through
conditions. Given the data in Display 11.56
on a sample of ten toxicants, establish a
model that will allow adjustment of the static
values to corresponding flow-through values.
Display 11.54 Number of police offcers and violent

crime rate for a sample of states.
[Source: U.S. Census Bureau, Statistical Abstract
of the United States, 20042005.]
Display 11.56 Flow-through and static LC50s. [Source:

R. Sheaffer and James McClave,Probability and
Statistics for Engineers (Boston: Duxbury Press,
1995), p. 486.]
Display 11.55 Violent crimes and property crimes

in a sample of cities (per 100,000
population). [Source: U.S. Census Bureau,
Statistical Abstract of the United States, 2006,
Table 296.]
E28. Metals, especially copper and its alloys, are

often made by heating (sintering) powders
of essentially pure ore. As the generally
round particles of powder soften and merge,
the spaces (voids) between them become
smaller. The volume of the voids per unit
volume of the metal is a measure of the
porosity of the sintered metal, and porosity is
related to the strength of the metal. Porosity
is measured by the weight of liquid wax that
is taken up by the sintered metal.
The data in Display 11.57 (on the next page)
are from a process for sintering copper and tin
powders to make bronze bearings. Model the
relationship between the sintering time and
the weight of wax taken up by the bearings.
Construct a confidence interval for the slope
of the true regression line and interpret it.
787
22-03-2009 21:26
Lesson
18 de 19
in a 95% confidence interval. Interpret

the observed slope and the confidence
interval estimate.
c. Now plot the data from Experiment 1
and try the same transformation that you
used in part b. Does this transformation
linearize the relationship here? Does the
shelf appear to make a difference in the
relationship between height and distance?
Display 11.57 Sintering time and weight of wax

from a sintering process. [Source:
R. Sheaffer and James McClave, Probability
and Statistics for Engineers (Boston: Duxbury
Press, 1995), p. 536.]
E29. At the age of 31, Galileo conducted a series

of experiments on the path of projectiles that
eventually led to his formulation of theories
for the motion of falling bodies. In Experiment
1, a ball was released at a set height on an
inclined ramp and allowed to roll down a
groove set into the ramp. After leaving the
ramp, the ball fell to the floor unobstructed.
The measurement of the release height (H)
and the horizontal distance traveled (D)
between leaving the ramp and hitting the floor
were recorded, measured in punti.
Realizing that the ball leaving the ramp
in the first experiment had a downward
velocity when it left the ramp, Galileo
carried out Experiment 2, in which a narrow
horizontal shelf was placed at the end of
the ramp. When the ball reached the edge
of the inclined plane, it rolled across the
shelf before starting its fall, neutralizing
the downward force. The data for both
experiments are shown in Display 11.58.
a. For Experiment 2, set up a plot to predict
distance from height. Would a simple
linear model be a good fit for these data?
If not, find a transformation that will
linearize the relationship.
b. Fit a linear equation to the transformed
data of part a and estimate the slope
Display 11.58 Galileos (distance, height) data for

two experiments. [Source: D. A. Dickey
and T. Arnold, Teaching Statistics with Data
of Historic Significance: Galileos Gravity and
Motion Experiments, Journal of Statistics
Education 3, no.1 (1995).]
E30. Rivers and streams carry sediment (small

particles of rock and mineral) downhill
as they flow. It seems reasonable that
fast-moving streams would carry larger
particles than would slower-moving streams.
Knowing the relationship between the speed
of the water and the size of the sediment
particles would be of great value to those
studying, for example, the effects of dikes
and buildings on stream flow and the
resulting sediment carried off by the stream.
Display 11.59 shows the sample data on the
diameters of particles moved and the speed
of the water moving them.
a. Set up a plot to predict the size of objects
moved from the speed of the current.
Is a linear model a good fit? If not, find
a transformation that linearizes the
relationship.
b. Fit a line to the transformed data of part a
and estimate the slope in a 95% confidence
interval. Interpret the observed slope and
the confidence interval.
22-03-2009 21:26
Lesson
19 de 19
Display 11.59 Diameters of particles and the

speed of moving water. [ Source:
www.seattlecentral.edu.]
E31. Data on the worlds women can be found

in Display 11.41 on page 777. Display 11.60
shows a regression analysis of the relationship
between fertility rate and life expectancy.
a. Do the data appear to meet the

conditions for a regression analysis?
b. Verify the computations in the test of
significance for the slope in Display
11.60. Interpret the result of the test.
c. Can you infer that an increase in life
expectancy causes the fertility rate to
decrease?
d. Does the scatterplot suggest a different
way of looking at these data?
E32. Display 11.61 shows a regression analysis
of the percentage of parliamentary seats
in a single or lower chamber occupied
by women (ps) versus the girls share of
secondary school enrollment (se) from the
worlds women data on page 777. Perform a
significance test for the slope. Does it agree
with the results shown? Make the necessary
dot plot of the residuals by estimating them
from the plots. Interpret your results.
Display 11.61 Regression analysis of ps versus se .

Display 11.60 Scatterplot, residual plot, boxplot
of residuals, and analysis for
(life expectancy, fertility rate) data.
789
22-03-2009 21:26
Lesson
1 de 8
Chapter Summary
When two variables are both quantitative, you can display their relationship in a
scatterplot, and you might be able to summarize that relationship by fitting a line to
the original data or to the data after a transformation. Although many relationships
are more complicated than this, linear relationships occur often enough that many
questions can be recast as questions about the slope of a fitted line.
That is old news. What is new in this chapter is that you can use the t-statistic
to test whether the slope of the true regression line is 0. If you can reject that
possibility, your regression line will be useful in predicting y given x. If you cant,
then its plausible that the positive or negative trend you see is due solely to the
chance variation that always results when you have only a sample.
You also have learned to find a confidence interval for the slope of the
true regression line, which helps you decide whether the linear relationship is
statistically significant and whether it has any practical significance.
One way to put the methods of this chapter into a larger picture is to remind
yourself of the inference methods youve learned so far: for one proportion and
for the diference of two proportions (Chapter 8), for one mean and for the
difference between two means (Chapter 9), for the relationship between two
categorical variables (Chapter 10), and now for the relationship between two
quantitative variables. In the next chapter, you will work on four case studies that
bring all these ideas together.
Review Exercises
E33. Study the scatterplots in Display 11.62.
a. In which scatterplots is it reasonable to

model the relationship between y and x
with a straight line?
b. If you fit a line through each scatterplot
by the method of least squares, which plot
will give a line with slope closest to 0?
c. Which plot shows a correlation
coefficient closest to 1?
d. For each scatterplot that does not look as
if it should be modeled by a straight line,
suggest a way to modify the data to make
the shape of the plot more nearly linear.
E34. More on pesticides in the Wolf River.
Display 11.63 shows four scatterplots
of HCB concentration versus aldrin
concentration. The four scatterplots are for
the measurements taken on the bottom, at
mid-depth, at the surface, and for all three
locations together. Based on the plots, which
depth do you expect to give the narrowest
Display 11.62 Five scatterplots.
22-03-2009 21:27
Lesson
2 de 8
confidence interval for the slope? The

widest? Give your reasoning.
E36. As in all wars, espionage played a critical

role in the American Civil War. One of the
more interesting examples concerns the role
of detective Allan Pinkerton in using spying
methods to estimate the number of troop
regiments in the various states. Display 11.65
shows his estimates for September 1861,
along with the actual number as discovered
after the war.
Display 11.63 Four scatterplots of HCB concentration

versus aldrin concentration. [Source: R. V.
Hogg and J. Ledolter, Engineering Statistics (New
York: Macmillan, 1987). Original source: P. R.
Jaffe, F. L. Parker, and D. J. Wilson, Distribution
of Toxic Substances in Rivers, Journal of the
Environmental Engineering Division 108 (1982):
63949.]
E35. If I change to a brand of pizza with

lower fat, will I also reduce the number of
calories? Display 11.64 shows the printout
for a significance test of calories versus fat per
5-oz serving for 17 popular brands of pizza.
Display 11.64 Regression analysis for pizza data.
a. Use Display 11.64 to estimate, in a 95%

confidence interval, the reduction in
calories you can expect per 1-g decrease
in the fat content of the pizza.
b. Use your answer to part a to estimate, in a
95% confidence interval, the reduction in
calories you can expect per 5-g decrease
in the fat content of the pizza.
Display 11.65 Actual and estimated numbers of

Confederate regiments. [Source: Chris
Olsen, Was Pinkerton Right? STATS
(Winter 2000): 24.]
a. Plot the data, with Pinkertons number as

the explanatory variable. Does the plot
have a definite linear trend? If so, fit a
least squares regression line to the data.
b. Although these observational data cover
all states with Confederate regiments,
inference might still be meaningful in
deciding if a slope of this size could have
happened merely by chance. Conduct
a test to see if the observed slope was
likely to occur by chance. What is your
conclusion?
c. Interpret the slope of the least squares
line in part a. Estimate plausible values
for the true slope in a 95% confidence
interval. If Pinkerton was on target with
Chapter Summary
791
22-03-2009 21:27
Lesson
3 de 8
his estimates, what value would you get

for the slope of the line? Is this value in
the confidence interval?
d. Virginia contributed by far the most
troops to the Confederate war effort.
What influence does Virginia have on the
analysis in parts ac?
E37. How are film ratings associated with film
lengths, and are either of these variables
associated with the year of the films release?
The data in Display 11.66 are a random
sample of 20 films (TV films excluded) from
the 1996 Movie and Video Guide by Leonard
Maltin, which includes over 19,000 films.
Each movie is rated on the scale 1, 1.5, 2, 2.5,
3, 3.5, 4, with 4 being excellent.
Display 11.66 Film lengths and ratings. [Source:

Thomas L. Moore, Paradoxes in Film Ratings,
Journal of Statistics Education 14, no. 1 (2006).]
a. Is there a statistically significant linear

relationship between year and length?
Remember to do all parts of the analysis,
including the construction of appropriate
plots.
b. Is there a statistically significant
relationship between length and rating?
Again, do all parts of the analysis.
c. Given the results from parts a and b, what
do you think the relationship between
year and rating will be? Is this, in fact,
the case? Make a plot and conduct a
significance test of a slope as part of your
answer.
d. Display 11.67 shows a plot of ratings
by year, separated by short versus long
movies, with short being defined as less
than 90 minutes running length. (In the
plot, a circle represents short.) Do the
decreasing trends over the years within
the length groups appear to be stronger
than the decreasing trend in the overall
sample not accounting for length? Test
for significance of these two slopes and
compare the results to your answer in
part c.
Display 11.67 Ratings by year, separated by short

and long movies.
22-03-2009 21:27
Lesson
4 de 8
E38. The data in Display 11.68 extend the data

on the number of police offcers and the
violent crime rate in Display 11.54 on page
787 and come from a random sample of the
50 U.S. states. The variable expenditures is
the amount (in millions of dollars) it takes
to keep the given number of police officers
on the job for 1 year, population is the
population of the state in millions of people.
E40. The data in Display 11.69 represent nearly

200 years of earthquake activity in the
New Madrid area of southeastern Missouri
and adjacent states. The magnitude of
the earthquakes is based on the Richter
scale. The frequency of each magnitude
is expressed as the cumulative relative
frequency of earthquakes of magnitude
greater than or equal to that size. (For
example, about 77.6% of the recorded
earthquakes were of magnitude 1.66
or greater.) A good model for relating
magnitude to cumulative relative frequency
would be of help to scientists who study
earthquakes. Can you find such a model and
justify it statistically?
Display 11.68 Number of police officers,

expenditures, and population for a
sample of U.S. states. [Source: U.S. Census
20042005.]
a. How many new police officers would you

expect could be added for an increase
of $1 million in expenditure?
b. Does removal of the most influential
point in the set have much of an effect
on your answer to part a?
c. If one state has 1 million more people
than another state, how many more
police officers would the first state be
expected to have?
d. Does removing the most influential point
have much of an effect on your answer
to part c?
E39. In a bivariate population, what is the
relationship between the standard deviation
of the responses and the standard deviation
of the prediction errors?
Display 11.69 Earthquake magnitude and

cumulative relative frequency on
the New Madrid fault. [Source: www
.seattlecentral.edu.]
Chapter Summary
793
22-03-2009 21:27
Lesson
5 de 8
AP1. A statistics exam has two parts, free

response and multiple choice. A regression
equation for predicting the score, f, on the
free response part, from the score, m, on
the multiple choice part is f = 50 + 0.25m.
This equation was based on the scores of
17 students. The standard deviation of their
multiple choice scores was 30, the standard
deviation of their free response scores was
16, and the sum of the squared residuals
was 2940. What is the estimated standard
error of the slope?
0.117
0.133
0.219
0.467
3.5
AP2. Which of the following is not an important
condition to check before constructing a
confidence interval for the slope of the true
regression line?
You have a simple random sample.
The points fall in an elliptical cloud.
The residuals for small values of x
have about the same variability as the
residuals for large values of x.
The sum of the squared residuals is small.
The residuals are approximately
normally distributed.
AP3. On the first day of statistics class, data on
shoe size and height were gathered for
82 randomly selected female high school
seniors. Part of a regression analysis is given
below. Which of the following is the best
interpretation of the P-value for shoe size?
An error has been made in the data

collection or analysis.
There is statistically significant evidence
of a non-zero slope in the true linear
relationship between shoe size and height.
There isnt statistically significant
evidence of a non-zero slope in the true
linear relationship between shoe size and

height.
A confidence interval for the slope
would have a width of 0.
A confidence interval for the slope
would include 0.
AP4. Refer to AP3. What is the best explanation
of what is measured by s = 1.87862?
the variability in the slope from sample
to sample
the variability in the y-intercept from
sample to sample
the variability in the shoe sizes
the variability in the heights
the variability in the residuals
AP5. Refer to AP3. What is the best explanation
of what is measured by Std Error = 0.1438?
the variability in the slope from sample
to sample
the variability in the y-intercept from
sample to sample
the variability in the shoe sizes
the variability in the heights
the variability in the residuals
AP6. Refer to AP3. Which of the following is
the appropriate computation for a 95%
confidence interval for the slope?
1.2116 1.990 1.87862
1.2116 1.990 0.1438
1.2116 1.96 0.1438/
1.2116 1.990 0.1438/
1.2116 1.990 1.87862/
AP7. In an attempt to predict adult heights,
researchers randomly selected men and
collected their heights at age two and
their adult heights, then computed a
least squares regression equation,
adult height = 1.9 height at age two + 3.5.
A 95% confidence interval for the slope was
given as (1.8, 2.0). Which of the following
is the best interpretation of this confidence
interval?
22-03-2009 21:27
Lesson
6 de 8
If the researchers took 100 more random

samples, they would expect 95 of the
regression equations to have a slope
95% of two-year-old boys have heights
The researchers are pretty sure that if
they studied all men, the slope of the
true regression line would be between
1.8 and 2.0.
If a two-year-old boys height is known,
theres a 95% chance that his adult height
will be between 1.8 and 2.0 times his
current height.
In predicting the height of an adult man
using his height at age two, 95% of the
errors will be between 1.8 and 2.0 inches.
AP8. A friend is doing a regression analysis to
predict the mass of a small bag of fries given
the number of fries in the bag. He finds a
linear regression equation and then makes
the following residual plot. He asks your
advice about whether to proceed with a test
of significance of the slope. Which is the
best advice you could give?
The relationship doesnt have constant

variability across all values of x. Its not
okay to proceed.
Try a transformation before proceeding.
Its clear even without a significance test
that the true slope isnt 0.
Investigative Tasks
AP9. Do animals make optimal decisions? Tim
Penning of Hope College studied the strategy
his dog, Elvis, uses in retrieving a ball thrown
into the water of Lake Michigan. Looking at
the diagram in Display 11.70, suppose Tim
and Elvis stand on the edge of the lake at A
and the ball is thrown to B in the lake. Elvis
could jump into the water immediately at A
and swim to the ball, but he seems to know
that he can run faster than he can swim. He
could run all the way to C and then swim the
perpendicular distance to the ball, but that
is not the most time-effcient strategy either.
What Elvis actually does is run to a point D
and then swim diagonally to the ball at B.
But, does he determine D so as to minimize
the time it takes him to get to the ball?
Display 11.70 Diagram of paths Elvis might

follow to get from A to B.
Somethings wrong here. This cant be a

residual plot.
The relationship is linear. Its okay to
proceed with the significance test.
Methods of calculus can be used to show

that the time is minimized when y is related
to x by the formula
AP Sample Test
795
22-03-2009 21:27
Lesson
7 de 8
where r is Elviss running speed and s is his

swimming speed. Further experiments with
Elvis determined that his running speed
was about 6.40 m/s and his swimming
speed was about 0.91 m/s, which result in
the optimal relationship for minimizing
time to the ball of
y = 0.144x
Display 11.71 shows the measurements
x and y (in meters) for 35 ball-tossing
experiences with Elvis.
a. Describe the pattern in the scatterplot.

b. Assuming this to be a random sample of
Elviss work, construct a 90% confidence
interval estimate of the slope of the true
regression line relating distances x and
y. Does it look like Elvis is making an
optimal decision?
c. Assuming this to be a random sample
of Elviss work, construct a test of the
hypothesis, at the 0.05 significance level,
that the slope of the true regression line
relating distances x and y is the optimal
0.144. Now does it look like Elvis is
making an optimal decision?
d. Choose the point that you think is most
influential on the slope of the regression
line and the correlation. Describe what
will happen to the slope of the regression
line and the correlation if it is removed
and the data re-analyzed.
e. The analysis below shows that a
regression model fit to the data in
Display 11.71 does not have a y-intercept
of 0, as the optimal model does. Explain
what will happen to the slope of the
regression line shown in the plot if you
force the line to go through the origin.
Display 11.72 Analysis of Elviss data.

Display 11.71 Data from Elvis. [Source: The College
Mathematics Journal 34, May 2003.]
22-03-2009 21:27
Lesson
8 de 8
AP10. Display 11.73 shows birthrate versus gross

national product for a sample of countries.
The overall shape of the scatterplot looks

roughly like that of exponential decay
(exponential with a negative exponent),
except for one problem: The curve should
be close to the x-axis as it moves out
into the right tail, a pattern that would
correspond to birthrates dropping to zero as
richer and richer countries are considered.
That doesnt happen, even approximately,
for the countries in the data set.
a. How might you first transform the data
to give an exponential model a chance of
fitting?
b. Make the transformation you identified
in part a first, and then transform again
to see if an exponential model is a
reasonable one for these data.
c. What birthrate would you predict for a
country with a gross national product of
$15,000 per capita?
d. Does the same principle apply to the
scatterplot in Display 11.74 of life
expectancy at birth for men versus
infant mortality rate for boys for the
sample of countries in Display 11.41
on page 777?
Display 11.74 Scatterplot of life expectancy

at birth for men versus infant
mortality rate for boys for a
random sample of countries.
Display 11.73 Birthrate versus gross national product.

[Source: U.S. Census Bureau, Statistical Abstract of
the United States, 1995.]
AP Sample Test
797
22-03-2009 21:27
Lesson
1 de 2
12
Statistics in Action:
Case Studies
Can you grow

bigger flowers by
reducing the length
of their stems? Plant
scientists designed
an experiment to
test different growth
inhibitors to see which
were most effective in
reducing the length of
stems, anticipating that
reduced stem growth
would enhance the
quality of the flower.
22-03-2009 21:27
Lesson
2 de 2
Statistics today is big business. Nearly every large commercial enterprise and
government agency in the United States needs employees who understand how
to collect data, analyze it, and report conclusions. The language and techniques of
survey and experimental design are part of politics, medicine, industry, advertising
and marketing, and even, as you will see, flower growing. Consequently,
a statistics course typically is required of college students who major in
mathematics or in fields that use data, such as business, sociology, psychology,
biology, and health science.
In this chapter, you will examine four examples of statistical

practice in todays world: case studies about how to
produce bigger and better flowers
gather information on Americans
evaluate the economics of Major League Baseball
study possible discrimination in employment
In each of these four real-world situations, statisticians have worked with the
business or government agency involved. You will be following in their steps, so
think of yourself as a consultant to the business or government agency. Compared
to those in previous chapters, the data will be more complex and which statistical
technique to use will be less clear-cut. Several statistical techniques may be
appropriate, and you will have to decide which would be best. Often, the data
dont quite satisfy the conditions for the techniques you have learned, and you
will have to think about what can be done in such situations, which come up in
practice more often than not. If you do all four case studies, this chapter can serve
as a comprehensive review of the statistical concepts and techniques that you
have learned in this course.
22-03-2009 21:27
Lesson
1 de 3
12.1
Mums the Word!

The commercial flower industry is large and growing in the United States,
particularly in Florida. The chrysanthemum, or mum, is a popular flower for
bouquets and potted plants because it is hardy, long-lasting, and colorful. If
you visit a flower shop to choose a bouquet of mums for someone special, you
probably will look for the largest, most fully developed, most colorful blooms
you can find. Flowers with these characteristics are in demand, and they dont
show up in your flower shop by accident. Considerable research goes into
producing fine blooms.
Plants can produce and use only so much energy. So within the same species,
a plant with a long stem is likely to have a smaller flower than a plant with a short
stem. The Environmental Horticulture Department of the University of Florida
experimented with growth inhibitors to see which were most effective in reducing
the length of stems, anticipating that reduced stem growth would enhance the
quality of the flower. There were many treatments in the complete study: this
investigation will consider only seven of them. The patterns you will see here
remain much the same when all treatments are considered.
Individual plants of the same age were grown under nearly identical conditions,
except for the growth-inhibitor treatment. Each treatment was randomly assigned
to ten plants, whose heights (in centimeters) were measured at the outset of the
experiment (Hti) and after a period of 10 weeks (Htf ). (Heights actually were
measured at intervening times as well, but those measurements are not part of
this analysis.)
You can find the raw data for this experimentthe treatment each of the
70 plants received and the height of each plant before and after treatmentin
Analyzing the Experiment

D1. Can you suggest possible improvements to this experimental design?
D2. What response variable would be best to use for comparing treatment
groups in this experiment?
800 Chapter 12 Statistics in Action: Case Studies
22-03-2009 21:28
Lesson
2 de 3
D3. What statistical method(s) would you suggest for comparing the
effectiveness of, say, Treatments 1 and 2? What conditions must you check
before proceeding?
Display 12.1 shows the statistical summaries of the growth,
, during
the 10 weeks for each of the seven treatment groups. Notice that the means
fluctuate quite a bit from group to group, as do the variances.
, for
Display 12.1 Summary statistics for plant growth,

seven treatment groups.
Display 12.2 shows the boxplots of the differences in height,

, by
treatment. The boxplots give a better view of the key features of the data than does
the summary table.
Display 12.2 Boxplots of the differences in height (in centimeters)

of mums for the seven treatment groups.
Display 12.3 shows the raw data for this experiment.
12.1 Mums the Word!
801
22-03-2009 21:28
Lesson
3 de 3
Display 12.3 The treatment each of the 70 plants received and its
height before and after treatment. [Source: University of
Florida Institute for Food and Agricultural Sciences, 1997.]
22-03-2009 21:28
Lesson
1 de 7
Practice
Analyzing the Experiment
significant? Be sure to check the conditions
for the test you select. Also write the
P1. Which two treatments appear to be most
hypotheses in words.
effective in reducing the stem length? Least
effective? Explain your choices.
P5. The two treatments that differ most appear
to be Treatments 4 and 5. Why is it clear
P2. Compare the distributions in terms of center
from the plot that the difference in mean
and spread. Do you see any pattern relating
growth between these two treatments
the spreads to the centers? Make a statistical
is statistically significant? Construct a
graphic that shows how the mean and
95% confidence interval to support your
standard deviation are related. Describe this
reasoning.
relationship. Is this the kind of relationship
P6. By hand, make boxplots of the initial
you would have expected? Explain.
heights of the mums assigned to Treatments
P3. Refer to P2. If one treatment has a mean that
4 and 5. Show any outliers determined by the
is 1 cm larger than that of another treatment,
1.5 IQR rule. Describe any differences in
how would you expect the standard
the initial heights of the plants assigned
deviations to differ?
to these two treatments. Is there enough
P4. Is the difference in mean growth between
difference to call into question the
the two most effective treatments statistically
conclusion in P5?
12.2
Keeping Tabs on Americans

The American Community Survey (ACS) is a nationwide sample survey that
collects fairly detailed data on persons and households at nearly 3 million
addresses across the United States. The goals of the annual survey are to
provide federal, state, and local governments with an information base for the
administration and evaluation of government programs; to facilitate improvement
of census data; and to provide timely demographic, housing, social, and economic
statistics that can be compared across states, communities, and population groups.
To keep things manageable, in this case study you will work with a sample of
40 households and with only 12 of the more than 50 variables measured by the ACS.
This sample of 40 households comes from Florida (where over 27,000 households
are sampled in the complete ACS). The 40 households are a random sample of a
random sample, and so you can think of them as a simple random sample of Florida
households. The data are provided in Display 12.4 (on the next page) for these
variables:
NP
number of persons
BDS number of bedrooms
BLD building type: (1 = mobile home, 2 = single-family detached house,
3 = apartment or other attached structure)
ELEP mean monthly electricity payment
INSP hazard insurance payment for the year
MRGP monthly mortgage payment
RMS number of rooms
12.2 Keeping Tabs on Americans
803
22-03-2009 21:28
Lesson
2 de 7
OWN
VEH
HHL
NOC
R65
ownership (1 = own with a mortgage, 2 = own free and clear, 3 = rent)

number of motor vehicles
household primary language (1 = English, 2 = other)
number of children under the age of 18
at least one householder age 65 or older (0 = no, 1 = yes)
Display 12.4 Random sample of 40 ACS households in Florida.

[Source: American Community Survey, 2003, www.census.gov.]
22-03-2009 21:28
Lesson
3 de 7
Analyzing the Survey Data

D4. Give at least two possible reasons for the missing data in the table in Display
12.4. Do you suppose such gaps are common in surveys of this type?
D5. Based on these data and the techniques of this book, can you construct a
confidence interval for the difference between the proportion of households
residing in detached houses and the proportion residing in mobile homes?
D6. Can you construct a confidence interval for the difference between the mean
hazard insurance payments for owners and renters with these data? Explain
why or why not.
D7. In addition to retired folks on limited incomes, Florida has some very
wealthy residents. Why, then, is the largest number of rooms among
households in this data set only nine?
Inference with Univariate Categorical Data

Florida has the reputation of being a state full of elderly retired folks, so lets
begin by estimating the proportion of households that actually contain at least
one person age 65 or older. The sample proportion, , is 13/40, or 0.325. The
conditions for estimation are met because this is a random sample and both
= 13 and n(1 ) = 27 are greater than 10. The 95% confidence interval
estimate turns out to be (0.18, 0.47), a fairly wide interval because of the relatively
small sample size. So it is plausible that nearly half of Florida households contain
a person age 65 or older.
Inference with Bivariate Categorical Data

Many of the elderly live on nearly fixed incomes, and mobile homes are relatively
inexpensive compared to houses. Because the retired tend to have less income,
perhaps a higher proportion of them live in mobile homes. To check this out,
you can look for a possible association between R65 and BLD (building type) by
summarizing the data in a two-way table. Display 12.5 (on the next page) gives
such a table, with the expected cell frequencies under the null hypothesis of no
association and the resulting chi-square test of independence.
The P-value is small, but not small enough to reject the null hypothesis of
independence between the categories at the 0.05 significance level. In fact, this
statistic is inflated a bit because three of the cells have expected frequencies less
than 5, so the actual P-value might be even greater than the one shown. In short,
there is little evidence to support an association between whether a household has
an older member and the type of buildings in which the household resides.
As in Chapter 1, you can use randomization as a basis for making a decision
about two-way tables. This method does not depend on the distribution and,
in particular, does not require expected cell frequencies of 5 or greater. If there is
no association between BLD and R65, then the observed value of the statistic
(5.432 in this example) is simply a random occurrence and does not depend
805
22-03-2009 21:28
Lesson
4 de 7
Display 12.5 Summary of households with members at least age

65 versus building type.
on the way the pairs of data on BLD and R65 were matched up in the observed
sample. To check the possibility that a statistic this large reasonably could
occur by chance, you must generate a randomization distribution of possible
values. Begin by listing the BLD values in a column. Write the R65 values on cards
and shuffe them. Deal them out, one by one, so that you pair each BLD value
with a randomly assigned R65 value. Then compute for this randomization.
You do this many times. Display 12.6 shows 200 values for such a randomization
distribution.
Display 12.6 Randomized values of
for BLD versus R65.
22-03-2009 21:28
Lesson
5 de 7
The observed value, 5.432, was met or exceeded 15 times out of 200 runs,
making the approximate P-value 0.075. Notice that this is larger than the P-value
from the formal chi-square test (0.066) and provides even less evidence against
the null hypothesis of no association between whether the household has a
member age 65 or older and the type of building in which the household resides.

D8. Create one additional value for the randomization distribution in
Display 12.6.
D9. Florida sometimes conjures up the image of many retired people living
in mobile homes. How would you proceed if you wanted to determine
if householders age 65 and older are more likely to live in mobile homes
than not?
Inference with Univariate Measurement Data

How many rooms, on average, does a household in Florida have for its use? A plot
of the sample data on RMS, given in Display 12.7, shows that the sample might
well have come from a normal distribution.
Display 12.7 Number of rooms in the sample of households from

the American Community Survey.
Conditions for a confidence interval estimate of plausible values for the

population mean are met, and the 95% confidence interval for the number of
rooms per household in Florida is (4.76, 5.79). Even though there are some very
large houses in Florida, the preponderance of smaller places of residence keeps
the mean number of rooms per household rather small.
Inference with Bivariate Measurement Data

Most households in Florida use a lot of air conditioning, so electricity costs can
be considerable. It seems reasonable that the electricity bill should increase with
the number of rooms, but will the pattern of association be linear?
Display 12.8 (on the next page) shows a fairly pronounced linear trend for
these data. The regression equation is ELEP = 17.1 RMS + 34.7. You will work
with this regression in P18 and P19.
807
22-03-2009 21:28
Lesson
6 de 7
Display 12.8 Electricity payment versus number of rooms for the

ACS sample.
Practice
Inference with Univariate Categorical Data
P7. Suppose someone claims that a quarter of
Florida households have a primary language
other than English. Can you refute this claim
based on the data in Display 12.4 on page 804?
P8. Estimate with 95% confidence the
proportion of Florida households residing
in apartments or other attached structures.
P9. What plausible proportions of Florida
households own their home outright?
P10. Construct a two-way table of BLD versus
OWN. By looking carefully at your two-way
table but not actually conducting a test,
does it appear that these two variables are
associated? If so, explain the nature of the
association.
P11. In some cultures it is common to have
extended families living in the same
household, which might suggest that a
greater proportion of nonEnglish speaking
households would contain an older resident.

Do the data in Display 12.4 contain any
evidence of this? Which two equivalent
significance tests could you use to answer
this question? What concerns do you have,
if any, about conducting and interpreting
these tests?
P12. Is there a significant association between
primary language spoken in the household
(HHL) and the type of building in which
the household is found (BLD)? If so, explain
the nature of the apparent association. Are
you concerned about the accuracy of the
reported P-value here? Why or why not?
P13. Display 12.9 (on the next page) shows 200
runs of the randomization distribution for
values with BLD randomly paired with HHL.
a. Describe how this distribution was
constructed.
b. Does this distribution alleviate any
concerns you might have about the
results of P12? Explain.
22-03-2009 21:28
Lesson
7 de 7
Display 12.9 Randomized values of
Inference with Univariate

Measurement Data
P14. Estimate the mean number of people per
household in Florida, with 90% confidence.
Be sure to check the conditions first.
P15. Focus on the monthly mortgage payments,
for those households that have them.
a. Plot the data on monthly mortgage
payments and describe the distribution.
Do these data appear to meet the
conditions for inference for a mean? Will
a transformation help?
b. In 2003, the mean monthly mortgage
payment nationwide was $840. Is there
statistically significant evidence that the
mean was less than that in Florida?
P16. Florida is a hot spot for hurricanes, so
hazard insurance for a persons place of
residence is highly recommended. Consider
the yearly amounts paid for hazard insurance
in the sampled households in Display 12.4.
a. Plot the data on insurance payments
and describe the distribution. Is a
transformation needed here, in order to
allow you to make inferences about the
mean? If so, find a transformation that
works well.
b. Estimate the population mean of the
appropriately transformed data in a
for BLD versus HHL.
P17. Estimate the difference between the mean

number of rooms in detached houses and in
apartments in a 95% confidence interval.
Inference with Bivariate Measurement Data
P18. Interpret the slope of the regression line in
Display 12.8.
P19. Is the slope in Display 12.8 significantly
different from 0? Be sure to do all steps of
the significance test.
P20. Make an appropriate plot and study the
relationship between the number of
people and the number of bedrooms in
the households in Display 12.4 on page 804.
Would you say that the number of bedrooms
in a residence is a strong predictor of the
number of people in a household? Explain
your reasoning.
P21. One possible use of the data in Display 12.4
is to build a model to predict the number
of vehicles that might be associated with a
household.
a. Which is the better predictor of the
number of vehicles attached to a
household: the number of people, or the
number of rooms?
b. Test for the significance of the slope in
each relationship in part a. What are your
conclusions? Give a possible practical
reason for this state of affairs.
809
22-03-2009 21:28
Lesson
1 de 7
P22. How does the yearly hazard insurance

payment relate to the number of rooms?
a. Plot INSP versus RMS, fit a line to the
data, and interpret the slope.
b. Do the conditions for a confidence interval
for the slope and for a test of significance
of the slope appear to be satisfied here? If
not, explain what is amiss.
12.3
c. Find a transformation that cures

whatever is amiss in part b. Then redo
the regression and estimate the slope of
the line relating the two variables in a
95% confidence interval and interpret
the result.
Baseball: Does Money Buy Success?

Salaries of professional athletes are astronomical, but huge payrolls are not
necessarily bad for team owners if the salaries are associated with great team
performance or with other variables that improve income, such as attendance.
What are the relationships between some of these key variables for Major League
Baseball?
Display 12.10 gives the total
payroll, the average salary per
player, the total attendance, the
team batting average (percentage
of hits, without the decimal), and
the percentage of games won in
regular-season play (without the
decimal) of teams for the 2001
season. Teams are divided into
two major leagues (the American
and the National), and each league
is divided into three divisions
(East, Central, and West).
Exploring the Table of Data

D10. Looking only at Display 12.10, do you see any interesting features of the
data? Do you see any possible patterns or associations between variables?
D11. Using statistical software, explore these data and comment on any
interesting patterns or possible relationships you find. Are there any
surprises?
D12. Which variable was most strongly associated with payroll? Does it matter
much whether you use total team payroll or average salary per player in
looking for relationships with other variables in the data set? Explain.
D13. Discuss whether you think it is appropriate to do inferential analyses on
these data.
22-03-2009 21:28
Lesson
2 de 7
Display 12.10 Data on Major League Baseball for 2001. [Sources:

www.CNNSI.com; www.slam.ca; www.mlb.com, June 20, 2002.]
12.3 Baseball: Does Money Buy Success?
811
22-03-2009 21:28
Lesson
3 de 7
Is Inference Appropriate?
This is an observational study because there was no random sampling and no
random assignment of treatments. (Owners probably would object to a random
assignment of a payroll to each team.) The analysis should start, as you already
have done, with the exploration of possible relationships among the variables,
but some inference might be possible if the nature of the inference is carefully
phrased. Payroll is certainly not random; for the most part, these values are fixed
at the beginning of the season. Even though attendance, batting average, and wins
are somewhat random events that occur throughout the season, they were neither
randomly sampled nor randomly assigned. You can, however, still ask, How
likely am I to observe by mere chance a pattern like the one seen in the data? In
studying bivariate relationships, the null hypothesis in the significance test is that
the data look like a random sample of ys for fixed values of x from a population
with no linear trend. You can test to see whether this is a reasonable model or
whether a linear trend appears to be a better explanation. In comparing two
means, the null hypothesis is that the data look like independent random samples
from two populations with the same mean.
This case study is similar in construction to common investigations of
industrial processes in which yields are measured after temperature and pressure
gauges have been set at fixed levels determined by an engineer. Often, these
temperature settings cannot be randomized because of other considerations in the
process, such as the time it takes to get a process up to the required temperature
and pressure levels. Even with no randomness in the design, the statistical analysis
of yields does help decide whether an observed association can be attributed to
chance alone. But it cannot answer questions of cause and effect.
Differences Between the Leagues

Of interest in this study is whether possible associations among the variables are
different for the two leagues. Display 12.11 shows a scatterplot, residual plot, and
regression analysis of percent wins versus payroll for all 30 teams.
Display 12.11 Percent wins versus payroll for all 30 teams.
22-03-2009 21:28
Lesson
4 de 7
The P-value for the

American League is
0.4545; for the National
League, its 0.0350.
The observed trend is increasing, but a closer scrutiny of the test of significance
for the slope shows that the null hypothesis of a zero slope has a two-sided
P-value of 0.0674. This is small, but not small enough to make a bold declaration
in favor of a linear trend with nonzero slope. In other words, the trend here is no
more pronounced than you would expect if you randomly assigned the winning
percentages to the different teams while keeping their payroll fixed.
Separating the leagues, however, gives the plots and analyses in Display 12.12.
While the American League shows no significant linear trend, the National League
does show a significant trend. That is, if payroll were kept constant for each team
in the American League but percentage of wins were reassigned at random to
the teams, then it is quite likely that you would get a slope as far from 0 as the
American Leagues 0.782. If you did the same thing for the teams in the National
League, it isnt likely that you would get a slope as far from 0 as 1.49, indicating that
the positive trend in the National League cannot reasonably be explained by chance.
American League
National League
Display 12.12 Percent wins versus payroll for the American League and the National League.
813
22-03-2009 21:28
Lesson
5 de 7

D14. Compare the plots in Display 12.12 for the two leagues. What is it about
these data that produces the drastic difference in results for the two leagues?
D15. Besides having a random sample, what other conditions need to be met for a
regression analysis? From the plots in Display 12.12, does it seem reasonable
to assume that these other conditions are met?
Simulating a P-Value
The null hypothesis for percent wins versus payroll is this:
The observed slope is no farther from 0 than you would be reasonably likely
to get if you randomly reassigned the values of percent wins to different teams
while keeping each teams payroll fixed.
The P-value for the test of significance for the slope measures how unusual the
observed slope would be under those conditions. You can estimate the P-value
by repeatedly rearranging the y-values and observing what happens to the
slopes. Using this idea, Display 12.13 shows a set of 100 slopes found by 100
rerandomizations of the values of percent wins for the American League.
Display 12.13 Stem-and-leaf plot of 100 simulated slopes for

percent wins versus payroll for the American
League.
In the 100 trials, the observed slope, 0.782, for the American League was
equaled or exceeded 21 times, giving a simulated one-sided P-value of 0.21. The
t-test in Display 12.12 gives a two-sided P-value of 0.4545, which is almost
equal to the P-value from the simulation. So with either method, the question
of whether the observed pattern reasonably could be attributed to mere chance
is answered in the affirmative for the American League.
Simulating a P-Value
D16. The stem-and-leaf plot in Display 12.14 shows a simulation for the National
League that is parallel to the simulation for the American League.
a. Describe how the simulation was conducted.
b. Conduct one more trial and show where it would go on the stemplot.
22-03-2009 21:28
Lesson
6 de 7
c. What is the simulated one-sided P-value for these 101 trials? How does
it compare to the two-sided P-value given in Display 12.12? What is your
conclusion about the observed pattern for the National League?
Display 12.14 One hundred simulated slopes for percent wins

versus payroll for the National League.
Ecological Correlations
Each of the two major leagues is divided into three divisions, so another way
of looking at the baseball data is to explore what happens at the division level.
Display 12.15 shows the averages for the data in Display 12.10 on page 811 by
division. (All teams play a 162-game schedule, with similar numbers at bat for the
season, so it is fair to take simple averages of team batting averages and winning
percentages.)
Display 12.15 Data on Major League Baseball, by division,

in 2001.
In analyzing how percent wins is related to batting average, you could use the
teams as cases or use the divisions as cases. The scatterplot for each analysis is
815
22-03-2009 21:28
Lesson
7 de 7
Display 12.16 Scatterplots of percent wins versus batting

average, by team and by division.
Each plot in Display 12.16 shows an increasing trend, but the correlation in
the first is about 0.49, while the correlation in the second is about 0.66. In this
situation, using the division averages instead of the individual team data inflates
the correlation and makes the linear trend appear to be stronger.
The correlation computed using the divisions as cases and the division
averages as variables is an example of an ecological correlation, or a correlation
between group averages. Such correlations are often used in subjects such as
sociology and political science, where it is easier to get information about groups
of people than about individuals. (For example, its easy to find the percentage
of people in your state who voted Republican in the last presidential election but
almost impossible to get the same information for each individual voter.) As you
have seen, correlations based on groups and averages can be quite different from
correlations based on individuals.
Ecological Fallacy
Ecological correlations use groups as cases and averages as variables. For many
situations, you get quite different values than you would if you used individuals
as cases. The mistake of using ecological correlations to support conclusions
about individuals is called the ecological fallacy.
Of course, the team statistics are themselves averages (or totals), and we
could have analyzed the relationship between salaries and batting averages by
using individual players as cases. For the purpose of studying teams as business
entitiestaking into account variables such as attendance and percent winsit
makes sense to use teams as cases. Team statistics are of little use, however, in
studying player performance. So the choice of what to use as cases depends on
the objectives of the study.
Ecological Correlations
D17. Do both lines in Display 12.16 have slopes that are significantly greater than
0? Does this suggest another problem with using ecological correlations?
22-03-2009 21:28
Lesson
1 de 7
D18. Calculate the correlation between batting average and payroll using teams
as cases and then using divisions as cases. Comment on possible reasons for
any difference.
D19. Calculate the correlation between percent wins and payroll using teams as
cases and then using divisions as cases. Does the latter correlation appear to
have a meaningful interpretation?
Practice
For P23P25: Analyze the given relationship
a. for all teams
b. for the American League
c. for the National League
d. Compare the three relationships.
P23. payroll versus attendance
P24. batting average versus payroll
P25. percent wins versus batting average
P26. Use an appropriate test to answer these
questions about means.
a. Is the difference between mean attendance
for the two leagues statistically significant?
b. Is the difference between mean batting
average for the two leagues statistically
significant?
12.4
P27. In the 2001 World Series, the New York

Yankees lost to the Arizona Diamondbacks.
Is the difference in their percentage of wins
during the regular 162-game 2001 season
statistically significant?
P28. For the 2001 regular season of 162 games,
the three National League division winners
were Atlanta, St. Louis, and Arizona. For the
American League, the division winners were
New York, Cleveland, and Seattle.
a. Are there statistically significant
differences among the proportions of
games won by the three National League
division winners?
b. Are there statistically significant
differences among the proportions of
games won by the three American League
division winners?
Martin v. Westvaco Revisited: Testing for Discrimination

Against Employees
In Chapter 1, you read about Robert Martin, who was laid off from his job at
the Westvaco Corporation at age 54. The statistical analysis presented during
the lawsuit was quite a bit more involved than the simplified version you saw
in Chapter 1. With what you have learned, you can now carry out a much more
thorough analysis while reviewing some inferential techniques as well as some
ideas from probability theory.
In the Westvaco case, Martin claimed that he had been terminated because of
his age, which, if true, would be against the law. His case eventually was settled
out of court, before going to trial, but not until after a lot of statistical analysis and
some arguments about the statistics.
Display 1.1 on page 5 shows the data used in the lawsuit, arranged in a
standard cases by variables format. Each row of the table represents a caseone
of the 50 people who worked in the engineering department of the envelope
division of Westvaco when layoffs began.
12.4 Martin v. Westvaco Revisited: Testing for Discrimination Against Employees
817
22-03-2009 21:29
Lesson
2 de 7
Comparing Termination Rates for Two Age Groups

By law, all employees age 40 or older belong to what is called a protected class:
To discriminate against them on the basis of their age is against the law. At the
time of the layoffs at Westvaco, 36 of the 50 people working in the engineering
department were age 40 or older. A total of 28 workers were terminated, and
21 of them were age 40 or older.
Comparing Termination Rates for Two Age Groups

D20. Construct a table to satisfy each of these descriptions. In each table, keep
the marginal totals the same as in the Westvaco case, presented in
Display 12.17.
a. Make a copy of Display 12.17 and fill in the cells in order to present
the strongest possible case for discrimination against workers age 40
or older.
Display 12.17 Two-way table of layoff status and age, with

marginal totals from the Westvaco case.
b. Make another table, but this time fill in the cells in a way that makes the
variables terminated and age as close to independent as possible. If you
select an employee at random from those represented in your table, are
the variables terminated and age independent events according to the
definition in Section 5.5?
c. Make a final table using the actual data from Westvaco. Does it look
more like the table in part a or the table in part b? Do you have strong
evidence that older workers were more likely to be chosen for layoff?
D21. Consider which statistical test would justify your opinion in D20, part c,
about whether it looks as if older workers were more likely to be chosen
for layoff .
a. Which tests of significance are possibilities in this situation? Give the
strengths and weaknesses of each test as applied to this scenario. (Dont
worry about conditions for now.) Should you use a one-sided test or a
two-sided test?
b. Using the data from your table in D20, part c, find the P-value for each
of the tests in part a and compare them.
c. Select the test you think is most appropriate. What is your conclusion
if you take the test at face value (that is, if you dont worry about the
conditions being met)?
22-03-2009 21:29
Lesson
3 de 7
Conditions Rarely Match Reality

The major differences between reality and a model are often the basis of heated
arguments between opposing lawyers in discrimination cases. One test you might
have chosen in D21 is the z-test for the difference between two proportions. This
test is based on several assumptions.
Independent Random Samples
A simple random sample of size n1 is taken from a large population with
proportion of successes p1. A second, independent random sample of size n2 is
taken from a large population with proportion of successes p2.
Large Samples from Even Larger Populations
In the z-test, you use the normal distribution to approximate the sampling
distribution of
, where is the proportion of successes in the first sample
and is the proportion of successes in the second sample. This approximation is
reasonable provided that all of n1 , n1(1 ), n2 and n2(1 ) are at least 5
and that each population is at least 10 times the sample size. On the surface, the
mismatch between the two sets of conditions is striking:
This mismatch does not make the test invalid, however. You can still use the
test to answer this question: If the process had been random, how likely would it
have been to get a difference in proportions as big as the one Westvaco got just by
chance? As long as it is made clear that this is the question being answered, the
test is valid and can be very informative.
So you can proceed with a significance test, but you must make the limitations
of what you are doing very clear. If you reject the null hypothesis, all you can
conclude is that something happened that cant reasonably be attributed to
chance alone.
Another alternative is to use Fishers exact test, in which the sampling
distribution of the difference of two proportions is constructed exactly rather than
relying on a normal approximation. Fishers exact test requires few assumptions
and uses no approximations. In contrast, the z-test for the difference between two
proportions is an approximation and requires strong assumptions. So why dont
we always use Fishers exact test? We can, but it requires some computing power.
You saw a simulation of this test in E19 in Chapter 1. As technology becomes
more powerful, statisticians increasingly are turning to methods such as Fishers
exact test rather than using approximations based on the normal distribution.
819
22-03-2009 21:29
Lesson
4 de 7
Conditions Rarely Match Reality

D22. Evaluate the conditions necessary for your significance test in D21, part c.
Give a careful statement of the conclusion you can make.
D23. Agree or disagree, and tell why: A hypothesis test is based on a probability
model. Like all probability models, it assumes certain outcomes are random.
But in the Westvaco case, the decisions about which people to lay off werent
random. Theres no probability model, so a statistical test is invalid.
Looking for a Better Approach

The test you did in D21 was based on dividing workers into two age groups,
under 40 and 40 or older. This replaces a quantitative variable (age) with a
categorical variable (age group). The cutoff age that defines who is classified as
an older worker can be arbitrary and, as youll see in P30P35, can change
the results of the analysis. Looking at the mean ages of those laid off and those
retained can help you avoid making such an arbitrary decision. You took that
approach in Chapter 1 when you used the average age of the workers laid off by
Westvaco as the test statistic. Youll explore this approach further in P36P38.
Conditions for a t-Test
A t-test for the difference between two means is based on independent random
samples from two large populations, which must be approximately normal if the
sample sizes are small. If you were to test that the difference between the mean
ages of those laid off and those retained is 0, strict application of the t-test would
require the conditions that those workers terminated and those retained are
independent random samples from two large, approximately normal, populations.
In actual fact, there is only one population that is finite rather than infinite, and
this population is sorted into two groups. Again, the mismatch between the two
sets of conditions is striking:
On the surface, theres no apparent reason to think the t-test is appropriate

for answering the question Can the observed difference in means be attributed
simply to chance, or should we look for another explanation? Remarkably,
though, extensive simulations have shown that despite the mismatch between the
two sets of conditions, the t-test tends to give a good approximation of the P-value
22-03-2009 21:29
Lesson
5 de 7
you would get using an approach that satisfies the actual set of conditions for
situations like the Westvaco case.
You used the simulation approach in Section 1.2 when you selected three
workers for layoff using a completely random process and computed their average
age. After you did this step many times, you compared the average age of workers
actually laid off at Westvaco to your distribution of the average ages of workers
selected at random for layoff. The null hypothesis was that the process used by
Westvaco was equivalent to randomly choosing the employees for layoff from
the ten hourly workers remaining at the beginning of Round 2. This approach
is called the randomization test or permutation test. As with Fishers exact test,
randomization tests avoid the need to approximate the sampling distribution
with a normal distribution. You will be seeing randomization tests more often as
computing becomes more powerful.

D24. Describe how to use the randomization test to determine if an average age as
large or larger than that of the 28 workers laid off at Westvaco, 49.86 years, can
reasonably be attributed to chance. Each student in your class should perform
one trial of your simulation.
D25. A stem-and-leaf plot for the ages of the 50 Westvaco employees is shown
in Display 12.18. One condition for a t-test is that your data come from a
normal distribution. How appropriate is that assumption here? (In what
ways is the shape of the distribution different from normal? How important
do you consider these departures from normal?)
Display 12.18 A stem-and-leaf plot of the ages of the

50 Westvaco employees.
The End of the Story

The planning that led to the layoffs at Westvaco took place in several stages. In the
first stage, the head of the engineering department made a list of 11 employees
to lay off. His boss reviewed the list and decided it was too short: They needed
to reduce the size of their work force even further. The department head added a
second group of people to the list and checked again with his bossstill too few.
He added a third group of names, then a fourth, and finally one more person in
the fifth round of planning. Display 1.1 on page 5 shows this information in the
column headed Round: An entry of 1 means chosen for layoff in Round 1 of the
planning, and similarly for 2, 3, 4, and 5. An entry of 0 means Retained.
821
22-03-2009 21:29
Lesson
6 de 7
As it turned out, older employees fared much worse in the earlier rounds of
the planning than in the later rounds, as you can see in Display 12.19.
Display 12.19 Breakdown of layoffs by round and age group.
Time plots give another view of the same situation. Each plot in Display 12.20
shows the number of employees remaining at Westvaco after each round of layoffs.
Display 12.20 A time plot of the reduction in work force by

round and age group.
The End of the Story

D26. Describe the patterns in Display 12.20. What do they suggest to you?
About all you can ever show by using statistical methods in discrimination
cases is that the process doesnt look like random selection. Statistical analysis
alerts us to questionable situations, but it cannot reconstruct the intent of the
people who did the laying off. Knowing the intent is crucial because it might
be perfectly legal: Perhaps employees in obsolete jobs were the ones picked
for termination, and it just happens that the obsolete jobs were held by older
employees.
The Westvaco case never got as far as a jury. Just before it was about to go
to trial, the two sides agreed on a settlement. Details of such settlements are not
public information, so, like many problems based on statistics, this case has no
final answer.
22-03-2009 21:29
Lesson
7 de 7
Practice
Comparing Termination Rates for Two
Age Groups
P29. Consider two companies: Seniors, Inc.,
with almost all of its 50 employees age
40 or older, and Youth Enterprises, with
roughly half of its 50 employees under age
40. Suppose both companies discriminate
against older workers in a layoff. If you do
a significance test of the difference between
two proportions, will you be more likely to
detect the discrimination at Seniors, Inc., or
at Youth Enterprises? Explain.
P30. Refer to the stem-and-leaf plot of the ages of
the 50 Westvaco employees in Display 12.18
on page 821.
a. Use it to construct a boxplot of the ages.
b. What characteristics of the distribution
can you see from the boxplot that you
could not see from the stem-and-leaf
plot? From the stem-and-leaf plot that
you could not see from the boxplot?
P31. If you use age 40 as the cutoff for defining
your age groups, what percentage of the
workers are in the older age group? If you
use age 50 instead of age 40, what percentage
are in the older age group?
P32. Of the 28 workers who were age 50 or older,
19 were laid off. Use this information to
construct a table similar to that in D20,
part c, but use age 50 as your cutoff age.
P33. Using your table in P32, state appropriate
null and alternative hypotheses and carry out
a test of the null hypothesis. What do you
conclude if you take the results at face value?
P34. Compare your analyses in D21, in which age
40 was the cutoff age, and in P33, in which
age 50 was the cutoff age. Which cutoff
value, 40 or 50, leads to stronger evidence
of discrimination? Which test do you think
is more informative about what actually
happenedusing age 40 as your cutoff age
or using age 50? Explain your reasoning.
P35. According to the U.S. Supreme Court, if
you do a statistical test of discrimination,
you should reject the null hypothesis if your
P-value from a one-sided test is less than or

equal to 0.025 or if your P-value from a
two-sided test is less than or equal to 0.05.
a. According to this standard, should the
null hypothesis be rejected in either of the
two testsusing age 40 as your cutoff (as
in D21), or using age 50 (as in P33)?
b. Which of these statements correctly
completes this phrase: If you use a
P-value of 0.025 to determine guilt, then
I. 2.5% of not guilty companies
would be declared guilty by the test
II. 2.5% of guilty companies would be
declared not guilty by the test
c. Is 2.5% the probability of a Type I or a
Type II error?
d. Describe a Type II error in this situation.
P36. Construct a back-to-back stemplot of the ages
of the Westvaco workers laid off and those
retained. Compare the distributions. (Use split
stems: 2 | for 2024, | for 2529, and so on.)
P37. Use your stemplot from P36 to construct
side-by-side boxplots.
P38. Display 12.21 shows the means and standard
deviations of the ages for the laid-off and
retained workers. Use the summary data to
carry out a two-sample t-test. (Again, dont
worry about conditions for now. Should you
use a one-sided test or a two-sided test?)
Display 12.21 Means and standard deviations

of ages for laid-off and retained
employees.
P39. Compare your two-sample t-test in P38

with the randomization test in D24. How do
the hypotheses, the conditions you need to
check, the sampling distributions, and the
conclusions differ?
823
22-03-2009 21:29
Lesson
1 de 5
Appendix: Statistical Tables
Table entry for z is the

probability lying below z.
TABLE A
824
Standard Normal Probabilities
22-03-2009 21:29
Lesson
2 de 5
Table entry for z is the

probability lying below z.
TABLE A
Standard Normal Probabilities (continued)
Appendix: Statistical Tables
825
22-03-2009 21:29
Lesson
3 de 5
Table entry for p and C is the value of

t* with probability p lying above it
and probability C lying between
t* and t*.
TABLE B
t-Distribution Critical Values
826 Appendix: Statistical Tables
22-03-2009 21:29
Lesson
4 de 5
22-03-2009 21:29
Lesson
5 de 5
TABLE D
828 Appendix: Statistical Tables
Random Digits
22-03-2009 21:29
Lesson
1 de 10
Glossary
Symbols
b0
y-intercept of the sample regression

line
b1
slope of the sample regression line
E
margin of error
E(X)
expected value (mean) of the
random variable X
H0
null hypothesis
Ha
alternative hypothesis (sometimes H1)
IQR
interquartile range
n
sample size or the number of trials
N
population size
p
population proportion or the
probability of a success on any one
trial
p0
hypothesized value of the population
proportion
sample proportion
estimate of the common value of
a population proportion found by
combining two samples
P(A)
probability that event A happens
P(A or B) probability that event A happens
or event B happens or both (also
P(A B))
P(A and B) probability that event A and event B
both happen (also P(A B))
P(B | A)
conditional probability that event B
happens given that event A happens
P(x)
probability that the random variable
X takes on the value x (also P(X = x))
Q1
first or lower quartile
Q3
third or upper quartile
r
sample correlation
s
sample standard deviation; in
regression, the estimate from the
sample of the common variability
of y at each value of x
s2
SD
SE
SRS
SSE
t
Var(X)
x
X
y
z
z*
sample variance
estimated standard error of the slope
of the regression line
standard deviation
standard error
simple random sample
sum of squared errors
a test statistic using a t-distribution
variance of the random variable X
sample mean
observed value of a variable
a random variable
observed value of a variable
predicted value of a variable
standardized value (z-score)
critical value
Greek Letters
significance level; probability of a
Type I error
probability of a Type II error
y-intercept of the population
regression line
slope of the population regression
line
difference of the value of y for a
point and the value predicted by the
population regression line
population mean
the sample mean
mean of the conditional distribution
of y given x
population standard deviation; in
regression, the common variability
of y at each value of x
standard error of the mean
829
22-03-2009 21:30
Lesson
2 de 10
population variance
box-and-whiskers plot
variance of the random variable X
boxplot (or box-and-whiskers plot) A graphical

display of the five-number summary. The box extends
from the lower quartile to the upper quartile, with a
line across it at the median. The whiskers run from
the quartiles to the minimum and maximum.
capture rate
The proportion of confidence intervals
produced by a particular method that capture the
population parameter. See also confidence level.
case
The subject (or unit) on which a measurement
is made.
categorical variable A variable that can be grouped
into categories, such as yes and no. Categories
sometimes can be ordered, such as small, medium,
and large.
census A collection of measurements on all units in
the population of interest.
Central Limit Theorem
The shape of the sampling
distribution of the sample mean becomes more normal
as n increases.
chance model See probability model.
test statistic for chi-square tests

Addition Rule For any two events A and B,
P(A or B) = P(A) + P(B) P(A and B). If events A
and B are disjoint, then P(A or B) = P(A) + P(B).
adjacent values On a modified boxplot, the largest
and smallest non-outliers.
alternative hypothesis The set of values, as compared
to that of the null hypothesis, that an investigator believes
might contain the plausible values of a population
parameter in a test of significance. Sometimes called
the research hypothesis.
average
See mean.
balanced design An experimental design in which

each treatment is assigned to the same number of units.
bar chart (or bar graph) A plot that shows frequencies
for categorical data as heights or lengths of bars, with
one bar for each category.
bias
The difference between the actual value of a
parameter being estimated and the average value, over
repeated sampling, of an estimator of that parameter.
bias due to sampling See sample selection bias.
bimodal Describes a distribution with two
well-defined peaks.
binomial distribution A distribution of the random
variable X where X represents the number of successes
in n independent trials, with the probability of a success
the same on each trial.
bins
The intervals on the real number line that
determine the width of the bars of a histogram.
bivariate data Data that involve two variables per
case. For quantitative variables, often displayed on a
scatterplot. For categorical variables, often displayed in
a two-way table.
blind Describes an experiment in which the subjects
do not know which treatment they received.
blocking
The process of setting up an experiment by
dividing the units into groups (blocks) of similar units
and then assigning the treatments at random within
each block.
blocks In an experiment, groups of similar units, with
treatments randomly assigned within these groups.
830 Glossary
See boxplot.
chi-square test of homogeneity A chi-square test

used to determine whether it is reasonable to believe
that when several different populations are broken
down into the same categories, they have the same
proportion of units in each category.
chi-square test of independence A chi-square test of
the hypothesis that two categorical variables measured
on the same units are independent of each other in the
population.
clinical trial A randomized experiment comparing
the effects of medical treatments on human subjects.
cluster(s) On a plot, a group of data clustering
close to the same value, away from other groups. In
sampling, non-overlapping and exhaustive groupings of
the units in a population.
cluster sampling Selecting a simple random sample
of clusters of units (such as classrooms of students)
rather than individual units (students)
coeffcient of determination
The square of the
correlation r. Tells the proportion of the total variation
in y that can be explained by the relationship with x.
column chart A three-dimensional plot of frequencies
taken from a two-way table, which depicts those
frequencies as heights of columns.
22-03-2009 21:30
Lesson
3 de 10
column marginal frequency

The total of all
frequencies across row categories for a particular
column of a two-way table of frequencies for
categorical variables.
critical value
The value to which a test statistic is
compared in order to decide whether to reject the null
hypothesis. Or, the multiplier used in computing the
margin of error for a confidence interval.
comparison group In an experiment, a group that

receives one of the treatments, often the standard
treatment.
cumulative percentage plot

frequency plot.
complement In probability, the outcomes in the

sample space that lie outside an event of interest.
completely randomized design An experimental
design in which treatments are randomly assigned to
units without restriction.
conditional distribution of y given x With bivariate
data, the distribution of the values of y for a fixed
value of x.
conditional probability The notion that a probability
can change if you are given additional information. The
conditional probability that event A happens given that
event B happens is given by
as P(B) > 0.
as long
conditional relative frequency The joint frequency

in a column divided by the marginal frequency for that
column, or the joint frequency in a row divided by the
marginal frequency for that row.
confidence interval A set of plausible values for a
population parameter, any one of which could be used
to define a population for which the observed sample
statistic would be a reasonably likely outcome.
confidence level
The probability that the method
used will give a confidence interval that captures the
parameter.
confounding variables (or confounding) Two
variables in an observational study whose effects on
the response are impossible to separate.
continuous variable A quantitative variable that can
take on any value in an interval of real numbers.
control group In an experiment, a group that
provides a standard for comparison to evaluate the
effectiveness of a treatment; often given a placebo.
convenience sample A sample in which the units
chosen from the population are the units that are easy
(convenient) to include, rather than being selected
randomly.
correlation A numerical value between 1 and 1,
inclusive, that measures the strength and direction of a
linear relationship between two variables.
See cumulative relative
cumulative relative frequency plot (or cumulative

percentage plot) A plot of ordered pairs in which
each value x in the distribution and its cumulative
relative frequency, that is, the proportion of all values
less than or equal to x, are plotted.
data A set of numbers or observations with a context
and drawn from a real-life sample or population.
data analysis
See statistics.
degrees of freedom
The number of freely varying
pieces of information on which an estimator is
based. For example, when using a sample to estimate
the variability in the population, the number of
independent deviations from the estimate of center.
dependent events
Events that are not independent.
deviation
The difference from the mean, x
from some other measure of center.
, or
disjoint events (or mutually exclusive events) Events

that cannot occur on the same opportunity. If event A
and event B are disjoint, P(A and B) = 0.
distribution, data
The set of values that a variable
takes on in a sample or population, together with how
frequently each value occurs.
distribution, probability The set of values that a
random variable takes on, together with a means of
determining the probability of each value (or interval of
values in the case of a continuous distribution).
dot plot A graphical display that shows the values of
a variable along a number line.
double-blind Describes an experiment in which
neither the subjects nor the researcher making the
measurements knows which treatment the subjects
received.
ecological fallacy The mistake of using ecological
(group-level) correlations to support conclusions about
individuals.
event Any subset of a sample space.
expected value, X or E(X)
The mean of the probability
distribution for the random variable X.
experimental units In an experiment, the subjects or
objects to which treatments are assigned.
Glossary 831
22-03-2009 21:30
Lesson
4 de 10
explanatory variable (or predictor) A variable

used to predict (or explain) the value of the response
variable. Placed on the x-axis in a regression analysis.
exploratory analysis (or data exploration) An
investigation to find patterns in data, using tools such
as tables, statistical graphics, and summary statistics to
display and summarize distributions.
exponential relationship A relationship between
two variables in which the response variable, y, is
multiplied by a constant for each unit of increase in the
explanatory variable, x. Mathematically, y = ab x where
a and b are constants.
extrapolation Making a prediction when the value of
the explanatory variable, x, falls outside the range of the
observed data.
factor An explanatory variable, usually categorical, in
a randomized experiment or an observational study.
first quartile, Q1 See lower quartile.
fitted value See predicted value.
five-number summary A data summary that lists the
minimum and maximum values, the median, and the
lower and upper quartiles for a data set.
fixed-level test A test in which the null hypothesis is
rejected or not rejected based on comparison of the test
statistic with the critical value for some predetermined
level of significance.
frame See sampling frame.
frequency (or count)
The number of times a value
occurs in a distribution. With categorical data, the
number of units that fall into a specific category.
frequency table A table that gives data values and
their frequencies.
Fundamental Principle of Counting If there are
k stages in a process, with ni possible outcomes for
stage i, then the number of possible outcomes for all
k stages taken together is n 1n 2n 3 n k.
gap On a plot, the space that separates clusters
of data.
geometric (waiting-time) distribution
The
distribution of the random variable X in which X
represents the number of trials needed to get the first
success in a series of independent trials, where the
probability of a success is the same on each trial.
goodness-of-fit test A chi-square test used to
determine whether it is reasonable to assume that
a sample came from a population in which, for
832 Glossary
each category, the proportion of outcomes in the

population that fall into that category is equal to some
hypothesized proportion.
heteroscedasticity The tendency of points on a
scatterplot to fan out at one end, indicating that the
relationship varies in strength.
histogram A plot of a quantitative variable that
groups cases into rectangles or bars. The height of the
bar shows the frequency of measurements within the
interval (or bin) covered by the bar.
homogeneous populations Two or more populations
that have nearly equal proportions of units in each
category of study.
hypothesis test
See test of significance.
incorrect response bias A bias resulting from

responses that are systematically wrong, such as from
intentional lying, inaccurate measurement devices,
faulty memories, or misinterpretation of questions.
independent events Events A and B for which the
probability of event A happening doesnt depend
on whether event B happens. Events A and B are
independent if and only if P(A | B) = P(A) or,
equivalently, P(B | A) = P(B) or, equivalently,
P(A and B) = P(A) P(B).
inference (or inferential statistics) Using results
from a random sample to draw conclusions about
a population or using results from a randomized
experiment to compare treatments.
influential point On a scatterplot, a point that
strongly influences the regression equation and
correlation. To judge a points influence, you compare
the regression equation and correlation computed first
with and then without the point.
interpolation Making a prediction when the value of
the explanatory variable, x, falls inside the range of the
observed data.
interquartile range, IQR A measure of spread equal
to the distance between the upper and lower quartiles;
IQR = Q3 Q1 .
joint frequency
The frequency within a particular
cell of a two-way table of frequencies for categorical
variables.
judgment sample A sample selected using the
judgment of an expert to choose units that he or she
considers representative of a population.
Law of Large Numbers A theorem that guarantees
that the proportion of successes in a random sample
22-03-2009 21:30
Lesson
5 de 10
will converge to the population proportion of

successes as the sample size increases. In other words,
the difference between a sample proportion and a
population proportion must get smaller (except in
rare instances) as the sample size gets larger, if the
sample is randomly selected from that population.
place where you would put a pencil point below the

horizontal axis in order to balance the distribution.
measure of center A single-number summary that
measures the center of a distribution; usually the
mean (or average). Median, midrange, mode, and
trimmed mean are other measures of center.
least squares line See regression line.

level One of the values or categories making up a
factor.
measure of spread(or measure of variability) A

single-number summary that measures the variability
of a distribution. Range, IQR, standard deviation, and
variance are measures of spread.
median A measure of center that is the value that
divides an ordered set of values into two equal halves.
To find it, you list all the values in order and select
the middle one or, if the number of values is even, the
average of the two middle ones. If there are n values,
the median is at position (n +1) / 2. On a plot of a
distribution, the median is the value that divides the area
between the distribution curve and the x-axis in half.
method of least squares A general approach to
fitting functions to data by minimizing the sum of the
squared residuals (or errors).
midrange The midpoint between the minimum and
maximum values in a data set, or (max + min) / 2.
level of significance,
The maximum P-value for
which the null hypothesis will be rejected.
line of averages See line of means.
line of means(or line of averages) Another term for
the regression line, if points form an elliptical cloud.
In theory, the population regression line contains the
means (expected values) of the conditional distribution
of y at each value of x.
linear shape The characteristic of an elliptical
cloud of points where the means of the conditional
distributions of y given x tend to fall along a line.
lower quartile(or first quartile, Q1) In a distribution,
the value that separates the lower quarter of values from
the upper three-quarters of values. The median of the
lower half of all the values.
lurking variable A variable other than those being
plotted that possibly can cause or help explain the
behavior of the pattern on a scatterplot. More generally,
a variable that is not included in the analysis but, once
identified, could help explain the relationship between
the other variables.
margin of error, E Half the length of a confidence
interval; E = (critical value) (standard error).
marginal frequency The total of the joint frequencies
across row categories for a given column or across
column categories for a given row of a two-way table of
frequencies for categorical variables.
marginal relative frequency The marginal frequency
of a two-way table of categorical data divided by the
total frequency (number of units represented in the
table).
matched pairs design See randomized paired
comparison design.
maximum The largest value in a data set.
mean, x A measure of center, often called the
average, computed by adding all the values of x and
dividing by the number of values, n. On a plot, the
minimum The smallest value in a data set.

mode A measure of center that is the value with
the highest frequency in a distribution. On a plot of a
distribution, it occurs at the highest (maximum) peak.
modified boxplot A graphical display like the basic
boxplot except that the whiskers extend only as far as
the largest and smallest non-outliers (sometimes called
adjacent values) and any outliers appear as individual
dots or other symbols.
Multiplication Rule For any two events A and B,
P(A and B) = P(A) P(B | A) = P(B) P(A | B).
If events A and B are independent, then
P(A and B) = P(A) P(B).
mutually exclusive events See disjoint events.
negative trend The tendency of a cloud of points
to slope downward as you go from left to right, or the
tendency of the value of y to get smaller as the value of
x gets larger.
nonresponse bias A bias that can occur when people
selected for the sample do not respond to the survey.
normal distribution A useful probability distribution
that has a symmetric bell or mound shape and tails
extending infinitely far in both directions.
Glossary
833
22-03-2009 21:30
Lesson
6 de 10
null hypothesis The standard or status quo value

of a parameter that is assumed to be true in a test of
significance until possibly refuted by the data in favor
of an alternative hypothesis.
observational study A study in which the conditions
of interest are already built into the units being studied
and are not randomly assigned.
one-sided (one-tailed) test of significance A test
in which the P-value is computed from one tail of the
sampling distribution. Used when the investigator
has an indication of which way any deviation from
the standard should go, as reflected in the alternative
hypothesis.
outlier A value that stands apart from the bulk of
the data.
parameter A summary number describing a
population or a probability distribution.
percentile The quantity associated with any specific
value in a univariate distribution that gives the
percentage of values in the distribution that are equal
to or below that specific value. The median is the 50th
percentile.
placebo A nontreatment that mimics the treatment(s)
being studied in all essential ways except that it does not
involve the crucial component.
placebo effect The phenomenon that when people
believe they are receiving the special treatment, they
tend to do better even if they are receiving the placebo.
plot of distribution (or graphical display or statistical
graphic) A graphical display of the distribution of
a variable that provides a sense of the distributions
point estimator A statistic from a sample that
provides a single point (number) as a plausible value of
a population parameter.
point of averages The point (x, y), where x is the
mean of the explanatory variable and y is the mean of
the response variable. This point falls on the regression
line.
pooled estimate The weighted average of two
statistics estimating the same parameter, with the
weights usually determined by the sample sizes or
degrees of freedom.
population The entire set of people or things (units)
that you want to know about.
population regression line See true regression line.
834 Glossary
population size
The number of units in the
population.
population standard deviation,
See standard
deviation of a population.
positive trend
The tendency of a cloud of points
to slope upward as you go from left to right, or the
tendency of the value of y to get larger as the value of x
gets larger.
power of a test
The probability of rejecting the null
hypothesis.
power relationship A relationship between
two variables in which the response variable, y, is
proportional to the explanatory variable, x, raised to
a power. Mathematically, y = ax b, where a and b are
constants.
predicted (or fitted) value An estimated value of the
response variable calculated from the known value of
the explanatory variable, x, often by using a regression
equation.
prediction error The difference between the actual
value of y and the value of y predicted from a regression
line. Usually unknown except for the points used to
construct the regression line, whose prediction errors
are called residuals.
predictor See explanatory variable.
probability A number between 0 and 1, inclusive
(or between 0% and 100%), that measures how likely
it is for a chance event to happen. At one extreme,
events that cant happen have probability 0. At the
other extreme, events that are certain to happen have
probability 1.
probability density A probability distribution,
such as the normal or distribution, where x is a
continuous variable and probabilities are identified as
areas under a curve.
probability distribution See distribution,
probability.
probability model (or chance model) A description
that approximatesor simulatesthe random behavior
of a real situation, often by giving a description of all
possible outcomes with an assignment of probabilities.
probability sample A sample in which each unit in
the population has a known probability of ending up in
the sample.
protocol A written statement telling exactly how an
experiment is to be designed and conducted.
22-03-2009 21:30
Lesson
7 de 10
P-value For a test, the probability of seeing a result

from a random sample that is as extreme as or more
extreme than the one computed from the random
sample, if the null hypothesis is true. (Sometimes called
the observed significance level.)
quantitative variable (or numerical variable)
A variable that takes on numerical values.
quartiles Three numbers that divide an ordered set
of data values into four groups of equal size.
questionnaire bias Bias that arises from how the
interviewer asks and words the survey questions.
random sample A sample in which individuals are
selected by some chance process. Sometimes used
synonymously with simple random sample.
random variable A variable that takes on numerical
values determined by a chance process.
randomization (or random assignment) Assigning
subjects to different treatment groups using a random
procedure.
randomized block design An experimental design
in which similar units are grouped into blocks and
treatments are then randomly assigned to units within
each block.
randomized comparative experiment An
experiment in which two or more treatments are
randomly assigned to experimental units for the
purpose of making comparisons among treatments.
randomized paired comparison (matched pairs)
An experimental design in which two different
treatments are randomly assigned within pairs of
similar units.
randomized paired comparison (repeated measures)
An experimental design in which each treatment is
assigned (in random order) to each unit.
range A measure of spread equal to the difference
between the maximum and minimum values in a
data set.
rare events Values or outcomes that lie in the outer
5% of a distribution or in the upper 2.5% and lower
2.5% of a distribution. Compare reasonably likely.
reasonably likely Describes values or outcomes that
lie in the middle 95% of a distribution. Compare rare
events.
recentering Adding the same number c to all the
values in a distribution. This procedure doesnt change
the shape or spread but slides the entire distribution by

the amount c, adding c to the measures of center.
rectangular distribution A distribution in which all
values occur equally often.
regression The statistical study of the relationship
between two (or more) quantitative variables, such
as fitting a line to bivariate data. (Can be extended to
categorical variables.)
regression effect (or regression toward the mean)
On a scatterplot, the difference between the regression
line and the major axis of the elliptical cloud.
regression line (or least squares line or least squares
regression line) The line for which the sum of squared
errors (residuals), SSE, is as small as possible.
regression toward the mean See regression effect.
relative frequency A proportion computed by
dividing a frequency by the number of values in the
data set.
relative frequency histogram A histogram in which
the length of each bar shows proportions (or relative
frequencies) instead of frequencies.
repeated measures design See randomized paired
comparison (repeated measures).
replication Repetition of the same treatment on
different units.
Multiplying all the values in a distribution
rescaling
by the same nonzero number d. This process doesnt
change the basic shape but instead stretches or shrinks
the distribution, multiplying the IQR and standard
deviation by |d| and multiplying the measures of center
by d.
residual (or error) For points used to construct the
regression line, the difference between the observed
value of y and the predicted value of y, that is, y y.
residual plot A scatterplot of residuals, y y, versus
predictor values, x, or versus predicted values, y. A
diagnostic plot used to uncover nonlinear trends in a
relationship between two variables.
resistant to outliers Describes a summary statistic
that does not change very much when an outlier is
removed from the data set.
response variable The outcome variable used
to compare results of different treatments in an
experiment or the outcome variable that is predicted
by the explanatory variable or variables in regression
analysis. Placed on the y-axis in a regression analysis.
Glossary
835
22-03-2009 21:30
Lesson
8 de 10
robustness The comparative insensitivity of a

statistical procedure to departure from the assumptions
on which the procedure is based.
row marginal frequency The total of all frequencies
across column categories for a particular row of a twoway table of frequencies for categorical variables.
sample The set of units selected for study from the
population.
sample selection bias (or sampling bias or bias
due to sampling) The extent to which a sampling
procedure produces samples for which the estimate
from the sample is larger or smaller, on average, than
the population parameter being estimated.
sample space A complete list or description of
disjoint (mutually exclusive) outcomes of a chance
process.
sampling bias
See sample selection bias.
sampling distribution The distribution of a sample

statistic under some prescribed method of probability
sampling.
sampling distribution of a sample proportion,
The theoretical distribution of the sample proportion in
repeated random sampling.
shape One of the characteristics, along with center

and spread, that is used to describe distributions.
Univariate distributions sometimes have a standard
shape such as normal, uniform, or skewed. Bivariate
distributions may form an elliptical cloud. Descriptions
of shape should consider possible outliers, clusters,
and gaps.
simple random sample, SRS A sample generated
from a sampling procedure in which all possible
samples of a given fixed size are equally likely.
simulation A procedure that uses a probability
model to imitate a real situation. Often used to
compare an actual result with the results that are
reasonable to expect from random behavior.
size bias A type of sample selection bias that gives
units with a larger value of the variable a higher chance
of being selected.
skewed Describes distributions that show bunching
at one end and a long tail stretching out in the other
direction. Often happens because the values bump up
against a wall and hit either a minimum that values
cant go below or a maximum that values cant go
above.
skewed left A skewed distribution with a tail that
stretches left, toward the smaller values.
sampling distribution of the sample mean,

The
theoretical distribution of the sample mean in repeated
random sampling.
sampling frame (or frame) The listing of units from
which the sample is actually selected.
sampling with replacement In sequential sampling
of units from a population, a procedure in which each
sampled unit is placed back into the population before
the next unit is selected.
slope
For linear relationships, the change in y (rise)
per unit change in x (run).
sampling without replacement In sequential

sampling of units from a population, a procedure in
which each sampled unit is not placed back into the
population before the next unit is selected.
spread
scatterplot A plot that shows the relationship

between two quantitative variables, usually with each
case represented by a dot.
segmented bar graph (or stacked bar graph) A plot
in which categorical frequencies are stacked on top of
one another.
sensitive to outliers Describes a summary statistic
that changes considerably when an outlier is removed
from the data set.
836 Glossary
skewed right A skewed distribution with a tail that

stretches right, toward the larger values.
split stem A stem-and-leaf plot in which the leaves

for each stem are split onto two or more lines. For
example, if the second digit is 0, 1, 2, 3, or 4, it is placed
on the first line for that stem. If the second digit is 5, 6,
7, 8, or 9, it is placed on the second line for that stem.
See variability.
stacked bar graph See segmented bar graph.

standard deviation of a population,
A measure
of spread equal to the square root of the sum of the
squared deviations divided by n. For a probability
distribution, it is the square root of the expected
squared deviation from the mean.
standard deviation of a sample, s A measure of
spread equal to the square root of the sum of the
squared deviations divided by n 1.
standard error
distribution.
The standard deviation of a sampling
22-03-2009 21:30
Lesson
9 de 10
standard error of the mean,

The standard
deviation of the sampling distribution of x or
standard error of the mean ( estimated )
The
estimated standard deviation of the sampling
distribution of x or
standard normal distribution A normal distribution
with mean 0 and standard deviation 1. The variable
along the horizontal axis is called a z-score.
standard units, z The number of standard deviations
a given value lies above or below the mean:
standardizing Converting to standard units; the

two-step process of recentering and rescaling that
turns any normal distribution into a standard normal
distribution.
statistic A summary number calculated from a
sample taken from a population. For example, the
sample mean, x, and standard deviation, s, are statistics.
statistically significant Describes the situation
when the difference between the estimate from the
sample and the hypothesized parameter is too big to
reasonably be attributed to chance variation.
statistics (or data analysis) The study of the production,
summarization, and analysis of data, along with the
processes for drawing conclusions from the data.
stem-and-leaf plot (or stemplot) A graphical display
with stems showing the left most digit of the values
separated from leaves showing the next digit or set
of digits.
stemplot See stem-and-leaf plot.
strata (singular, stratum) Subgroups of the population,
usually selected for homogeneity or sampling
convenience, that cover the entire population. See also
stratified random sampling.
stratification A classification of the units in a
population into homogeneous subgroups, known as
strata, prior to sampling.
stratified random sampling Stratifying the
population and then taking a simple random sample
from within each stratum.
strength In the context of regression analysis, two
variables are said to have a strong relationship if there
is little variation around the regression line. If there
is a lot of variation around the regression line, the
relationship is weak.
sum of squared errors, SSE

residuals:
The sum of the squared
summary statistic See statistic.

systematic sampling with random start A sample
selected by taking every nth member of the population,
starting at a random spotfor example, having people
count off and then picking one of the numbers at
random.
table of random digits A string of digits constructed in
such a way that each digit, 0 through 9, has probability
of being selected and each digit is selected
independently of the previous digits.
t-distribution The distribution, for example, of the
statistic below, when the data are a random sample
from a normally distributed population:
test of significance (or hypothesis test) A procedure

that compares the results from a sample to some
predetermined standard in order to decide whether the
standard should be rejected.
test statistic Typically, in significance testing, the
distance between the estimate from the sample and the
hypothesized parameter, measured in standard errors.
third quartile, Q3 See upper quartile.
treatment group In an experiment, a group that
receives an actual treatment being studied. Compare
with control group.
treatments Conditions assigned to different groups
of subjects to determine whether subjects respond
differently to different conditions.
tree diagram A diagram used to calculate
probabilities for sequential events.
trend On a scatterplot, the path of the means of the
vertical strips (conditional distributions of y given x)
as you move from left to right. More simply, the path
taken by a line through the center of the data in a
scatterplot.
true regression line (or population regression
line) The regression line that would be computed if
you had the entire population. Theoretically, the line
through the means of the conditional distributions of y
given x. See also line of means.
Glossary
837
22-03-2009 21:30
Lesson
10 de 10
t-test A test of significance of a population mean (or

comparison of means) using the t-distribution. See also
test of significance and t-distribution.
two-sided (two-tailed) test of significance A test
in which the P-value is computed from both tails of
the sampling distribution. Used if the investigator is
interested in detecting a change from the standard in
either direction.
two-stage sampling A sampling procedure that
involves two steps. For example, taking a random
sample of clusters and then taking a random sample
from each of those clusters.
two-way table A table of frequencies that lists
outcomes in the cells formed by the cross-classification
of two categorical variables measured on the same units.
The error made when the null
Type I error
hypothesis is true and you reject it.
The error made when the null
Type II error
hypothesis is false and you fail to reject it.
unbiased Describes an estimator (statistic) that has
an average value in repeated sampling (expected value)
equal to the parameter it is estimating.
uniform distribution A distribution whose frequencies
are constant across the possible values. Its plot is
rectangular.
unimodal Describes a distribution of univariate data
with only one well-defined peak.
units Individuals that make up the population from
which samples may be selected or to which treatments
may be applied.
upper quartile (or third quartile, Q3) In a

distribution, the value that separates the lower
three-quarters of values from the upper quarter of
values. The median of the upper half of all the values.
variability (or spread) The degree to which values
in a distribution differ. Measures of variability for
quantitative variables include the standard deviation,
variance, interquartile range, and range.
variability due to sampling (or variation in
sampling) A description of how an estimate varies
from sample to sample.
variable A characteristic that differs from case to case
and defines what is to be measured or classified.
variance A measure of spread equal to the square of
the standard deviation.
voluntary response bias The situation in which
statistics from samples are not fair estimates of
population parameters because the sample data came
from volunteers rather than from randomly selected
respondents.
voluntary response sample A sample made up of
people who volunteer to be in it.
waiting-time problems Problems in which the
variable in question is the number of trials you have
to wait until the event of interest happens. See also
geometric (waiting time) distribution.
z-score See standard normal distribution and
standard units.
univariate data Data that involve a single variable

per case. A quantitative variable often is displayed on a
histogram. A categorical variable often is displayed on
a bar chart.
838 Glossary
22-03-2009 21:30
Lesson
1 de 44

The answers below are not complete solutions but are
meant to help you judge whether you are on the right
track. If you round computations in intermediate steps
or use tables rather than a calculator, your numerical
answers might not match those given exactly.
Chapter 1
Section 1.1
P1.
P2.
P3.
E1.
E3.
E5.
E7.
Older hourly workers were far more likely to be

laid off in Rounds 13 than were younger hourly
workers.
a.
b.
c. A higher proportion of those age 50 and older
were laid off than those under age 50 (0.875
versus 0.50).
d. Hourly workers. The difference in the
proportions for salaried workers (0.60 versus
0.375) is smaller than in part c. But note the
small number of hourly workers.
hourly, although the patterns are similar
a.
b.
c. A higher proportion of those age 40 and older
were laid off than those under age 40 (0.52
versus 0.44).
d. age 50, because the difference in proportions
is greater
a. The hourly workers who kept their jobs
tended to be younger than the salaried
workers who kept their jobs.
b. no ( Salaried workers tended to be older even
before layoffs. )
b. The percentage of those laid off who were
age 40 or older, by round, were 82%, 89%,
67%, 50%, and 0%. Most layoffs came early,
and older workers were hit harder in earlier
rounds.
a. .267; 626 or 627; 180
b.
Section 1.2
P4. a. about 37 out of 200, or 0.185
b. An average age this high would be easy to get
just by chance, so there is no evidence of age
discrimination.
P5. a. 48.6
b. Write 14 ages on cards. Draw 10 at random
and find the average age. Repeat many times
and find where 48.6 falls in the distribution.
c. about 45 out of 200, or 0.225
d. no
E9. b. 24 out of 50, or 0.48
c. The proportion can reasonably be attributed
to chance.
E11. C
d. weak
E13. a. 45
b. 4
c.
Chapter Summary
E15. b. There is no reason to look for an explanation,
because there isnt much difference in centers
or spreads.
E17. B
E19. a. B
d. 13%; no
E21. a. 1001
b. 5, 6, 7, 8, or 9
c. i. 360 ii. 90 iii. 5
d.
E23. a.
c. yes
d.
Chapter 2
Section 2.1
P1. a. 1
b. 0.5, 1, and 1.5
c. 0.5 and 1.5
d. 15%
e. 0.05 and 1.95
P2. The number of deaths per month is fairly
uniform, with about 190,000200,000 per month.
Summer months have the smallest numbers of
deaths, and winter months the largest.
839
22-03-2009 21:30
Lesson
2 de 44
P3. a. A typical SAT math score is roughly 500, give

or take about 100 or so.
b. A typical ACT score is about 20, give or take
5 or so.
c. A typical college-age woman is about
65 inches tall, give or take 2.5 inches or so.
d. A typical professional baseball player in the
1910s had a single-season batting average
of about .260 or .270, give or take about
.040 or so.
P4. The middle 50% of students had GPAs between
2.9 and 3.7, with half above 3.35 and half below.
P5. a. IV b. II c. V d. III e. I
E1. a. skewed left b. skewed right
c. approximately normal d. skewed right
E5. a. each of the approximately 92 officers; the age
at which the officer became a colonel
b. This distribution is skewed left, with no
outliers, gaps, or clusters. The middle 50%
of the ages are between 50 and 53, with half
above 52 and half below.
c. mandatory retirement, age discrimination,
an up or out rule by which if you havent
been promoted beyond colonel by your 55th
birthday you must retire
E7. approximately normal; too many outliers
E9 a. For example, if a case is a business in the
United States and the variable is the number
of employees, the distribution will be skewed
right. There would be a wall at 1, because
thats the smallest number of employees a
business could have (and many businesses
have only one employee).
b. For example, if a case is an AP Statistics class
and the variable is the percentage of students
who did their homework last night, the
distribution will be skewed left. There
would be a wall at 100%, because in most
AP Statistics classes almost all students do
their homework.
E11. Births tend to be more frequent in the summer.
E13. a. Within each cluster on either side of the gap
from 18,000 to 27,000, the plot is skewed
right. If the two clusters are combined (which
might not be a good idea), the median
is 12,350, with the middle 50% of values
between 2,533 and 30,355.
840 Brief Answers to Selected Problems
b. Norway and Switzerland; no, they are part of

the thinning tail.
c. higher cluster; Eastern Europe, Asia, and the
Middle East
d. not when economic conditions of different
continents are different
Section 2.2
P6. Quantitative: year of birth, year of hire, and age;
categorical: row number, job title, round, and pay
category. Month of birth and month of hire are
best called ordered categories.
P7. The distribution is skewed right, with no obvious
gaps or clusters and a wall at 0. The elephant
is the only possible outlier. About half the
mammals have gestation periods of more than
160 days, and half less. The middle 50% have
gestation periods between 63 and 284 days. Large
mammals have longer gestation periods.
P8. The average longevity distribution is skewed
right, with two possible outliers, while the
distribution of maximum longevity is more
uniform but has a peak at 2030 years and a
possible outlier. The center and spread of the
distribution of maximum longevity are larger.
P9. no
P10. about 0.15; about 30; skewed left, with median
between 70 and 75 and the middle 50% between
about 60 and 75
P11. See P8.
P12. The cases are the individual males in the labor
force age 25 and older. The variable is their
educational attainment. The proportion increases
through the first three levels with a huge jump
at the high school graduation level. Then it
decreases except for a spike at bachelors degree.
The distributions for males and females are
similar in shape. Relative frequency bar charts
account for the different numbers of males and
females.
P13. Westvaco laid off the majority of workers in
Rounds 1 and 2.
E15. a. collected by a statistics class; a penny; age of
the penny
b. The shape is strongly skewed right, with
a wall at 0. The median is 8 years, and the
22-03-2009 21:30
Lesson
3 de 44
spread is quite large, with the middle 50% of

ages falling between 3 and 15 years; however,
it is not unusual to see a penny that is more
than 30 years old.
c. If the same number of pennies is produced
each year and if a penny has the same chance
of going out of circulation each year, then
the height of each column would be a fixed
percentage of the previous height.
E17. I. D
II. A
III. C
IV. B
E19. a. Twenty percent (or 0.20) of the class got an A.
b. About 0.47 of the people at the concert
bought a T-shirt.
E21. b. about 70.2 in.; about 2.8 in.
c. about 0.93
d. about 0.16
e. The distribution isnt smooth and has a
domain of only 60.5 to 78.5.
E23. c. Both distributions are slightly skewed right,
with possible outliers on the high end. The
median of both distributions is 12, but the
spread of the distribution of values is larger
for wild mammals.
E25. a. the number of nonpredators that fall into the
categories Domesticated and Wild, and the
total number of nonpredators
b. wild, because the second bar is taller than the
first
c. predator, because the second bar is a larger
fraction of the third bar than is the case for
nonpredators
E27. a. 12
b. two-tenths of 18, or 3.6
c. The speeds must be estimates for the wild
mammals.
Section 2.3
P14. a. 2.5; 2.5
b. 3; 3
c. 3.5; 3.5
d. 49.5; 49.5
e. 50; 50
P15. about 4 ft 4 in.; about 4 ft
P16. a. Africa: 50; Europe: 80
b. For Africa, the median is smaller than the
mean because of the right skew. For Europe,
the mean is slightly lower than the median
because of the left skew.
P17. a. 2 and 5; 3
b. 2 and 6; 4
c. 2.5 and 6.5; 4
d. 2.5 and 7.5; 5
P18. a. predators: 12, 7 and 15; nonpredators: 12,

8 and 15
b. The distributions are centered at exactly the
same place and have about the same spread,
but the distribution for nonpredators has two
outliers on the high side. Both are essentially
mound-shaped.
P19. a. about 76 million
b. Half the programs had more than 10 million
viewers, and half had fewer.
P21. a. min: 1; Q1: 8; median: 12; Q3: 15; max: 41
b. 7
c. 2.5; no outliers on the low end
d. Outliers are elephant at 35 years and
hippopotamus at 41 years; grizzly bear at
25 years is the largest non-outlier.
P22. The SD is 3.21.
P23. a. i
b. iii
c. iv
d. vii
e. ii
f. v
g. vi
P24. a. skewed right
b. 3
c. 3.24; 1.89
P25. a. 2.3; 1.84
b. 2 children
c. positions: 25.5 and 75.5; 2 and 4, with IQR 2;
1 and 3, with IQR 2
d. The median number of children for 1997 is
two, one fewer than in 1967; the mean also
went down by about one child per family,
from 3.2 to 2.3. The distributions kept the
same shape and about the same spread.
E29. 10
E31. a. boxplot III
b. boxplot I c. boxplot II
E33. The back-to-back stemplot is better because there
are only a few values.
E35. laid-off workers: 53.5, 42 and 61; retained
workers: 48, 37 and 55. The median age of the
workers laid off was 5.5 years greater than the
age of those retained, but the distributions have
about the same shape and about the same IQR.
E37. a. II; III
b. II and III
E39. heights of basketball players; heights of all athletes
E41. a. 39.29; 30.425; yes, because it is computed
using only the maximum and minimum
b. 10.54
E43. a. 3.11 g
b. 0.043 g
c. yes
E45. a. 72.9
b. Finding the median of the combined groups
requires having the ordered values.
841
22-03-2009 21:30
Lesson
4 de 44
Section 2.4
P26. a. Values tend to be skewed right, because some
houses cost a lot more than most houses in
a community while very few houses cost a
lot less.
b. $43,964,124.05
c. $4,508.68
P27. a. The distribution of car ages is strongly skewed
right.
b. vehicles proving more durable; people
unwilling or unable to buy new cars
P28. a. 4 ft; 3.75 ft; 0.2 ft; 0.25 ft
b. 50 in.; 47 in.; 2.4 in.; 3 in.
c.
; 0.2 ft; 0.25 ft
P29. a. 2; 1
b. 12; 1
c. 20; 10
d. 110; 5
e. 900; 100
P30. a. Outliers occur above 30 + 1.5(21), or 1.5,
so Hawaii, at 12, is an outlier.
b. count 49; mean 41.5; median about 40;
SD about 16; min 80; max between 5 and
0; range between 75 and 80; Q1 about 51;
Q3 about 30
P31. 425 and 590; about 505; about 165
E47. a. It depends on the purpose of computing a
measure of center. Real estate agents usually
report the median because it is lower and tells
people that half the houses cost more and half
less. The tax collector wants the mean price
because the mean times the tax rate times the
number of houses gives the total amount of
taxes collected.
b. If the reason is to find the total crop in Iowa,
use the mean. An individual farmer might
use the median to see whether his or her yield
was typical.
c. Survival times usually are strongly skewed
right. Telling a patient only the mean survival
time would give too optimistic a picture. The
smaller median would inform the person that
half the people survive for a longer time and
half shorter. On the other hand, the mean
would help a physician estimate the total
number of hours he or she will spend caring
for patients with this disease.
E49. a. The scale goes from 36.67 to 58.89.

b.
c. yes
E51. Show that the mean of (x1 + c) + (x2 + c) +
(x3 + c) + (x4 + c) + (x5 + c) is equal to the
original mean, x plus c.
E53. a. .23 or .24
b. about 10 and about 70; 60
c. skewed right
d. no
E55. about 32 mi/h or, if rounded down, 30 mi/h
E57. A
Section 2.5
d. 78.81%
P32. a. 1.29%
b. 4.75%
c. 34.46%
P33. a. 0.47
b. 0.23
c. 1.13
d. 1.555
P34. a. 85.58%
b. 99.74%
P35. a. 1.645 to 1.645
b. 1.96 to 1.96
P36. a. cancer, because it is 1.35 SDs below the mean,
compared to 0.808 SD for heart disease
b. cancer, because it is 1.23 SDs above the mean,
compared to 1.097 SDs for heart disease
c. The death rate for heart disease in Colorado
(1.83 SDs below the mean) is more extreme
than the death rate for cancer in Hawaii
(1.29 SDs below the mean).
P37. a. about 24.2%
b. about 63.8 in.
P38. a. about 152 to 324
b. about 186 to 290
P39. A. outside both
B. not outside either
C. not outside either
D. not outside either
E59. a. 84.13%; 99.43%
b. 15.87%; 0.57%
c. 93.32%
d. 68.27% (68.26% using Table A)
E61. a. 2
b. 1
c. 1.5
d. 3
e. 1
f. 2.5
E63. a. i. 0.6340 (calculator: 0.6319)
ii. 0.0392 (calculator: 0.0395)
iii. 0.3085 (calculator: 0.3101)
b. about 287 to 723
E65. 76%; 87%
22-03-2009 21:30
Lesson
5 de 44
E67. a. 0.1587
b. 8.16 (or 8.17)
c. 10.02
d. 8 (or 7.89)
E69. 68%; 95%; 16%; 84%; 97.5%; 2.5%
E71. a. 0.2177 (or 0.2183)
b. 1,203,800 (or 1,207,650)
c. 73.56 in.
E73. a. skewed right
b. 0.1151
c. If all values in the distribution must be
positive and if two standard deviations below
the mean is less than 0, the distribution isnt
Chapter Summary
E75. b. min: 0; Q1: 2; median: 10; Q3: 35; max: 232
c. Florida and Texas
e. Both show the strong skewness and outliers.
In the stemplot, you can see that half the states
have fewer than ten tornadoes. The cleanness
of the boxplot makes it clear how much of an
outlier Texas actually is. However, you cant see
from the boxplot that many states have at most
one tornado. Because the stemplot is easy to
read while showing the values, it is reasonable
to select it as the more informative plot.
f. The distribution is strongly skewed right, with
two outliers and a wall at 0. The median
number of tornadoes is 10, with the middle 50%
of states having between 2 and 35 tornadoes.
E77. a. scores below 457.5 or above 797.5
b. probably skewed right
E79. a. Region 1 is Africa; Region 2 is the Middle
East; Region 3 is Europe.
b. Region 1 is A; Region 2 is B; Region 3 is C.
E81. No; for example, values {2, 2, 4, 6, 8, 8} are
symmetric with mean and median 5. Only two of
the six values, or about 33%, are within one SD,
2.76, of the mean, 5.
E83. a. Half the cities have fewer than 47 pedestrian
deaths per year, and half have more.
b. The three outliersChicago, Los Angeles,
and New Yorkare the three most populous
cities in the United States.
c. A stemplot reveals the three outliers and the
right skew and retains the original values.
d. If the data were presented in rates per 100,000
population, the three cities in part b might
not be outliers.
E85. 2.98; 1.33

E87. An example is {1, 1, 1, 1, 1, 2, 2, 10}.
E89. a. i. 6.325 vs. 6.667
ii. 2.000 vs. 2.010
iii. 0.632 vs. 0.633
b. No; as n gets larger, the difference between s
and approaches 0.
E91. a. The state with the lowest average income in
dollars in 1980 had an average income of
$7,007.
b. There is at least one outlier on the high end
for both years. There are no outliers on the
low end for either year.
c. no
E93. 126.65 mg/dL and 225.35 mg/dL.
E95. a. about .260; about .040.
b. Both distributions are approximately normal
in shape. The distribution for the American
League has a higher mean (by about .010)
and less spread.
c. about .300
Chapter 3
Section 3.1
P1. b.
c.
d.
P2. a.
b.
c.
e.
linear, positive, strong

yes; no
maturation during the early years of life
Delta; Northwest
upper left; United and America West
false
d. negative; weak
no, because these are the largest carriers in
the United States; no; yes
E1. a. positive, strong, linear
b. negative, strong, linear
c. positive, moderate, linear
d. negative, moderate, linear
e. positive, strong, linear
f. negative, strong, curved
g. negative, strong, curved; more variability
among values of y for smaller values of x than
for larger values of x
h. positive, strong, curved
E3. A. II
B. IV
C. III
D. I
843
22-03-2009 21:30
Lesson
6 de 44
E5. a. i. A, B, and C; none

ii. A, B, and C; D
iii. A and C
b. Positive: A and C; negative: B; D has little
trend.
c. A and C; D
d. no, because these are highly rated universities
e. Graduation rates may increase as SAT scores
increase because better-prepared students
are more successful in college. Alumni giving
rates may increase as graduation rates increase
because there are more happy alumni.
E7. a. Cases are the individual employees at the
time of the layoffs; variables are the age at hire
and the year of hire. There is a weak positive
association with heteroscedasticity. There
are no points in the upper left because these
employees have reached retirement age.
b. No; you need a plot of the age at hire of all
people hired, not just those who remained at
the time of layoffs.
c. People hired earliest were more likely to be
laid off. Perhaps they had obsolete job skills.
Section 3.2
P3. Each day the eraser tended to lose around 0.0135 g.
P4. a. about 0.8
b. If one student has a hand length that is 1 in.
longer than that of another student, the first
students hand tends to be 0.8 in. wider.
c. hand width = 1.7 + 0.8 hand length
d. The students in the lower cluster did not
spread their fingers. The slope and intercept
would be larger.
P5. a. student/faculty ratio; alumni giving rate
b. A horizontal run of 5 corresponds to a drop
of about 10 percentage points.
c. no, because no university has a student/
faculty ratio of 0 (i.e., no students)
d. about 23%; large
e. about 1 or 2; about 16
f. 11
g. positive
P6. b. y =279.75 + 2.75x
c. For every additional gram of fat, a pizza tends
to have 2.75 more calories. Five ounces of
pizza with no fat is predicted to have
279.75 calories.
d. 279.75 + 2.75 (11) = 310

e. 0.5 + ( 1.0 ) + 0.5 = 0
P7. % on time = 87.2 2.15 mishandled baggage;
4.08, 2.31, 0.71, 6.51, 1.56, 0.66, 0.11, 0.18,
2.99, 8.47
P8. yes; in row Error and column Sum of Squares
E9. a. IE; IIC; IIIA; IV D; VB
b. IA; IIE; III B; IVD; VC
E11. a. about 2.5 in./yr
b. Boys tend to grow a median of about 2.5 in.
per year from age 2 to 14.
c. height = 31.5 + 2.5 age
d. An average newborn is 31.5 in. long. No; this
is too long.
E13. a. The points lie on a line.
b. 0
c. For every 1 mi/h increase in speed, the
reaction distance is an additional 1.1 ft.
d. y =1.1x, where y is predicted reaction distance
in feet and x is speed in miles per hour
e. 60.5 ft; 82.5 ft
f. The equation would be y = 1.47x.
E15. a. arsenic concentration in the well water;
concentration of arsenic in the toenails of
people who use the well water
b. moderate, positive, linear; cluster in lower left
c. about 0.3 ppm
d. about 0.4 ppm
e. exceeded by seven wells
E17. a. y = 38 6x, where y is the predicted number
of days in Detroit with AQI greater than 100
and x is the number of years after 2000
b. The number of days with AQI greater than
100 in Detroit tended to decrease by 6 per
year on average.
c. 2002; 2
d. 6
e. 1 + 2 + ( 1 ) = 0
f. For y = 40 6x, SSE = 18. So the first line fit
better.
g. y = 37 6x; 25; 0, 0, 3; 9
h. Yes; for both of the other lines, all points are
on or to the same side of the line.
E19. a. height = 31.57 + 2.43 age
b. positive; 1.20 (or 1.21 with no rounding)
c. 51 = 31.57 + 2.43 8 (except for rounding
error)
d. almost the same
22-03-2009 21:30
Lesson
7 de 44
All costs have gone up over the years.
E21. a. height = 31.5989 + 2.47418 age

b. 2.4; yes P19. a. 0.747
E23. 1.33, 0.69, 1.99; possible curvature
b. The value of r2 means that 55.8% of the
state-to-state variability in the percentage of
E25. a. Yes; y = 195 + 10.05x, where y is the
families living in poverty can be explained
predicted number of calories and x is the
by the percentage of adults who are high
number of grams of fat; for every additional
school graduates. In other words, there
gram of fat, a pizza has about 10 more
is 55.8% less variability in the differences
calories.
between
y and y than between y and y . So,
b. Yes; y = 10.7 + 2.41x, where y is the predicted
by knowing the high school graduation
number of grams of fat and x is the cost; for
percentage for a state and using the regression
every additional dollar in cost, the number of
line, you tend to do a better job of predicting
grams of fat tends to increase by 2.41 g.
y than if you used just y as the prediction for
c. no
that state.
d. Calories has a moderately strong positive
c. no
association with fat, which makes sense
d. percentage of high school graduates;
because fat has a lot of calories. Fat has a weak
percentage of families living in poverty;
positive association with cost. There appears to
percentage of families living in poverty per
be no association between cost and calories.
percent of high school graduates; no units
P20.
yes;
yes
Section 3.3
P21. Yes; an ellipse around the cloud of points
P9. a. 0.5
b. 0.5
c. 0.95
will have a major axis that is steeper than the
regression line. Also, for the vertical strip
e. 0.95
d. 0
containing exam 1 scores above 95, the mean
P10. b. 0.908
exam 2 score is about 93. For exam 1 scores less
P11. a. 1
b. 0.5
c. 0.5
d. 1
than 70, the mean exam 2 score is about 76.
e. 1
f. 0.5
g. 0.5
h. 0.5
E27. a. 0.66
b. 0.25
c. 0.06
d. 0.40
P12. 0.908; all but one are positive, resulting in a
e. 0.85
f. 0.52
g. 0.90
h. 0.74
positive correlation.
E29. a. 0.707
b. 0.707
P13. a. positive
b. about (4.8, 5)
E31. a. about 0.95
c. Quadrants I and III; 20
b. All but three of the points lie in Quadrants I
d. Quadrants II and IV; 7
and III (based on the origin ( x, y )), where
each zx zy is positive.
P14. no for the top plot because of the curvature; yes
for the bottom plot
c. The point in the lower-left corner of
P15. a. 0.650
Quadrant III is the most extreme in both
x and y, so zx zywill be largest.
b. Exam 2 = 48.94 + 0.368 Exam 1; 78.38
d. For the point just below ( x, y ), zx and z y
c. Exam 1 = 14.1 + 1.149 Exam 2
both near 0, so zx zy will be quite small.
P16. a. citys population
b. Divide each number by the population of the
E33. a. no
b. yes
city to get the number of fast-food franchises
E35. a. yes
b. sx = 25
c. 0.081
d. 0.0183
per person and the proportion of the people
E37.
a.
overall
size
of
the
animal
who get stomach cancer.
b. Inflation: All costs have gone up over the years.
P17. Parents tend to give higher allowances to older
c. Time: Stock prices generally go up due to
children, and vocabulary is larger for older
inflation; also the Internet is new technology,
children than for younger children.
so the number of Internet sites is increasing.
P18. that people are too busy watching television to
have babies; how affluent the people are
845
22-03-2009 21:30
Lesson
8 de 44
E39. a. r = 0.9030
b. 0.0523
c. Yes; hot weather causes people to want to eat
something cold.
d. degrees Fahrenheit; pints per person; pints
per person per degree Fahrenheit; no units
e.
E41. Scoring exceptionally well involves some luck
in the questions asked, feeling well, having no
distractions, and so on. Its unlikely that this
combination will happen again on the next test
for the same student.
Section 3.4
P22. a. There is little pattern except for one outlier in
the upper right.
b. y = 680 + 2.85x; r = 0.694
c. y = 1350 2.14x; r = 0.501. Now the slope
is negative and the correlation is low. Titanic
has a huge influence.
P23. a. Not well; the estimates were low.
b. (180, 350)
c. With the point: actual = 12.23 + 1.92 estimate
and r = 0.975. Without the point: actual =
27.10 + 3.67 estimate and r = 0.921. This
point pulls the right end of the regression line
down, decreasing the slope and increasing the
correlation.
P24. b. Residuals are 0.5, 0.5, 0, and 0.
d. A residual plot eliminates the tilt in the
scatterplot so that the residuals can be seen as
deviations above and below a horizontal line.
Here the symmetry of the residuals shows up
better on the residual plot.
P25. a. AIV; BII; CI; DIII
b. i. opens upward, as in II
ii. fans out or in, as in I
iii. opens downward
iv. V-shaped, as in III
c. plot D; residual plot
E43. a. no; not elliptical and has an influential point
b. y = 161.90 + 0.954x; r = 0.49
c. With Antarctica removed, the slope of the
regression line changes from positive (0.954)
to negative (1.869) and the correlation
becomes negative, r = 0.45. Without a plot,

you cant see that there is little relationship.
E45. a. y = 1.63 + 0.745x; residuals: 0.567,
4.331, 0.096, 0.803, 1.529, 1.159,
7.178, 1.841, 1.433, 0.605
b. more variability in the middle than at either
end, partly because of more cases in the
middle
c. (8, 0); (10, 13)
E47. b. Pizza Huts Stuffed Crust; Pizza Huts Pan
Pizza; no (make a boxplot)
c. Removing Dominos Deep Dish increases the
slope from about 14.9 calories per gram of
fat to around 18 calories and decreases the
correlation from 0.908 to 0.893.
E49. AI; BIV; CIII; DII; C and D
E51. a. residuals: 66.67, 133.33, 66.67
b. only the horizontal scale
E53. To start, the predicted weight of a person whose
height is 64 in. is about 145 + 1.2, or 146.2 lb.
The slope of the regression line must be
about
, or 3.5.
Section 3.5
P26. c. ln y = 5.22 0.435x; 1 e0.435 0.35, or
35%, per time period
d. some curvature, indicating a death rate of
more than 0.35 in the early rolls and less than
0.35 in the later rolls
P27. a. log transformation
b. ln( pop) = 54.9342 + 0.03583 year;
3.6% per year
c. Florida grew less rapidly than the model
predicts until about 1845, then grew more
rapidly than predicted, then less, then more.
There was a big jump in growth between 1950
and 1960 and a big drop in 2000.
P28. (2, 3), (1, 2), (0, 1), ( 1, 0); slope 1 and
y-intercept 1
P29. a. (6, 3), (4, 2), (2, 1), (0, 0); slope 0.5 and
y-intercept 0
b. ( 5, 4 ), ( 6, 2 ), ( 8, 2 ); slope 2 and
y-intercept 14
P30. for P28: y =
; for P29 part a: y =
; for P29 part b: y =
22-03-2009 21:30
Lesson
9 de 44
P31. flight length = 4.807(1.012) speed

P32. Taking the natural log of the consumption gives
ln(consumption) = 0.39 + 0.143 trips, with
r 0.69. However, the transformed points dont
form an elliptical cloud. Some fishers families
eat essentially no fish, even if the person fishes as
many as 11 times a month.
P33. about 54.9 lb (or 54.8 lb with no rounding)
P34. a. There is a strong positive curved relationship,
with the rate of change in velocity decreasing
as the depth increases.
b. ln(velocity) = 0.146 + 0.175 ln(depth)
P35. a.
(cube root)
b. 1 (reciprocal)
c. 2 (square)
P36. a. 3 (cube)
b. 1 (reciprocal)
They are reciprocals.
c. (square root)
P37. a. (square root)
b. (cube root)
c. (square root)
P38. diameter squared versus age; diameter versus
square root of age
P39. a. less than 1
b. nearly linear
c. log(brain) = 0.908 + 0.76 log(body) or
brain = 8.10 body0.76; yes
E55. no
E57. A log-log transformation does not remove all the
curvature. If you split the ages into two groups,
between 8 and 9, and fit a linear equation to
each group separately, the residuals show less
curvature and are smaller.
E59. a. strong positive relationship with some
curvature
b. A line is not a bad fit here and would predict
reasonably well for parties of 16 or fewer
chimps because the residuals are small.
Because of the curvature, it probably would
not predict as well for parties much larger
than 16 chimps.
c. A log-log transformation works pretty well.
The model would be ln(percent) =
0.524 ln(chimps) + 2.9575 or percent =
19.25 chimps0.524.
d. The residual plot shows a random scatter,
which is good; however, there is more spread
for the smaller hunting parties than for the
larger ones, so a transformation that reduces
this would be better.
E61. There is essentially no relationship between the

day these passengers bought their tickets and
the price they paid, with the exception of five
passengers who bought their tickets within 9 days
of the flight and paid more than double what any
other passenger paid.
E63. a. negative and curved
b. log(GNP) = 1.87 0.0674 birthrate or
GNP = 74.13 0.856 birthrate; log(GNP)
decreases, on average, 0.0674 unit for every
1 unit increase in birthrate. A country with
no births is predicted to have a log(GNP) of
1.87 (which doesnt make sense).
E65. a. There is an increase in CO2 over the years,
with upward curvature.
b. CO2 increased at a lower rate than the overall
average from 1967 to about 1994 and at a
higher rate than the overall average during
the very beginning and the very end.
c. Use different lines, one line for 1959 to
about 1976 and the other for 1977 to 2002.
Or recognize that, while an exponential
model has an asymptote at 0, the CO2
level in the atmosphere was never near 0.
Preindustrial levels of CO2 were around
250 ppm. Adjust for this by taking the natural
log of (CO2 level 250).
d. The linear model for years after 1976 gives an
average increase of about 1.57 ppm CO2 per
year. Using the exponential model with an
asymptote at 250 ppm gives a growth rate of
about 1.5% per year.
Chapter Summary
E67. a. Arm span from height: y = 5.81 + 1.03x;
kneeling height from height: y = 2.19 + 0.73x;
hand length from height: y = 2.97 + 0.12x;
yes, the slopes are about what he predicted
(1, 0.75, and ), and the y-intercepts are close
to 0 in each case.
b. If one student is 1 cm taller than another,
the arm span tends to be 1.03 cm larger.
If one student is 1 cm taller than another,
kneeling height tends to be 0.73 cm larger. If
one student is 1 cm taller than another, hand
length tends to be 0.12 cm longer.
847
22-03-2009 21:30
Lesson
10 de 44
c. arm span and height: 0.992 (strongest),

kneeling height and height: 0.989, hand length
and height: 0.961 (weakest)
E69. a. Yes; (52, 83) lies away from the general
pattern on the scatterplot; the residual plot
shows that the student scored much higher
than expected on the second exam.
b. The slope increases from 0.430 to 0.540, and
the correlation increases from 0.756 to 0.814.
c. yes
d. Yes; the student who scored lowest on Exam
1 did much better on Exam 2, and the highest
scorer on Exam 1 was not the highest scorer
on Exam 2.
E71. a. Match each value with itself.
b. You can get a correlation of 0.950.
c. You can get a correlation of 0.1.
d. Match the biggest with the smallest, the next
biggest with the next smallest, and so on.
E73. a. true
b. true
c. false
d. true
E75. a. Public universities that have the highest
in-state tuition also tend to be the universities
with the highest out-of-state tuition, and
public universities that have the lowest
in-state tuition also tend to be the universities
with the lowest out-of-state tuition. This
relationship is quite strong.
b. no; yes c. no; yes
E77. C, B, A
E79. For example, stocks that do best in one quarter
may not be the ones that do best in the next
quarter.
E81. a. very strong linear relationship; one influential
point
b. very strong linear relationship; one influential
point
c. Yes; in general, the more police, the
higher the rate of violent crime; a log-log
transformation straightens these data.
E83. a. moderate positive correlation of 0.66; cause
and effect, because if teachers are paid more
the cost per pupil has to go up (unless class
sizes are increased proportionally)
b. moderate positive correlation of 0.577; about
the same; yes, because the number of pupils
in the state is pretty much proportional to the

number of people in the state
c. no
Chapter 4
Section 4.1
P1.
a. all the households in your community;

individual households
b. quicker, cheaper, and easier
c. getting a response from each household
P2. a. Al
b. Cal
c. Bal
d. Dal
e. Al
P3. a. size bias
b. The bias isnt obvious, but it is likely there was
some because of the judgment sampling.
c. voluntary response bias
d. size bias
e. voluntary response bias
P4. too high; convenience sampling
P5. question I
P6. incorrect response bias
E1. a. too high
b. Adults are more likely to be home.
c. sampling bias
E3. convenience sample; sample selection bias;
probably too high
E5. too high; too high
E7. a. voluntary response
b. No; people with stronger feelings are more
likely to respond.
c. quite a bit less than 92%
E9. Too high; the more children, the more likely a
family is to be in the sample.
E11. ABC News poll
Section 4.2
P7.
a. no
b. Although this produces a random selection
of students, it does not produce a simple
random sample of a fixed sample size.
c. No; two students sitting in different rows
cannot both be in the sample.
d. yes
22-03-2009 21:30
Lesson
11 de 44
e. No; a group of six girls cannot all be in the

sample.
f. No; two students with last names starting with
different letters cannot both be in the sample.
P8. a. 516; 384; 300
b. age
P9. 0.79
P10. a. Choose a random start between persons 1
and 20, and take every 20th person.
b. Choose a random start between persons 1
and 5, and take every 5th person.
P11. Choose a random start between 1 and 17, and
take every 17th person.
P12. a. Make an estimate of the number of mortgages
and divide by the sample size desired. Suppose
you get k. Then pick an integer at random from
1 through k, say you get j. The sample consists
of the mortgage at position j in chronological
order, and every kth mortgage after that.
b. Pick a random sample of dates. The sample
consists of all mortgages assumed on those
dates.
c. Start as in part b, then take a random sample
of mortgages from within each cluster.
P13. a. Use pages as clusters.
b. Take an SRS of pages. Then take an SRS of
lines from each of those pages.
c. Take an SRS of characters from each line
selected in part b.
E15. a. stratified random sample with strata of
grocery store owners and restaurant owners
E17. Stratification by gender is likely to be the best
strategy.
E19. Consider the farms as the five strata, and take a
random sample of, say, 10 acres from each farm.
E21. systematic sampling with random start in both
cases
Section 4.3
P14. a. children who live near a major power line
and the matching children; living near a
major power line or not; whether the child
gets leukemia
b. no, because the children are not randomly
assigned to the two conditions
c. For example, major power lines often are near a
major highway (and hence in a polluted area).
P15. a. Randomly select four test sites for each type

of glass. Measure the amount of energy lost
at each site, and compare the results from the
three types of glass.
b. the twelve site-generator combinations; the
type of glass
c. Randomly assign each treatment to two
generators at each site.
P16. a. The lurking variable is the persons age.
b. I causes II.
c. II causes I.
P17.
P18. a. Brightness of the room and type of music; for

brightness, the levels are low, medium, and
high; for type of music, the levels are pop,
classical, and jazz.
b. Possibilities are heart rate, blood pressure,
and self-description of anxiety level.
P19. a. No; the pipe and cigar smokers may be older.
b. observational study
c. smoking behavior; nonsmoking, cigarette
smoking, and pipe or cigar smoking; number
of deaths per 1000 men per year
P20. Older men have a higher death rate, and the pipe
and cigar smokers are older; the new factor is age.
P21. a. observational
b. legal age for driving; the age groups; highway
death rate by state
c. for example, driver education, because states
with higher age limits may generally be more
restrictive and require more training
P22. a. green
b. The greater distances traveled by the green
bears is due to confounding of launch
order with bear color. As students got more
practice, they were able to launch the bear
farther.
c. Later launches tend to have greater distances.
d. Randomize the order in which the bears of
different colors are launched.
P23. adults who died of nonrespiratory causes; yes;
no; no
849
22-03-2009 21:30
Lesson
12 de 44
P24. a. No; students might have used other clues.

b. group with the magnets
c. whether the students were assigned randomly
to the treatments
d. The second design is better than the first.
e. no; no
P25. the two textbooks; the ten classes; five for each
treatment
P26. Carnation plants (if in separate containers); put
the plants in place and then randomly assign the
new product to half and leave the others growing
under standard conditions. There must be at
least two plants in each group, and preferably
many more. The plants receiving the standard
treatment can be used as a control.
P27. a. brand of paper towel, with levels Brand A
and Brand B, and wetness, with levels dry
and wet; number of pennies a towel can hold
before breaking
b. Randomize both the assignment of wet or dry
to five towels from each brand and the order
of testing; the experimental units are the 20
time-towel combinations for which the tests
will be performed.
c. experiment
E23. observational study
E25. a. yes
b. All ten must be dug up so that the variables
shade and being dug up are not confounded.
E27. a. dormitories
b. 20
c. experiment
E29. different population sizes; climate is confounded
with proportion of older people in the state;
observational study
E31. not unless the subjects are in random order to
begin with
Section 4.4
P28. a. Some dorms are less healthy than others (too
stuffy or contaminated); some may be for
athletes or others who tend to be healthier.
b. to equalize variability between the treatment
groups
P29. a. randomized paired comparison (repeated
measures)
b. randomized paired comparison (matched
pairs)
c. randomized paired comparison with matched

pairs, because the drug might not clear out of
the bloodstream in the time allowed between
treatments
P30. randomized paired comparison with repeated
measures
P31. a. There is a lot of variability in how well
students memorize, so use a randomized
paired comparison design with either
matched pairs or repeated measures. The
response variable could be the number of
words on a list that are remembered.
b. Bigger people tend to eat a lot more soup
than do smaller people. Blocking could be
done by estimating a persons weight. The
response variable is the amount of soup eaten.
P32. a. No; no; you cant tell which dots represent
measurements for the same patient.
b. yes; yes
c. Subtract the number of flicks per minute for
Treatment C from that for Treatment A.
d. Randomized paired comparison with repeated
measures; theres a lot of subject-to-subject
variability, as shown in Display 4.20.
E33. completely randomized; high school students
who want to take the special course; course-taking
assignment; take course or dont take course; SAT
score; no blocks
E35. two weeks of a patients time; low phenylalanine
diet or regular diet; one week of a patients time;
paired comparison (repeated measures) with, we
hope, randomization
E37. block design; rate of finger tapping; one day of a
subjects time; caffeine, theobromine, or placebo;
three days of a subjects time
E39. completely at random; quantity eaten (relative to
body weight); hornworm; diet of regular food or
diet of 80% cellulose; no blocks
Chapter Summary
E41. a. sample, because the measurement process is
destructive
b. census, because the population is small and
the information is easy to get
c. sample, because a census would be too costly
and time-consuming
22-03-2009 21:30
Lesson
13 de 44
E43. a.
b.
c.
e.
probably reasonably representative

not representative, because too young
probably reasonably representative
not representative, because blood pressure
tends to increase with age
E45. estimate of earnings too high
E47. owners who sold their 2000 model, who might be
the owners unhappy with high repair bills
E49. Your estimate will be too large because your net
allows tiny fish to escape.
E51. New York Times readers tend to have higher
incomes and more years of education.
E53. Choose an SRS of states; choose an SRS of
congressional districts from each state; choose an
SRS of precincts from each congressional district;
choose an SRS of voters from each precinct.
E55. a. observational study
b. The researcher cant tell whether it was the
diet or some other difference in lifestyles that
accounts for the result.
c. more physical activity and less stress in Greece
E57. a. experiment; factors: presence of mother,
with levels present and absent, and presence
of siblings, with levels present and absent;
difference in the mouses weight; total
number of baby mice
b. survey; the amounts of the ten different
nutrients; the number of species of plants; 4
c. observational study; the different types of
fruit; float or not; one observed unit for each
type of fruit
E59. a. location; urban or rural; one pair of twins
b. observational study
c. within the rural twins; yes
d. genetically identical; variability caused by
genetic differences
E61. Form four blocks of spaces with similar locations
in the gym: 1 and 4, 2 and 3, 5 and 8, 6 and 7.
Randomly assign two bikes within each block.
Chapter 5
Section 5.1
P1. a.
P2.
P3.
P4.
P5.
P6.
P7.
P8.
P9.
E1.
b.
c. Probably not; if no one can tell the difference,
there is 1 chance in 16 that all four will select
correctly.
a.
, or about 0.48
b. For example, for a temperature of 20F, the
probability is , or about 0.074.
c. too warm
a. H1, H2, H3, H4, H5, H6, T1, T2, T3, T4,
T5, T6
b. yes
c.
a. 28, 35; 28, 41; 28, 47; 28, 55; 35, 41; 35, 47;
35, 55; 41, 47; 41, 55; 47, 55
b. yes
c.
d.
a. yes
b. no
c. no
d. no
a. yes
b. no
a. tails; tails; heads; tails; tails
b. 0.44
b. 6
c. probably not
a. 21
b.
a. There are 32 outcomes.
b.
c.
E3. a.
f.
E5. a.
b.
c.
d.
e.
E7. a.
d.
b.
g.
c.
h.
d.
i.
e.
disjoint and complete;

not disjoint; complete
disjoint; not complete
no; no; yes; no
b. 6 c.
, or 0.12
As the number of rolls increases, one
additional roll has a smaller effect on the
cumulative proportion.
E9. a. 20
c. 40
E11. a. yes
b. They get the white marble on the same
draw; .
c. Add more non-white marbles to the bags.
E13. a. 64;
b. 46,656;
1200
c. 2 ; no, because outcomes arent equally
likely
851
22-03-2009 21:30
Lesson
14 de 44
Section 5.2
P10. a. Assign a digit from 1 to 8 to each worker.
Start at a random place in Table D and look
at the next three digits. Those digits will
represent the workers selected to be laid off.
Ignore 9, 0, and any digit that repeats, and
select another digit.
b. Assign a pair (or pairs) of digits to each
worker.
P11. a. Each teaspoon has probability 0.80 of
disappearing. Whether each spoon disappears
is independent of whether other spoons
disappear.
b. Assign digits 18 to disappeared and 9 and 0 to
did not disappear. Look at ten digits, allowing
repeats, and record how many were 18.
c. with the assignments given in part b:
disappeared, did not disappear, disappeared,
did not disappear, disappeared, did not
disappear, disappeared, disappeared, did not
disappear, disappeared
d. The estimated probability that all ten spoons
disappear within 5 months is 0.1034.
P12. a. Each possible set of eight catastrophic
accidents is equally likely to be in the sample.
b. Assign a triple (or triples) of digits to each
accident.
d. The estimated probability that at least half
of a random sample of eight catastrophic
accidents are from cheerleading is about 0.72.
P13. a. The probability that a given team wins a
particular game is 50%, no matter what the
results of previous games are.
b. Assign digits 15 to team A winning and
digits 69 and 0 to team B winning. Select
digits, allowing repeats, until one of the teams
has four wins. Record the total number of
digits needed.
d. The estimated probability that a World Series
of two evenly matched teams will go seven
games is 0.31.
E15. a. The probability that a randomly selected high
school girl reports that she rarely or never
wears a seat belt is 10%. The girls are selected
randomly and independently.
b. Assign the digit 0 to reports she rarely or never
wears a seat belt and the other digits to does
not report that she rarely or never wears a

seat belt. Select four digits, allowing repeats.
Record the number of digits that are 0.
d. The estimated probability that a random
sample of four girls contains no more than
one who says she rarely or never wears a seat
belt is about 0.95.
E17. a. Each question has a 25% chance of being
answered correctly, regardless of whether
other questions are answered correctly.
b. Assign pairs of digits 0125 to a correct
response and pairs 2699 and 00 to an
incorrect response. Select ten pairs of digits to
represent the ten questions, allowing repeats.
Record the number of pairs that are 0125.
d. The estimated probability of getting more
than half the questions correct by guessing is
about 0.08.
E19. a. Each baby has probability 0.49 of being a girl.
The sex of each baby is independent of that of
the other babies in that family.
b. Assign pairs of digits 0149 to a girl and pairs
5099 and 00 to a boy. Select pairs of digits,
allowing repeats, until a pair in the range
0149 is selected. Record the number of pairs
needed.
d. The estimated average number of babies per
family for families who keep having babies
until they have a girl is about 2.1.
E21. a. Each backpack is equally likely to be picked
up by each friend.The backpacks are selected
independently of one another.
b. Let backpack 1 go with person 1, backpack 2
with person 2, and so on. Assign digits 1 and 2
to backpack 1, 3 and 4 to backpack 2, 5 and 6
to backpack 3, 7 and 8 to backpack 4, 9 and
0 to backpack 5. Select a digit to represent
the backpack taken by person 1. Select digits
for the other four people, skipping digits that
represent a backpack already picked. Record
whether each person gets his or her own
backpack.
d. The probability that each person gets his or
her own backpack is 0.008.
E23. a. You grab a key at random. Each key is equally
likely to be picked on any one grab.
b. Assign the digit 1 to your car key and the
digits 26 to not your car key. Select two
22-03-2009 21:30
Lesson
15 de 44
d.
E25. a.
b.
d.
digits, skipping 7, 8, 9, 0, and any repeats, and

record whether the second digit is 1.
The probability that if you grab one key at
a time you will get your key on the second
draw is .
You are assuming that the prize is equally
likely to be behind any door and that Monty
knows which door it is.
Assign digits 13 to door 1, digits 46 to door
2 , and digits 79 to door 3. Ignore 0. Select a
digit at random to be the door hiding the car.
Select a second digit at random to be the door
you choose. If you choose the door with the
prize, there are two doors Monty might open.
In this case, select a third digit at random
to determine which door Monty opens. Let
digits 15 mean that he opens the door to
the left and digits 69 and 0 mean that he
opens the door to the right. Record whether
switching to the other unopened door results
in the car or a goat.
The probability of winning the car with the
switching strategy is .
Section 5.3
P14. a.
d.
P15. b.
P16. a.
P17. a.
P18. b.
d.
P19. a.
b.
no; yes
b. 0.266
c. no
3,419,000
about 0.3825
no
b. yes
c. yes
d. no
yes
b. 0.60
c.
These events are not disjoint.
no
c.
P20. a.
c.
b.
P21.
P22.
E27. The events in a and c are disjoint.
E29. a. 0.511; 0.0085; 0.515
b. 0.489; 0.210; 0.692
c. 0.017; 0.667
E31. b.
, or about 0.88
E33. a. 0.55 + 0.25 + 0.15, or 95%

b. 0.80 + 0.70 0.55, or 95%
E35. 0.648
E37. Both occur if she gets HH.
E39. when P(A and B) = 0, P(B and C) = 0, and
P(A and C) = 0
Section 5.4
c.
P23. a.
b.
d.
e.
P24. a. 0.027
b. 0.028
c. A white hourly worker is a bit more likely to
be paid at or below minimum wage than is an
hourly worker in general.
d. 0.131
e. 0.117
f. A worker who is paid at or below minimum
wage is a bit less likely to be black than is an
hourly worker in general.
P25. a.
b.
c.
d.
P26. a.
b.
c.
P27. b.
c. P(S) P(F | S) = P(S and F)
d. P(M and S) = P(M) P(S | M);
P(M and S) = P(S) P(M | S)
P28.
P29.
P30.
P31. P(doubles | sum of 8) =
P(sum of 8 | doubles) =
P32.
P33. 56.7%
P34. b. About 69% of samples that the technician
decides are contaminated actually are
contaminated.
c. About 97% of samples that the technician
decides arent contaminated actually arent
contaminated.
d. PPV = 0.90; NPV = 0.90
P35. PPV = 0.75; NPV = 0.20; sensitivity = 0.60;
specificity 0.33; no
853
22-03-2009 21:30
Lesson
16 de 44
P36. a.
b. 1
c. The probability depends on the probability
model used for selecting the coin.
d. same as part c
e. 1
f. 0
E41. a. 0.2883
b. 0.2420
c. 0.1367
d. 0.3548
e. ages 3544
E43. b. 0.154
c. 0
E45. a.
d.
E47. a. 0.0059
b. 0.0045
c. 0.0059
d. 0.0060
E49. a. 24%
E51. b. 0.088
c. 0.661
E53. a. no
b. This is a new test, so it is unlikely that false
negatives have shown up yet.
c. 177
E55. a.
Section 5.5
P37.
P38.
P39.
P40.
P41.
E57.
E59.
so not independent; no
a and c
b. 0.4872
a. 0.9957
b. 0.9998
a. 0.0121
b. 0.1958
only b
a.
b. no
c. No; six people are both.
E61. a. BB, BG, GB, and GG
b. two boys; two girls
c. 0.2601; births are independent.
E63. a. 0.021
b. 0.449
E65.
10 12 b. 0.753
E67. a. 0.1112 3.14
E69. It is not reasonable to assume that the results of
two surgeries on the same person at the same
time are independent, so the probability is
unknown.
E71. 21.4%
E75. b. P(win) = P(win | day) P(day) +
P(win | not day) P(not day) =
d.
P(win | day) =
Chapter Summary
E77. a.
e. no; no
f. no; yes
E79. a. Randomly select seven digits from 1 to 9
(with replacement). See if the number 3 is
among the seven digits selected. If so, the run
was a success. Repeat many times.
b. 0.562
E81. a. 0.675
b. 0.025
c. 0.975
d. 0.75
b. 0.249999975
E83. a. , or about 0.167
c. no; no; part b, because of the large population
size compared to the sample size
E85. a. 0.381; 0.310; 0.286; 0.405; 0.923; 0.75
b. no; P(Republican) P(Republican | voted
Republican)
c. no
E87. b. no;
E89. This is a reasonably good test, but if depression is
a serious impediment to recovery, a test should
have higher sensitivity than 86%. The number of
false positives, 8, isnt much of a problem because
these people will receive further testing.
E91. a. about 3000
b. about 0.3
Chapter 6
Section 6.1
P1.
P2.
P3.
P4.
2, 4
Let any digit represent smoking is not responsible.
The other nine digits represent smoking is
responsible.
For example, begin by assigning the triples
001524 to represent families with zero children.
Then the first triple, 488, represents a family with
no children.
0.007744, if its reasonable to assume
independence; very close
22-03-2009 21:30
Lesson
17 de 44
P6.
P(X = 0) = 0.027; P(X = 1) = 0.189;

P(X = 2) = 0.441; P(X = 3) = 0.343
P7. b.
c. P(X = 0) = ;P(X = 1) = ; P(X = 2) = ;
P(X = 3) = 0
P8. 0.873; 1.096
P9. sum: triangular and symmetric, 7, 2.415; larger
number: skewed left, 4.472, 1.404
P10. a. P(X = 0) = P(X
; = 1) = ;P(X = 2) =
b. 1; 0.632
c.
P11. a. $1.20
b. $3.60
P12. $116
P13. a. $33
b. $19.52 (after including the probability 0.2 of
no rental); yes
P14. a. triangular
b. 0; 5.833
c. 3.5 3.5 = 0; 2.917 + 2.917 5.834
(The difference is due to rounding.)
P15. $1060.80; $106.58, assuming savings from week
to week are independent
P16. a. 349
b. It would not be unusual to be off by 14.057
vehicles or so.
c. That the numbers of vehicles in the
households are independent of each other;
however, people living in the same apartment
complex probably will have similar income
levels and so are more likely to have closer to
the same number of cars.
P17. 3.49; 1.41; quite close
P18. a. 2.236; 1.115
b. 4.472; 1.577
E1. a. 2.16; 2.221
b. Assign random digits by pairs.
E3. 126.5; 29
E5. P(X = 0) = 0.729; P(X = 1) = 0.243;
P(X = 2) = 0.027; P(X = 3) = 0.001
E7. P(X = 0) = 0.6724; P(X = 1) = 0.2952;
P(X 2) = 0.0324
E9. a. $209.50; $4790.50
b. $214.50
c. for example, whether there is a burglar alarm
E11. a. for example, P(X = 0) = 0.0077,
P(X = 5) = 0.1440
b. quite close
c. 3.49 is equal to the mean found in P17

and is close to the mean of the simulated
distribution, 3.53.
E13. a. $0.185
b. $315,000
c. 37%
E15. a. Brand A: $1115; Brand B: $1200
b. The advantage of buying Brand A is that
you expect to pay less and there is a 70%
chance you will pay less with Brand A than
you will with Brand B, even with one repair.
The advantage of buying Brand B is that the
maximum you might spend is less.
E17. a.
b.
c. 1.25
E19. a. P(X = 2) = 0.02; P(X = 3) = 0.14;
P(X = 4) = 0.36; P(X = 5) = 0.48
b. 4.3 months; 0.781 month
c. 2.5 + 1.8 = 4.3
E21. a. for example, P(X = 2) = ; P(X = 3) =
b.
= 3.5 2.5 = 1 =
2.917 + 1.25 = 4.167 =
Section 6.2
P19. for example, P(X = 3) = 0.3125 and
P(X = 6) = 0.015625
P20. 0.2188; 0.1094; 0.0352
P21. 0.1641; 0.5
P22. a. 0.6309
b. 0.3691
c. for example, P(X = 3) 0.005668 and
P(X = 5) = 0.000005
P23. For example, assign the triples 001088 to
represent the dropouts and the other triples to
represent non-dropouts.
P24. P(X = 0) = 0.0053; P(X = 1) = 0.0575;
P(X = 2) = 0.2331; P(X = 3) = 0.4201;
P(X = 4) = 0.2840
P25. a. 0.1875
b. 8; 2; 0.227
c. that the sample was not selected randomly
P26. Five-question quiz; the probability of getting 3
or more correct on a 5-question quiz is 0.5. On a
20-question quiz, the probability of getting 12 or
more correct is only about 0.25.
E23. a. 0.4019
b. 0.0322
c. 0.5981
d. 0.8038
E25. a. 0.8446
b. 3.175; 1.6649
E27. a. $6.00
b. 0.0571
c. $9.20
855
22-03-2009 21:30
Lesson
18 de 44
E29. a. 0.973
b. 0.99927; no
c. 4
c.
E31. a. 7
b. 17
d.
E33. If 100 different polls each randomly selected
three adults, you would expect to see a total of
E49. a.
81 college graduates, or an average of 0.81 college
graduate per poll.
b.
Section 6.3
P27.
P28.
P29.
P30.
a.
a. 0.0191
b. 0.0225
a. 0.0074
b. 0.9993
a. 0.3
b. 0.21
c. 0.1029
d. Whether you get through is independent of
the outcome on previous calls, and each call
still has probability 0.3 of being completed.
P31. a. 10
b. 9.4868
c. 20; 30
P32. a. 1.1765
b. 0.4556
c. 11.765
d. 1.441
E35. a.
d. 0.5
E37. a. 1.1364; 3.4091
b. 0.3936
c. $113.64; $39.36
d. no; only 1.44% of the time
E39. 0.5528; this assumes the two test results are
independent, which is almost certainly not the
case.
E41. a. The probabilities for a geometric distribution
follow this same pattern: p, qp, q2p, q3p, q4p,
q5p, . . . .
b.
Chapter Summary
E43. a.
b. 6.5; 3.452
c. 13; 4.882
E45. a. P(X = 1) = 0.19; P(X = 2) = 0.45;
P(X = 3 ) = 0.36
b. P(X = 1) = 0.01; P(X = 2) = 0.15;
P(X = 3) = 0.84
c. 2.17 months; 2.83 months; System II
E47. a. 0.6513; the probability that a donation is type
B is independent of the type of the donations
previously checked.
b. 0.5487
c.
E51. a.
E53. a.
b.
c.
d.
e.
E55. a.
c.
d.
No; the probability is only 0.0399. Check

more donations.
0.729
0.9974; that in your region 80% of the girls
want to do their best in all classes
0.8782; that in your region 65% of the boys
want to do their best in all classes
0.9979; that the sample is about half girls
$31.804
b. about 128
300, 450, 900, 325, 475, 925, 700, 850, and
1300; 313.8028; 98,472.22
33,472.22; 65,000;
33,472.22 + 65,000 = 98,472.22
no; 182.9542 + 254.9510 313.8028
yes; 341.67 + 350 = 691.67
no; 225 + 250 700
2
b. 0.9990
1.001 flips; less, because there is a smaller
chance of all ten living than there is of a
single flipper living
about 4.8 rounds
Chapter 7
Section 7.1
P1. a. 50; 7300
b. Randomly select three planets. Sum the
numbers of moons. Repeat many times.
P2. a. IB; IIA; IIIC
b. no, because a mean of 86 or greater did not
occur in 100 random samples in Histogram C
P3. a. D; B; C; A
b. D; A
P4. a. $8 million; $1.9 million
b. There are 15 possible amounts, each with
probability
c.
P5. a. = 261.8; = 171.85
c. Both equal 261.8.
d. The SE, 105.23, is much less than = 171.85.
P6. a. 471
c. 360.3
d. No; the mean of the sampling distribution
of the range is much less than the actual
range, 471.
e. yes
22-03-2009 21:30
Lesson
19 de 44
E1. a. A20; B1; C2; D50

b. The means are all about 4.5, regardless of
sample size.
c. The larger the sample size, the smaller the
spread.
E3. b. compared to mean: more skewed left ,
centered higher, about equal spread
c. good estimator of the midrange of the
population, but tends to be too large
compared to the mean and median
E5. b. 62, 63; 62, 64; 62, 64; 62, 65; 63, 64; 63, 64;
63, 65; 64, 64; 64, 65; 64, 65
c. Sampling distribution: mean 63.6, SE 0.6245;
population: mean 63.6, SD 1.02; means are
equal, but SE < SD.
d. 63, 64, 64, 65, 64, 64, 65, 64, 65, 65
e. 1, 2, 2, 3, 1, 1, 2, 0, 1, and 1
E7. a. Unbiased; the mean of the sampling
distribution of the sample mean is equal to
b. biased; tends to be too small
c. biased; tends to be too small
E9. a. The sample of 48 packages will contain 0, 24,
or 48 lb of spoiled fish, but the probabilities
are unknown.
b. No, because the sample size was only 2;
sample one fish (or one package) from each
of a larger number of cartons.
E11. e. 4;
f.
g. by n 1, because the average of the sample
variances, , is equal to the population
variance
h. too small
E13. a.
N b. at N
c. N
(sample maximum)
Section 7.2
P7. a.
Ipopulation; II n = 10; III n = 4; mean

1.7 for each; SD for I, II, and III about 1, 0.3,
and 0.5, respectively
b. 1.7; yes
c. 0.5; 0.316; yes
d. The population is slightly skewed, and the
two sampling distributions get more nearly
normal as the sample size increases; yes.
P8. approximately normal; .266; .009553

P9. a. approximately normal; mean 0.9; SE 0.035
b. from 0.83 to 0.96
c. 0.002
P10. a. 0.3058 (0.3050 using Table A)
b. 0.1549 (0.1539 using Table A)
c. 0.15
d. 2.4 and 15.4
P11. a. 0.04
b. 0.0384
c. 0.64
d. 0.6384
e. 0.36
f. 0.3616
P12. would not change, because the sample size
in both cases is a very small fraction of the
population size
P13. a. 0.1226 (0.1230 using Table A)
b. 0.1635 (0.1645 using Table A)
P14. a. No; the probability is only 0.002.
b. Yes; the probability is 0.982.
P15. a. (0.469, 1.331)
b. (0.684, 1.116)
c. (0.832, 0.968)
d. (0.866, 0.934)
E15. a. I: B, 25; II: A, 4; III: C, 2
b. yes
c. The distributions for n = 4 and n = 2
reflect the skewness of the population.The
distribution for n = 25 looks approximately
normal.
d. works well for n = 25 and less well for n = 4
and n = 2
E19. a. No; two or more accidents happened in about
27% of the days.
b. The top plot is for 8 days.
c. No; an average of 1.75 or more accidents
occurred about 16 times out of 200.
d. Yes; an average of 1.75 or more accidents
occurred about 4 times out of 200.
e. It isnt reasonable to assume that the days are
independent; for example, if the first of the
days has ice or snow, its likely the other days
will also.
E21. a. 0.4602
b. 0.4207
c. 0.3085
d. Assign the pairs of digits 01 through 46 to
be a score of 510 or greater. The other pairs
represent a score lower than that. Take four
pairs of random digits and see if all four
represent scores of 510 or greater.
E23. a. 0.4729 (0.4714 using Table A)
b. about 384
857
22-03-2009 21:30
Lesson
20 de 44
E25. a. 134.8
b. almost 0
c. No; this group is special in that the employees
are all old enough to have a job.
c. The small number cant reasonably be
attributed to chance.
E41. 0.1191 (0.1188 using Table A); you are assuming
that the 75 married women were selected
E27. a. 27
b. $4025
randomly and independently from a large
E29. a. remains at the population mean, , for all
population in which 60% are employed.
sample sizes; shrinks by a factor of
E43. a. yes; no
b. increases by a factor of n; stretches by a factor
b. For a fixed p, the SE decreases as n increases.
of
For a fixed n, the closer p is to 0.5, the larger
E31. a. 1021; 161.23
the SE.
b. 0.0852 (0.0853 using Table A)
c. For a fixed p, the skewness decreases as n
c. (705, 1337)
increases. For a fixed n, the farther p is from
d. 0.2378 (0.2389 using Table A)
0.5, the more skewness.
e. mean 1021; SE unknown, because the two
d. for n = 25 or 100 when p = 0.4, for n = 100
scores arent independent
when p = 0.2
E33. a. 10.5; 8.75
b. 24.5; 20.417
E45. a. = 0.65, which is equal to the proportion of
successes,
c. 10,060; 255,380; 0.4527 (0.4522 using Table A)
b.
0.6, which is the same
Section 7.3
c. The sample proportion is a type of sample
mean, if successes are assigned the value 1
P16. 0.853
and failures the value 0.
P17. a. n = 100; n = 10; yes
b. 0.10 in each case
c. n = 10; n = 100
d. 10
Chapter Summary
P18. a. means 0.10; SEs 0.095, 0.067, 0.047, 0.03
E47. a. Assign each city a two-digit number from
c. 0.0999999; 0.095; yes
01 to 32. Generate pairs of random digits
until you get three distinct pairs between
P19. a. approximately normal; 0.53; 0.05
01 and 32. Ignore 3399 and 00 and repeats.
b. no, because the probability of getting nine or
fewer women just by chance is close to 0
b. 140.9; 68.94
c. no
P20. a. close to 0
b. yes
E49. a. approximately normal; 0.68; 0.033
P21. a. (0.448, 0.752)
b. (0.504, 0.696)
b. 0.17
c. (0.552, 0.648)
c. less than 0.615 or greater than 0.745
E35. a. approximately normal, mean 0.92, SE 0.00858
d. 0.8185 (0.8186 using Table A)
b. approximately normal, mean 920, SE 8.58
E51. a. regular increase from 40 to 60 followed by a
sharp drop after 60 and another sharp drop
c. 0.0099
after 65
d. 0.2800 (0.2810 using Table A)
b. 0.6048 (0.6065 using Table A)
e. 937 or more or 903 or fewer; larger than
E53. a. approximately normal, mean 4, standard
0.937 or smaller than 0.903
error 3.680
E37. a. slightly skewed left; mean still 0.92; SE larger
b.
0.7066 (0.7054 using Table A)
b. slightly skewed left; mean one-tenth as large;
c.
0.1385
(0.1379 using Table A)
SE smaller
E55.
a.
Select
two
integers at random from 0 through
c. larger
9 and compute their mean. Repeat. Subtract
E39. a. 0.5
the second mean from the first. You win that
b. 0.000011 (close to 0 using Table A)
amount.
22-03-2009 21:30
Lesson
21 de 44
>
b. shape slightly rounded-out triangular, mean

0, SE 2.03
E57. a. becomes more approximately normal; stays at
p; decreases
b. becomes more approximately normal;
increases; increases
E59. 0.739
E61. a. 12.8 and 4.56; 8 and 4
b. total = 20.8 and
total = 6.96
c. 12.8 + 8 equals 20.8, as you would expect;
but 4.56 + 4 is not equal to 6.96, and you
should not expect it to be because the morning
and afternoon times were not selected
independently when you computed 6.96.
Chapter 8
Lesson 8.1
P1. a. 0.264 and 0.536
b. yes
P2. Let random digits 1, 2, and 3 represent successes.
Count the number of successes in 40 digits.
P3. about 2242
P4. about 1426
P5. no
P6. about 2839; about 0.700.975
P7. about 15% to 40%
P8. about 30% to 60%
P9. a. about 50% to 75%
b. (0.475, 0.775)
P10. a. Conditions are met.
b. (0.612, 0.688); 0.038
c. (0.618, 0.682); 0.032
d. 95%, because a longer confidence interval has
a greater chance of including p
b. (0.024, 0.056)
c. narrower, because is farther from 0.5
P12. 72
P13. 1
P14. The horizontal line segments will be shorter; the
confidence intervals will be narrower.
P15. 1.28; smaller, because z* is smaller
P16. 9 times as large
P17. a. 2,401; 16,590; 27,061
P18. 879
P19. a. The second two conditions are met, but

random sampling isnt mentioned.
b. (73.3%, 78.7%)
c. the proportion of all students ages 1217 with
Internet access who would say this
d. If you could ask all students ages 1217 with
Internet access whether they go online to get
news or information about current events,
you are 95% confident that the proportion
who would say yes would be somewhere in
the interval from 73.3% to 78.7%.
e. Suppose you could take 100 random samples
from this population and construct the 100
resulting confidence intervals. Youd expect
the proportion of all students ages 1217 with
Internet access who would say this to be in 95
of these intervals.
E1. a. no
b. about 0.65; about 0.35
c. about 30% to 55%
E3. a. about 0.050.25
b. (0.039, 0.261)
c. They are similar. The chart would give a more
exact interval but doesnt include all values
of p. Also, n (1 ) < 10, so the normal
approximation used in part b is not reliable.
E5. a. no apparent randomization
b. no, because there is no indication of a
random sample
c. yes
d. the proportion of all U.S. teens who know this
e. Suppose you could take 100 random samples
from this population and construct the 100
resulting confidence intervals. Youd expect
the proportion of all U.S. teens who know this
to be in 95 of these intervals.
E7. Conditions have been met. (0.485, 0.555)
E9. No; Display 8.2 applies only to samples of size 40.
E11. Conditions are met. You are 95% confident
that, if you asked all adults, the percentage who
thought this would be between 78.6% and 83.4%.
E13. a. gets narrower
b. gets wider
E15. a. yes, because 0.042 rounds up to 5%
b. 1068
E17. A, D, F, H
E19. a. about 0.013 using 95% confidence
b. at least 212,345
859
22-03-2009 21:30
Lesson
22 de 44
E21. a. You want to maximize p(1 p); because p is

restricted to [0, 1].
b. quadratic, or parabola opening down
c. x =
P29. a. z = 5.01 is more extreme than z* = 1.96;

reject H0, that spinning a quarter is fair.
b. z = 1.34 is less extreme than z* = 1.96;
dont reject H0, that flipping a quarter is fair.
P30. a. about 1.55
b. 0.0836
Section 8.2
P31. a. 1%
b. 2.33
P32.
Conditions
are
met. H0: The probability of heads
P20. a. p = where p is the proportion of todays
when a quarter is spun is 0.5. Ha: The probability
teens who want to study more about medical
is not 0.5. Because z = 5.01 is more extreme
research
than z* = 1.96, reject H0, that spinning a
b. 0.575
quarter results in heads half the time.
c. not statistically significant
P33. Conditions are met. H0: The proportion of all
P21. a. p = 0.5, where p is the probability that the
U.S. bookstores that sell DVDs is 0.5. Ha: The
student gets an answer correct.
proportion is not 0.5. Because z = 1.34 is less
extreme than z*= 1.96, the result from this
b. 0.75
sample is reasonably likely to occur if half of all
c. statistically significant
bookstores sell DVDs. Do not reject H0.
d. No; the student might have been really lucky.
P34. a. p = 0.2, where p is the probability that the
P22. a. p = 0.75, where p is the proportion of all
student gets any one answer correct
juniors who will want extra tickets
b. p = 0.5, where p is the proportion of
b. 0.80
newscasters who are men (or women)
c. not statistically significant
c. p = where p is the proportion of weekly
d. no
car-washing people who wash on Saturday
P23. 1.23
P35. B
P24. 0.149
P36. a. With z = 1.43, Hila should not reject H0.
b. no
P25. If spinning a penny is fair, there is only a 0.0016
chance of getting 10 or fewer heads or 30 or more P37. a. Yes, they each have a 0.05 chance.
heads in 40 spins.
b. no, because H0 isnt false
P26. If the cats could not distinguish between the rakes,
P38.
a. no, because H0 isnt true
the probability of getting the rake with the food
b.
Both could, but Jeffrey is more likely to
28 or more times (or 22 or fewer times) is 0.3961.
because he has the smaller sample size.
P27. a. H0: p = 0.69, where p is the proportion of
P39. = 0.10
houses in your community that are occupied
by their owners; Ha: p 0.69
P40. n = 200
b. 1.376
P41. 10%, which is farther from p0
c. If the proportion of houses in your community
P42. a. Random sample; np0 = 500 and
that are occupied by their owners is 69%, then
n(1 p0) = 500 are both at least 10; the
the probability of getting 30 or fewer or 39 or
number of U.S. adults is at least 10,000.
more owner-occupied houses in a random
b. H0: The proportion p of all U.S. adults who
sample of 50 houses is about 0.1688.
would say they are satisfied is 0.5. Ha: p < 0.5.
d. With a P-value this high, there isnt sufficient
c. 2.53; 0.0057
evidence to reject H0, that the proportion
d. The P-value, 0.0057, is less than = 0.05,
of houses occupied by their owners in your
so reject H0. There is sufficient evidence to
community is 0.69.
support the claim that the proportion of all
P28. B
U.S. adults who would say they are satisfied is
less than 50%.
22-03-2009 21:30
Lesson
23 de 44
p is the proportion of all 0.
P43. a. Random sample; np0 = 500 and

n(1 p0) = 500 are both at least 10, and
the number of U.S. adults is at least 10,000.
b. H0: The proportion, p, of all U.S. adults
who would say they are dissatisfied is 0.5.
Ha: p > 0.5.
c. 0.632; 0.2635
d. The P-value, 0.2635, is more than = 0.05,
so do not reject H0. If the proportion of U.S.
adults who would say that they are dissatisfied
is 0.5, you are reasonably likely to get a
random sample of 1000 adults in which
51% say they are dissatisfied.
P44. Conditions are met if the number of teens in your
community is at least 1690. H0: 51% (or even
less) of teens in your community know this.
Ha: The percentage is greater than 51%. Because
the P-value, about 0.1473, is greater than 0.05,
dont reject H0. If it is true that the percentage of
teens in your community who know this is 51%,
then you are reasonably likely to get 55% who
know this in a random sample of 169 teens.
P45. a. false b. false c. false d. true
E25. D
E27. a. H0: p = 0.6, where
students in the school who carry backpacks
to class
b. z = 0.6455; if it is true that 60% of the
students carry a backpack to class, the
probability of getting 26 or more or 22 or fewer
in a random sample of size 40 is about 0.519.
c. Because the P-value is more than 0.05, do
not reject H0. The results from the sample are
consistent with an overall percentage of 60%
of students carrying a backpack to class.
E29. Conditions are met. H0: p = 0.25, where p is
the proportion of all U.S. adults who would pick
a cure for cancer. Ha : p 0.25. Because the
P-value, about 0.0002, is less than 0.05, reject H
In a random sample from a population with 25%
successes, there is almost no chance of getting
as many successes as Gallup got in this sample.
Thus, it is not plausible that 25% of all adults
would pick a cure for cancer.
E31. Conditions are met. H0 : p = , where p is the
proportion of all U. S. adults who would have
said this. Ha : p
.The P-value, 0.08, is larger
than 0.05, so dont reject H0. You have insufficient

evidence to conclude that if all U.S. adults had
been asked, the proportion who would have said
this would be different from .
E33. 5
E35. a. yes for both; Taline
b. no, because H0 isnt false
E37. a. For example, the probability of exactly 4 heads
is 0.00462.
b. 5 or fewer, or 15 or more
c. 0.1272 d. Increase the sample size.
E39 a. no
b. Yes, 0.50 is a plausible value of p.
c. (0.365, 0.657), which contains 0.5
d. A smaller margin of error means greater
power.
e. Use a larger sample.
E41. a. one-sided b. a random sample
c. almost 0
d. One of three conclusions could be drawn: This
area always could have had a larger proportion
of birds with the mutation, there might have
been something about the mutation that made
those birds more likely to be in the sample,
or the proportion of birds with the mutation
is higher since the radiation leak. If scientists
can rule out the first two conclusions, they can
go with the third.
E43. a. one-sided
b. = 0.668; p0 = 0.5
c. yes, except for rounding error
d. yes
e. Not really; a much smaller sample would
have been sufficient because 66.8% is so far
from 50%.
E45. a. The P-value is 0.65, so you cannot reject H0.
b. The P-value is 0.18, so you cannot reject H0.
c. This is why we dont accept H0. There isnt
enough evidence to reject either H0, yet at
least one must be false.
d. The P-value is 0.155, so you cannot reject H0.
e. The P-value is 0.00002, so you can reject H0.
f. Larger sample sizes make it easier to reject a
false H0.
861
22-03-2009 21:30
Lesson
24 de 44
Section 8.3
b. You are 95% confident that the difference in
the two rates, 1994 this year, is between
0.206 and 0.066.
c. Yes. There is insuffcient evidence to support a
claim that there was a change in the percentage.
P47. a. Conditions are met, because Gallup uses a
randomization that can be approximated by
random sampling.
b. 0.14 to 0.30
c. You are 95% confident that the difference
between the proportion of all 13- to
15-year-olds who respond yes and the
proportion of all 16- to 17-year-olds who
respond yes is between 0.14 and 0.30.
d. No. It is not plausible that there is no
difference between these proportions.
E47. B and C
E49. a. Conditions are met except whether this is a
random sample.
b. (0.009, 0.571)
c. You are 95% confident that if the judges
had been given a choice of two dogs for
each San Diego dog owner, the difference
in the proportion of correct guesses for all
purebreed owners and the proportion of
correct guesses for all mutt owners would be
d. No. This implies that the difference in the
proportions probably was not due to chance,
but it is a close call.
e. Take bigger samples to be sure 0 isnt in the
E51. a. Conditions are met.
b. (0.19, 0.03)
c. You are 99% confident that the difference in
the percentage of all men and the percentage
of all women who prefer to be addressed by
their last name is between 0.19 and 0.03.
d. No. You are convinced that there is a
difference between the percentage of women
and the percentage of men who prefer to be
addressed by their last name.
E53. Conditions are met. Because 0 isnt included in the
confidence interval (0.094, 0.126), it is acceptable
to use the term significantly more likely.
E55. Generally, the length of the interval gets smaller.

E57. The sampling distribution of p1 p2 is
E59. No, because the respondents werent selected
independently from two different populations;
use a confidence interval for a single proportion,
which is (0.140, 0.304).
Section 8.4
P48. a. 0
b. 0.0154
d. 0.0006
P49. a. false
b. true
c. true
d. true
e. true
P50. Conditions are met if the town has at least 7500
probable voters. H0: The proportion, p1, of all
potential voters who favored the candidate at
the time of the first survey is equal to the
proportion, p2, of all potential voters who favored
the candidate one week before the election.
Ha: p1 > p2. Do not reject H0. If there is no
difference between the proportion of potential
voters who favored the candidate at three weeks
and the proportion at one week, then there is a
0.1741 chance of getting a difference of 0.0257 or
larger with samples of these sizes.
P51. Type I
E61. a. 0.04
b. 0.0585
d. 0.432 (0.4325 using Table A)
E63. a. Its reasonable to assume that a larger tumor is
more likely to spread than is a smaller tumor.
b. no, because these are not independent
random samples of patients with tumors of
different sizes
c. yes
d. If large tumors and small tumors are equally
likely to metastasize, then there is about a
0.02 probability of seeing a difference in
proportions at least as large as that seen in
this study.
E65. a. Conditions are met. H0: The proportion,
p1, of all girls who would have said yes is
equal to the proportion, p2, of all boys who
would have said yes. Ha: p1 > p2. Reject H0.
If the proportions of all girls and all boys ages
1217 who would have said yes were equal,
the probability of getting a difference in
proportions of 13% or greater from samples
of these sizes would be only 0.0007.
22-03-2009 21:30
Lesson
25 de 44
b. Not all boys and girls are equally likely to be

plausible that there is no difference in the
in the sample.
proportions who would get arrested.
c. That about half of teens ages 1217 are girls
d. Although it seems like a big stretch to say that
and half boys; you are 95% confident that if
not going to preschool caused more children
you asked all teens this question, between
to get arrested, we have no other explanation
other than an unlikely Type I error.
65.5% and 73.4% would say yes.
E67. Conditions are met. H0: The proportion, p1,
P54. a. Both conditions are met. H0: If all patients
of all adult Americans who would give NASA
could have been given Lipitor, the proportion,
p1, who had heart attacks would be equal to
a favorable rating in 2005 is equal to the
proportion, p2, who would give NASA a favorable
the proportion, p2, who had heart attacks
rating in 1999. Ha: p1 > p2. With a P-value of
had they all been given Zocor. Ha: p1 p2.
0.0008, reject H0. There is strong evidence that
Because the P-value, 0.082, is greater than
NASA was being looked upon more favorably in
= 0.05, dont reject H0. There is not sufficient
2005 than it was in 1999.
evidence to conclude that, had all patients
been treated with Lipitor, the proportion who
E69. a. Conditions are met. H0: The proportion, p1,
had heart attacks would have been different
of adults who logged on in 2003 is equal to
than had all patients been treated with Zocor.
the proportion, p2, of adults who logged on
in 2005. Ha: p1 < p2. The P-value, 0.037, is
b. Both conditions are met. H0: If all patients
could have been given Lipitor, the proportion,
less than = 0.05, so reject H0. You have
p1, who had heart attacks would be equal to
sufficient evidence that the proportion has
the proportion, p2, who had heart attacks
increased.
had they all been given Zocor. Ha: p1 < p2.
b. No; percentages less than 50% are in the 95%
The test statistic is still 1.74. Because the
confidence interval (0.479, 0.541).
P-value, 0.041, is less than = 0.05, reject
E71. B
H0, that if all patients could have been given
E73. Conditions were checked. H0: p1 p2 = 0.02,
Lipitor the proportion who had heart attacks
where p1is the proportion of all males who
would be the same as the proportion who had
are left-handed and p2 is the proportion of all
heart attacks had they all been given Zocor.
A one-sided test makes it easier to reject H0 if
females who are left-handed. Ha : p1 p2 > 0.02.
the difference is in the direction Hastates.
The P-value, 0.284, is greater than = 0.05, so
dont reject H0, that the difference in the two
P55. Type II; Type I; a Type I error means the patient
proportions is 2%.
receives a more expensive drug even though its
not more effective, and a Type II error means
Section 8.5
the patient is not given the more effective drug.
Type II seems more serious.
P52. Conditions are met. Suppose all the subjects could
P56. No randomization was done, so that condition
have been given aspirin and all could have been
is not met. H0: The difference in the proportion
given the placebo. Then you are 95% confident
of cold-turkey quitters who were successful
that the difference in the proportion who would
and the proportion of planned quitters who
get ulcers is in the interval 0.0028 0.0031.
were successful can be attributed to chance
Because 0 is in this interval, it is plausible that
variation alone. Ha: The difference is too large
there is no difference in the two proportions.
to reasonably be attributed to chance. With a
P53. a. yes
b. ( 0.348, 0.062)
P-value so close to 0, reject H0. Those who chose
c. Suppose it were possible that all the children
to quit cold turkey were more successful than the
could have been in both treatment groups.
others, and this difference cannot reasonably be
Then you are 90% confident that the
attributed to chance.
difference in the proportion who would get
P57.
a. observational study
b. (0.010, 0.058)
arrested is in the interval ( 0.348, 0.062).
c.
Zero
is
not
in
this
interval,
so the difference
Because 0 is not in this interval, it is not
in the proportion of people in this study
863
22-03-2009 21:30
Lesson
26 de 44
abused as children who later committed

crimes and the proportion of people not
abused as children who later committed
crimes cannot reasonably be attributed
to chance; no, because other contributing
factors werent controlled.
P58. greenways
E75. a. Conditions are met. You are 95% confident
that if all the skiers in the experiment could
have been in both treatment groups, then
the difference in the proportions who got
colds would have been in the interval
(0.011, 0.187). Because 0 isnt in this
confidence interval, you believe that vitamin
C reduced the proportion who got colds.
b. Because 0 is in the interval ( 0.016, 0.214), it
is plausible that the same proportion of these
skiers would have gotten a cold whether or
not they took vitamin C.
E77. Conditions are met. You are 95% confident that
if all the children had been given the placebo and
if all had been given the vaccine, the difference
in the proportions who developed polio would
have been between 0.0005 and 0.0002. You
believe that the vaccine reduced the proportion
of children who developed polio. (Although this
looks like a small difference, the vaccine cut the
incidence of polio about in half.)
E79. Conditions are met. H0: If all the men had been
given aspirin, the proportion, p1, who had a heart
attack would have been equal to the proportion,
p2, who had a heart attack if all had been given
the placebo. Ha: p1 < p2. With a P-value close
to 0, reject H0. If there is no difference in the
proportion who would have had a heart attack
had they all taken aspirin and the proportion
who would have had a heart attack if they had all
taken the placebo, then there is almost no chance
of getting a difference of 0.0091 or smaller in
the two proportions from a random assignment
of treatments to these subjects.
b. H0: The difference between the proportion, p1
of dementia-free people who exercise three
or more times a week and the proportion, p2,
of those with signs of dementia who exercise
three or more times a week can reasonably
be attributed to chance variation. Ha: The
difference cannot reasonably be attributed to
chance. Because the P-value, 0.006, is less than

= 0.05, reject H0. No; there is evidence of
an association between exercise and a delay
of dementia for the group of persons in this
study, but the reason isnt known.
E83. A difference as large as Reggies between the
regular season and the World Series would
happen by chance to fewer than 17 players
in 1000. Therefore, Reggies record is indeed
unusual and cannot reasonably be attributed to
chance variation alone.
E85. a. 7; a value this small would occur only about
3.5% of the time if H0 were true.
b. With a P-value of 0.035, reject H0, which was
also the conclusion in the example.
Chapter Summary
E89. a. Conditions are met. (0.754, 0.806)
b. You are 95% confident that if you were able to
ask all teenagers with online access, between
75.4% and 80.6% would agree.
c. If you could repeat this survey with 100
different random samples of teens with online
access, each of size 971, you would expect
the (unknown) proportion of all teens with
online access who agree with the statement to
be in 95 of the resulting (probably different)
confidence intervals.
E91. a. H0: The difference in the proportion of
snowboard injuries and ski injuries that
were fractures can reasonably be attributed
to chance variation. Ha: The difference is too
large to be attributed to chance alone. With
a P-value this close to 0, the difference in the
proportions of injuries that were fractures is
b. No. You need to know the total number of
snowboarders and skiers.
E93. a. 5
b. 0.00592
E95. 0.90; See the explanation in the dialogue on pages
478479 and in E23.
E97. This sentence is not correct: To be more specific,
the laws of probability say that if we were to
conduct the same survey 100 times, asking people
in each survey to rate the job Bill Clinton is doing
as president, in 95 out of those 100 polls, we would
find his rating to be between 47% and 53%.
22-03-2009 21:30
Lesson
27 de 44
distance walked before crossing a sideline

would be between 50.69 yd and 66.05 yd.
Section 9.1
E3. Conditions are met. You are 95% confident that
the mean weight of the bottles produced that day
P1. a. (4.335, 5.704)
is between 15.92 oz and 16.03 oz. Because 16 oz
b. The center is the same; too narrow; smaller.
is one of the plausible values for the population
P2. a. 2.262
b. 2.447
c. 3.106
mean, there is no need to adjust the machine.
d. 2.920
e. 2.704
f. 2.639
E5. a. No. This is not a random sample of bags of
P3. a. (7.905, 46.095) (7.908 to 46.092 using Table B)
fries.
b. (3.694 to 8.306)
b. If this were a random sample, then you
would be 95% confident that the mean mass
c. (16.58, 34.577) (16.572 to 34.572 using
of a small bag of fries at this McDonalds is
Table B)
between 69.926 g and 77.511 g (69.92 g and
(4.23,
5.81);
wider
P4.
77.52 g using Table B).
P5. a. (97.969, 98.231)
c. No. The confidence interval contains 74 g, so
b. You are 95% confident that the mean body
74 g is a plausible value for the mean mass of
temperature of all men is between 97.969F
a small bag of fries.
and 98.231F.
E7. a. no, because there is no indication that this is
c. No; the mean body temperature of all men is
a random sample of brochures
less than 98.6F.
b. (8.895, 10.705)
P6. a. No, but a sample of size 61 is large enough
c. If these 30 could be considered a random
that this should not matter.
sample of all brochures and if the readability
b. (8.93, 11.59)
found on the sampled page for each is typical
c. You are 90% confident that the mean number
of that brochure, you are 95% confident that
of hours of study per week for all students
if you tested the readability of all cancer
taking this course is between 8.93 and 11.59.
brochures the average reading level would be
90% confidence means that if you were to
between grades 8.895 and 10.705.
take 100 random samples of size 61 from
E9. a. third interval
b. first interval
the population of all students who take this
c. third interval
course and calculate a confidence interval for
E11. B
each sample, you would expect 90 to contain
the mean number of hours of study for all
E13. a. increases
b. decreases
c. increases
students.
E15. a. Skewed toward larger values; s is too small on
P7. D
average.
P8. a. false
b. false
c. false
b. s is smaller than = 112 in at least 61 of the
100 samples.
d. false
e. true
c. The advertised 95% confidence intervals will
P9. a. men: 0.397; women: 0.377
be too narrow more often than they will be
b. for the men, because the SD is larger than
too wide when s is used as an estimate of .
that for the women
This means the actual capture rate will be less
E1. a. No; this is not a random sample but a group
than 95%.
of volunteers.
b. (50.69, 66.05)
Section 9.2
c. If this were a random sample of people,
P10. t = 2.9314
you would be 99% confident that if you
P11. t = 1.0853
blindfolded all people and had them attempt
to walk the length of a football field, the mean
P12. The lighter graph is the standard normal
distribution because it has less area in the tails.
Chapter 9
865
22-03-2009 21:30
Lesson
28 de 44
P13. a. One way is to use randNorm(100,15,7) and

then calculate t.
P14. If the mean temperature at your desk is actually
72 F, the probability that temperatures taken on
seven randomly selected days would give a value
of t greater than 2.9314 or less than 2.9314 is
about 0.0262.
P15. If the mean Pell Grant for Minnesota college
students is still $2178.82, the probability that
a random sample of 35 students would have a
t-statistic greater in absolute value than 1.0853 is
about 0.285.
P16. a. H0: The mean temperature at your desk, , is
72F. Ha : 72F.
b. yes; yes; no
P17. a. H0: The mean Pell Grant in Minnesota, , is
$2178.82. Ha : $2178.82.
b. no; no; no
P18. a. H0: The mean SAT score of State University
students, , is 1700. Ha : 1700.
b. yes; no; no
P19. This might not be a random sample, but other
conditions are met. Because the t-value, 2.918,
exceeds the critical value, t* = 2.262, reject H0.
The data suggest that the mean aldrin level differs
from 4 nanograms.
P20. For males, with t 4.10 and P-value 0.003,
reject H0: = 98.6, where is the mean body
temperature of all men in the population under
study. For females, with t 0.48 and P-value
0.64, do not reject H0.
P21. a. Type I error b. Type I error
c. Type I error
= 0.05
P22. a. n = 16
b.
P23. a. one-sided (right-tailed)
b. x is the mean selling price for a sample of
houses sold this month, and is the mean
selling price of all houses for this month.
c. H0: is the same as the mean for last month.
Ha: is greater.
P24. Conditions are met. Because the observed
t-statistic, 2.931, is farther out in the tail than
the critical value, t* = 1.943, reject H0. The
data suggest that the mean temperature at your
desk is less than 72F. Alternatively, if the mean
temperature at your desk is indeed 72F, there
is only a 1.3% chance of getting a result from a
sample of size 7 as extreme as or more extreme

than the result from your sample. So there is
sufficient evidence to support the claim that the
mean temperature at your desk is less than 72F.
E17. Conditions are met. If the mean bottle weight
were 16 oz, a t-statistic at least as large in absolute
value as 1.022 would occur about one-third of
the time. This means you do not have sufficient
evidence to reject H0. The data suggest that it is
plausible that the mean amount of water in the
bottles is 16 oz.
E19. This probably is not a random sample of students,
but other conditions are met. Reject H0. If it were
true that the mean average error is 0, a t-statistic as
extreme as or more extreme than 3.943 in absolute
value would occur only 0.15% of the time.
E21. a. one-sided (left-tailed)
b. x is the mean freezing point for a sample of
ten bowls of salt water; and is the freezing
point for salt water with this degree of salinity.
c. H0: = 32F; Ha : < 32F
E23. Conditions are met. If it is true that the mean
weight of Munchies Potato Chips is 10 oz, the
probability of getting a t-statistic less than 2.55
is only 0.012. Reject H0, because there is sufficient
evidence that the mean weight of a bag of
Munchies Potato Chips is less than 10 oz.
E25. The chance of getting a random sample of
young women with a t-statistic of 8.73 or larger
is 2.7 10 6.The difference in the mean height
of the players and the mean height of all young
women cannot reasonably be attributed to
chance variation.
E27. Because the P-value, 0.333, is greater than
= 0.10, do not reject H0. The data suggest that
it is plausible that the mean amount of water in
the bottles is 16 oz.
E29. a Because the P-value, 0.012, is less than
= 0.05, reject H0. There is sufficient
evidence that the mean weight of a bag of
Munchies Potato Chips is less than 10 oz.
b. No error was made.
c. The 95% confidence interval for the mean
weight is (9.73, 9.98), which suggests that the
mean weight is less than 10 oz.
E31. D
E33. a. The P-value is 2(0.057).
22-03-2009 21:30
Lesson
29 de 44
b. The one-sided P-value is half as large as the

two-sided P-value.
c. two-sided test
E35. Both have probability 0.05.
E37. a. = 0.10
b. n = 45
c. one-sided test (assuming the alternative
hypothesis is in the correct direction)
E39. a. They are both mound-shaped and centered at
about 0. A has a smaller spread.
b. B, because it has the larger spread
Section 9.3
D. I
P25. A. II
B. III
C. IV
P26. a. Do the analysis with and without the outlier.
b. This isnt a random sample of bags of fries,
but other conditions are met. In both cases,
the P-value is very high (0.71 with the outlier
and 0.87 without). Do not reject H0. Neither
analysis provides statistically significant
evidence that the mean number of fries is
different from 90.
E41. a IV
b. III
c. I
d. II
E43. a. yes
b. You are 90% confident that the mean LOS for
Insurer A is in the interval (2.22, 2.42).
c. no for n = 40, but yes for n = 4
E45. a. No; do a log transformation.
b. You are 90% confident that the mean log of
the mass (in 100,000 solar units) of a black
hole is between 7.70 and 8.21.
E47. a. No. Still more outliers are created, which is
typical of strongly right-skewed data.
b. (102, 687); (69, 229)
c. Without outliers, the center is much lower
and the width is much smaller. Confidence
intervals apparently are quite variable when
the distribution is highly skewed.
d. You are 95% confident that the mean of the
natural log of the weights of mammals brains
is between 2.33 and 3.63.
E49. a. transform
b. On average, it takes 0.081 year to get an inch
of rainfall; not quite, because there are still
two outliers and a slight skew to the right.
c. much better
d. no, because a sample of size 128 is large

enough that the sampling distribution of
would be approximately normal
E51. a. The stemplots show marked skewness for the
P/H data but near symmetry for the H/P data.
There are outliers in both distributions.
b. The interval (1.465, 3.674) contains the set
of plausible values for the mean number of
people per housing unit in the population of
Florida counties.
c. You are 95% confident that the true mean
number of housing units per person by
county is between 0.355 and 0.559.
d. (1.789, 2.817), which is quite different from
the confidence interval in part b
e. housing units per person
Section 9.4
P27. a. yes
b. ( 12.760, 15.899)
c. the difference between the mean distance
all left-handed volunteers will walk before
crossing a sideline and the mean distance
all right-handed volunteers will walk before
crossing a sideline
d. no, because 0 is in the confidence interval
P28. Conditions are met except for the outliers on the
high end. Including them, you are 90% confident
that 0.32 <
< 5.79. Excluding
them, the interval is 1.27 <
< 5.75
and no longer contains 0. Because the results are
so different, it would be good to gather more data
to resolve the issue.
P29. Conditions are met except for the outliers. With
the outliers, the P-value is 0.1392; without the
outliers, the P-value is 0.012. Because the results
are so different, it would be good to get more data.
P30. a. Results should be viewed with some caution
because of the outlier. The P-value, 0.0568, is
a bit larger than = 0.05, so this is not quite
sufficient evidence to reject the hypothesis
that any observed difference in means is the
result of the randomization of the assignment
of treatments to the babies.
b. The P-value is 0.4038, so there isnt sufficient
evidence to reject the hypothesis of equal
means for the exercise control and weekly
report groups. An observed difference in
867
22-03-2009 21:30
Lesson
30 de 44
means of this size could quite reasonably be

due to the random assignment of treatments
to the babies.
c. The special exercises group is very close
to having a statistically significantly lower
mean than the weekly report group. The next
strongest comparison is between the special
exercises and exercise control groups, and the
weakest is between the exercise control and
weekly report groups.
P31. The measurements from the bottom have greater
variability, so get most of the new measurements
at the bottom to give your test greater power.
P32. a. The plausible values for the true difference
in mean body temperatures, female male,
lie in the interval (1.149, 0.131). Or
(1.038, 0.242) if you use the summary
statistics.
b. Because this interval is entirely below
0, conclude that there is a statistically
significant difference between the mean
body temperature of men and of women.
c. one interval for the difference
E53. a. yes, but with some caution because of the
high value for short days
b. (0.50, 9.46)
c. the difference between the mean enzyme
concentration of all eight hamsters had they
all been raised in short days and the mean
concentration had they all been raised in
long days
d. yes, because 0 is not in the confidence
interval
E55. a. Conditions are met. A 95% confidence
interval for the difference in initial mean pain
levels is (7.092, 10.092), which includes 0.
There isnt statistically significant evidence
that the randomization failed to yield groups
with comparable means.
b. You are 95% confident that the difierence
between the mean body mass indices
of the two groups is in the interval
(1.587, 2.5867), which includes 0. There
isnt statistically significant evidence that
with comparable means.
E57. a. Use absolute values.
b. This amount of skewness should be okay for

inference. If there were no difference in the
means of the absolute values of the errors
of the vertical line group and the horizontal
line group, a t-statistic at least as large as 2.09
would occur only about 2.4% of the time.
With a P-value this low, reject H0 . There is
sufficient evidence to support the claim that
students marking the midpoint on vertical
line segments have a larger mean absolute
error than the students marking the midpoint
on horizontal line segments.
E59. a. No. No randomization is mentioned. Groups
of size 8 are not large enough to compensate
for this degree of skewness.
b. If the mean number of termites remaining
had all dishes received a dose of 5 mg of
resin is equal to the mean number remaining
had all dishes received a dose of 10 mg of
resin, a test statistic at least as extreme as
2.153 would occur in 5.89% of the possible
randomizations. You dont have statistically
significant evidence that the size of the dose
makes a difference in the mean number of
termites remaining.
c. The main concerns are the lack of
randomization and the skewness of the
distribution of termites remaining in the 10
mg group. Repeat the experiment, randomly
assigning treatments to the dishes and using a
larger number of experimental units.
E61. With 95% confidence, 4.71 < men women<
9.94, where men is the mean heart rate for all
men in this population and women is the mean
heart rate for all women in this population.
Because 0 is in the confidence interval, there is
insufficient evidence to say that the mean heart
rate for men differs from the mean heart rate
for women. Alternatively, the P-value is 0.4663,
so there isnt statistically significant evidence
that the mean heart rates differ. If the outlier
among the mens heart rates is dropped, the 95%
confidence interval for the difference in means
becomes 2.76 men women < 10.94, and the
conclusion remains the same.
E63. a. Type I
b. Decrease the level of significance.
c. There was no overlap in the concentrations
for the two groups.
22-03-2009 21:30
Lesson
31 de 44
E65. a. more like a survey of peoples ability to walk

straight while blindfolded
b. If these were independent random samples
and there was no difference between the
mean distance men and women could walk,
a t-statistic greater than 2.774 or less than
2.774 would occur with probability 0.0098.
Reject H0. There is strong evidence that there
is a difference between these mean distances.
c. The lurking variable is the persons height.
Taller people tend to walk farther and men
tend to be taller than women, so gender or
height could explain the difference.
E67. a. The problem does not say whether these are
independent random samples. A test statistic
of only t = 1.50 and a P-value of 0.0825
do not allow rejection of H0 , that coaching
improves SAT math scores by 25 points.
b. Using a one-sample t-test to compare the
average gain, 73, with the expected mean
gain, 25 + 13 or 38, yields a P-value of
0.0074. Reject H0 and conclude that coaching
improves the average gain by more than the
expected 38 points.
Section 9.5
P33. a. repeated measures; completely randomized
b. Conditions are met. With a P-value of 0.005,
reject H0. There is sufficient evidence to
conclude that standing increases the mean
pulse rate by more than would be expected by
chance for this group of subjects.
c. Conditions are met. With a P-value of 0.0023,
reject H0 . There is sufficient evidence to
conclude that standing does increase the
mean pulse rate by more than would be
expected by chance.
d. Conditions are met. With a P-value less
than 0.0001, reject H0. Once again, there is
sufficient evidence to conclude that standing
does increase the mean pulse rate by more
than would be expected by chance. The
evidence is even stronger than in the matched
pairs design.
e. yes
P34. a. independent samples
b. log transformation
c. Conditions are met. With a P-value of 0.0248,

reject H0. There is sufficient evidence to
conclude that the mean brain weights of
species of birds and species of fish differ.
P35. a. Conditions could not be randomly assigned.
b. With a P-value of 0.33, the difference in lung
clearance between the twins can reasonably
be attributed to chance alone.
c. Independently, take random samples from
rural and urban populations.
d. The analysis of differences reduces the
person-to-person variation.
e. You would not have to find a criterion on
which to pair the subjects.
P36. No; statistics isnt needed to know that hens eggs
are longer than they are wide.
P37. a. No; there is no random assignment into
treatment groups.
b. It is better, but there still is no random
assignment of treatments.
c. Randomly divide a group of students into the
two treatment groups.
E69. a. one-sample test of the mean of the
differences; one-sided test
b. Conditions are met. Because the P-value,
0.0016, is less than = 0.01, reject H0 . If
these groups can be considered a random
sample of all possible teams of students, you
have statistically significant evidence that the
mean distance on the tenth launch is greater
than the mean distance on the first launch.
c. yes, because teams should improve with
practice; Type I error
d. no, because the sets of distances are not
independent samples
E71. a. one-sample
b. yes
c. You are 95% confident that the mean
difference between the distances of bears
launched using four books and the distances
of bears launched using one book is between
23.07 in. and 146.74 in. Because 0 is in
the interval, there isnt sufficient evidence to
support the claim that launch angle makes a
difference in how far gummy bears soar.
E73. a. The 95% confidence interval for the
difference bottommid-depth is (0.5655, 1.4145).
There is statistically significant evidence that
Brief Answers to Selected Problems 869
22-03-2009 21:30
Lesson
32 de 44
the mean difference is not 0, if the conditions

for inference are met.
b. no
c. one-sample test of the mean difference
E75. a. two-sample test
b. Conditions are met. If the higher step did
not increase the mean gain in heart rate, a
t-statistic larger than 2.778 would occur with
probability 0.0051. This is strong evidence
against H0. There is statistically significant
evidence that using the higher step increases
heart rate more than does using the lower step.
E77. a. H0: d = 0; Ha: d > 0, where d is the mean
difference in the number of stings
b. It is unclear whether any randomization took
place. The distribution of the differences is
fairly symmetric with the exception of one
outlier. Do the analysis both with and without
the outlier.
c. With a P-value of 0.0542 or 0.0528 (without
the outlier), dont reject H0 . There isnt
statistically significant evidence that there
would have been more new stingers in the
previously stung cotton balls than in the fresh
cotton balls had Free been able to give all
cotton balls both treatments.
E79. a. two independent samples
b. Pair rooms, one with carpet and one without,
that are near each other and have similar
uses. Alternatively, use a repeated measures
design, giving each room both treatments
and randomizing which eight rooms get the
carpet first.
E81. It is not the center but the variability that is of
interest.
E83. a. two independent samples
b. Pair students on their ability to estimate, and
then randomly assign one of each pair to
feet and one to meters. Or have each student
estimate in both feet and meters.
c. Matching students on their ability to estimate
distances would be difficult. Repeated
measures might not work, because the
student might change one estimate to the
other by multiplying by a constant.
d. Because one mean is in feet and the other is
in meters, of course the means arent going
to be equal. Further, the question asked was
whether the students were as accurate in

meters as they were in feet, and these data
cant answer that question because you dont
know the true length.
E85. a. yes, if eruptions can be selected at random
b. unpaired
E87. a.
;
the first estimates the mean difference for
all accounts, and the second estimates the
mean difference for all accounts with nonzero
differences. The latter is the more reliable
estimate because the conditions are better met.
b. You would expect about (1000), or 375, of
the 1000 accounts to show a difference between
the book and audit values. Multiplying both
ends of the interval by 375 gives a confidence
interval for the total, (22.5, 12,727.5).
Chapter Summary
E89. a. yes
b. You are 95% confident that if you were to
ask all highly trained athletes how long they
sleep, the mean number would be between
5.63 h and 7.71 h.
E91. Conditions are met. If it is true that the mean
distance had all students used four books is
the same as the mean distance had all students
used one book, a t-statistic greater than 3.65 or
less than 3.65 would occur with probability
0.0005. Reject H0. There is statistically significant
evidence that the number of books changes the
mean distance of gummy bear flight.
E93. Conditions are met. You are 95% condent that
the mean difference between the life expectancies
of females and males is in the interval (1.493,
5.147). There is sufficient evidence to say that the
difference between the means is positive, which
implies that the population mean life expectancy
for females is larger than that for males. (Using a
test gives a P-value of 0.0026.)
E95. D
E97. a. Transform the salaries.
b. Proceed if the firm has more than 500 people.
E99. a. yes, because the chance of getting a randomly
selected team of 16 players with a batting
average of .276 or higher is only 0.0188 (Use
because is known to be .050.)
22-03-2009 21:30
Lesson
33 de 44
b. no, because the chance of getting a randomly

selected team of 16 players with a batting
average of .256 or higher is 0.3156
Chapter 10
Section 10.1
P1. a. 12.5
b. 2 = 3.28
P2. 0
P3. P-value is small (about 146/5000 or 0.0292).
Reject the H0 that the die is fair.
P4. P-value is near 0. Reject the H0 that the die is fair.
P5. a. 2 * with df = 3 is 7.81. Reject the H that the
0
die is fair.
b. 2 * with df = 19 is 30.14. Do not reject the H0
that the die is fair.
P6. a. 0.9194
b. 0.0304
P7. Conditions are not met, as this is a convenience
sample. With 2 = 1.25 and df = 3, the P-value is
0.7410. Do not reject the H0 that the blood types
O:A:B:AB occur in the ratio 9:8:2:1.
P8. a. Conditions are not met, as this is an
observational study. The expected count in
each category is 117.5. With 2 = 16.723
and df = 3, the P-value is 0.0008. Reject the
H0 that seizures are equally likely during
these four phases of the moon. The largest
deviation occurs during the last quarter.
There were fewer seizures during the full
moon than the model predicts.
b. Were the quarters of equal length? Over how
many months were these data collected?
P9. a. Conditions may not be met because your class
probably will not be large enough to have
expected frequencies of 5 in each category.
b. Combining the last three categories into 2 or
more, with probability 0.275, 2 = 44.20 with
df = 2 gives a P-value of 2.5 1010. Reject
the H0 that the distribution of the number of
children in the families of statistics students
is consistent with the distribution for families
nationwide.
P10. a. Conditions are met. With z = 2.80, the
P-value for a two-tailed test is about 0.005.
Reject the H0 that spinning a coin is fair.
b. Conditions are met. With 2 = 7.84 and

df 1, the P-value is about 0.005. Reject the
H0 that spinning a coin is fair.
c. The P-values are equal and z2 = 2.
P11. a. Conditions are met. With z = 5.4, the
P-value for a two-tailed test is 6.7 108.
Reject the H0 that spinning a checker with
putty on the crown side is fair.
b. Conditions are met. With 2 = 29.16 and
df = 1, the P-value is 6.7 108. Reject the H0
that spinning a checker with putty on
the crown side is fair.
c. The P-values are equal and z2 = 2.
E1. Conditions are met. With 2 = 94.189 and
df = 5, the P-value is 8.84 10 19. Reject the
H0 that the die is fair.
E3. Conditions are not met, as this is probably
an observational study. With time periods of
15, 46, 120, and 184 days, 2 = 1.39, df = 3,
P-value 0.7080. Do not reject the H0 that the
number of people admitted in each time period
is proportional to the length of that time period.
E5. With 2 = 0.47 and df = 3, the probability of a
chi-square statistic less than or equal to 0.47 is
0.0746. The data fit the model so well that it casts
some doubt on the conduct of the experiment.
E7. Crossley: With 2 = 56.67 and df = 3, the
P-value is about 3 1012
Gallup: With 2 = 108.27 and df = 3, the P-value
is about 3 1023
Roper: With 2 = 229.07 and df = 3, the P-value
is about 2 1049
None of the three polls are close to fitting the
actual results, but Crossley fits best.
E9. Conditions are not met, as these are the
populations of fatalities for the selected days
and hours.
a. With 2 = 47.10 and df = 1, the P-value is
6.7 10 12. Reject the H0 that the difference
between the number of traffic fatalities during
the first four hours after the Superbowl and
(half) the number during those hours on the
previous and the following Sundays can be
attributed to chance variation.
b. With 2 0.36 and df = 1, the P-value is
0.5485. Do not reject the H0 that the number
involved in traffic accidents fits the 1:2 model.
22-03-2009 21:30
Lesson
34 de 44
The observed difference in fatalities could

reasonably be attributed to chance variation
alone.
E11. Answers will vary.
E13. a. As the sample size increases, the values of
sample statistics such as proportions tend to
get closer to the true population values, thus
improving the power to reject a false H0 .
b. Yes, if H0 is false, all else being equal, the
2
statistic tends to be larger with larger n
than with smaller n, giving you more power
to detect the false H0 . Also, the sampling
distribution of 2 becomes closer to the
continuous chi-square distribution.
E15. a. Each digit should occur of the time.
b. The rivers are not randomly selected, but the
final digit of the lengths can be regarded as
somewhat random from the population of final
digits of measured rivers. With 2 = 131.77
and df = 9, the P-value is 5.13 1024. Reject
the H0 that the final digits are equally likely.
Section 10.2
P12. b. It appears that females are more likely than
males to prefer a bath.
P13. a. marginal
b. conditional
c. marginal
d. neither
P14. For both males and females, expected frequencies
are 13.5 and 13.5 for bath and 36.5 and 36.5 for
shower.
P15. b.
P20. Yes. With 2 = 4.82 and df = 1, the P-value is

between 0.025 and 0.05 (or 0.0281).
P21. With 2 = 11.11 and df = 2, the P-value is
between 0.0025 and 0.005 (or 0.0039). Reject
the H0 that if both treatments could be assigned
to all subjects, the distributions of outcomes
would be the same. The nitric oxide treatment
resulted in more Not severe hemorrhages and
fewer Severe ones.
E17. a. The observed frequencies for 2003 are 190,
200, 90, and 520.
b. the populations of adult residents of the
United States in the respective years
c. Conditions are met. With 2 = 113.4 and
df = 9, the P-value is almost 0. Reject the
H0 that the distributions of responses in the
populations are the same for each year.
E19. b. z = 2.915; If all subjects could have been
assigned each treatment and which treatment
the subject received made no difference
in survival, the probability of getting a
difference in sample proportions as large in
absolute value as from the actual experiment
is only 0.0036.
2
= 8.495; If all subjects could have been

assigned each treatment and which treatment
the subject received made no difference
in survival, the probability of getting a 2
statistic as large as or larger than from the
actual experiment is only 0.0036.
d. The P-values are equal and z2 = 2 .
E21. Conditions are met. With 2 = 34.575 and
df = 4, the P-value is close to 0. Reject the H0
c. first row: 11.33, 22.00, and 6.67
that the three treatments would give the same
d. No. Population III appears to have too many
distributions if all children could have received
members that fall into Categories A and B
each treatment.
and too few in the others.
E23. Conditions are met. With 2 = 17.57 and df = 6,
the P-value is between 0.005 and 0.01 (or 0.0074).
P16. 2 = 11.42; df = 1; If male and female students
Reject the H0 that the distributions of responses
have the same preferences, the probability of
getting a test statistic of 11.42 or larger is less
in the population are the same for these years.
than 0.001 (or 0.0007).
E25.
Conditions are met. Reject the H0. A value of 2
2
P17.
64.84; df = 8; yes
as large as or larger than 18.19 (or a P-value as
P18.
small as 0.0011) is extremely unlikely to occur if
the proportion of the population in each category
is the same for each country. The United States
P19. The P-value is between 0.15 and 0.20 (or 0.1826).
appears to have less violent crime and a larger
proportion of non-victimized households.
Dont reject the H0 that the proportion of people
in each category is the same for each country.
c.
22-03-2009 21:30
Lesson
35 de 44
E27. Conditions are met. Reject the H0 . A value of

2
as large as or larger than 271.56 (or a P-value
close to zero) is extremely unlikely to occur
if the distribution of responses throughout
the populations is the same for each country.
The United States appears to have a smaller
proportion of individuals who report drinking
every day, a few times a week, or about once a
week than either Canada or Great Britain, and a
larger proportion who report never drinking.
E29. a. No. Some Internet users will fall into more
than one category and some into neither.
b. Yes. If the proportion of the population that
uses the Internet for sending and reading
e-mail hasnt changed, the probability of
getting a test statistic as large as or even larger
than that from this test is 0.0914.
E31. One meaning of homogeneous is essentially
alike. Homogeneous populations would be
populations that are alike in the characteristics
you are observing.
Section 10.3
P22. a. not independent
b. not independent
c. depends on recent fashion trends
d. not independent
e. independent
f. not independent
P23. A and C
P24. Expected frequencies for right-handed are 955.86
and 1048.14.
P25. Conditions are met. Reject H0 . If gender and
handedness are independent, the probability of
getting a value of 2 as large as or larger than
11.81 from a random sample of size 2237 is only
0.0027. It appears that women are more likely to
be right-handed and men are more likely to be
left-handed or ambidextrous.
P26. a. test of independence
b. Yes, with 2 of 21.09, the P-value is about
0.0001 so there is evidence that gender is
associated with effort at eating a healthy diet.
c. Yes, but the evidence is much weaker.
(Depending on rounding, 2 is cut in half
to about 10 or 11 and the P-value is 0.012 to
0.019.)
P27. If the sample is random, the conditions are
met. Reject the H0 . If, among all students at this
middle school, having breakfast three or more

times a week and thinking they have a healthy
diet are independent, the probability of getting
a chi-square of 4.475 or higher from a random
sample is 0.0344.
P28. a. With 2 18.37, the P-value is 0.000018.
There is a statistically significant association
between gender and attainment of a high
school education. You have strong evidence
of a weak association.
b. With 2 475.13, the P-value is effectively 0.
There is a highly statistically significant
association between gender and attainment
of a high school education. The association is
stronger than in part a.
c. In part a, the difference is too small to be of
much practical significance. In part b, the
difference is larger and more meaningful.
E33. a. In some regions (such as Alaska and parts
of the Southwest), adults are younger than
in other regions and so would tend to have
children in lower grade levels.
b. for K8: NE 0.1159; MW 0.1525; South 0.2519;
West 0.1697
c. for K8: NE 5,630,698; MW 7,407,049;
South 12,233,362; West 8,244,951
E35. a. Conditions are met. Dont reject the H0. If
a random sample this size is taken from a
student population where class year and
favorite team sport are independent, there is
a 30.4% chance of getting a 2 value of 10.596
or larger.
b. no, as you havent rejected the H0
c. Take a random sample of, say, 80 freshmen
and ask them their favorite team sport.
Repeat with sophomores, juniors, and
seniors.
E37. b. With 2 456.9 and a P-value close to
0, reject the H0 that the difference in the
proportions of males and females who
survived can reasonably be attributed to
chance. These are not the results you would
expect if people were placed on lifeboats
without regard to gender.
b. homogeneity
c. You do not have a random sample of Sundays,
or even of Super Bowl Sundays. If the
probability that a person in an alcohol-related
22-03-2009 21:30
Lesson
36 de 44
accident is killed is the same the first four

hours after a Super Bowl as it is during the
same four hours on a Sunday before or after
a Super Bowl, a 2 as high as or higher than
138.69 is extremely unlikely. Reject the H0 .
Something other than random variation must
account for the high incidence of fatalities
following a Super Bowl.
Chapter Summary
E41. Conditions are met for a test of homogeneity.
With 2 66.03, the P-value is about 0.000015.
Reject the H0 that if in every year from 2001
through 2005 and in 1997, you had asked all
adults in the United States this question, the
distribution of responses would be the same for
each year. The most significant trends are that
fewer people have no opinion and more people
think global warming has already begun.
E43. b. The infirmary was used most heavily on
Monday and Thursday. There were relatively
few admissions on Friday. Perhaps students
saved their problems until Thursday after
classes were over and perhaps the infirmary
wasnt open weekends, so students who
got sick on the weekend had to wait until
Monday.
c. yes for test of independence, as there was one
large population that was classified according
to day and severity of problem; no, as this
isnt a random sample
d. Because the table includes all students who
visited the infirmary during a school year,
these data cannot reasonably be considered
a random sample taken from one large
population. With 2 11.023 and P-value
0.527, dont reject the H0. These are typical
of the results you would see if there were no
association between severity of problem and
day of the week.
e. You might group Monday, Tuesday, and
Wednesday together and Thursday and
Friday together.
f. With 2 3.463, and a P-value of 0.326, the
association is not statistically significant here
either. The differences in the proportions of
the various problems reported on weekdays
when students attend class and on weekdays
when students dont attend class can

reasonably be attributed to chance.
E45. With 2 0.88 and a P-value of 0.644, there
isnt sufficient evidence to reject the percentages
claimed.
E47. a. homogeneity
b. Conditions are met. Dont reject the H0 . You
can attribute the differences to the fact that
you have only a sample of adults from each
educational level and not the entire adult
population. A value of 2 of 3.202 or even
larger is quite likely to occur ( P-value close
to 0.20) for three samples of this size if the
same proportion of adults in each population
watched the Super Bowl.
E49. a. first row: 20, 16
c. no
E51. a. Yes, conditions are met for a test of goodnessof-fit. With 2 = 5.456 and a P-value of
0.2436, dont reject the H0. You dont have
statistically significant evidence that the
distribution of grades differs from that of 1996.
b. Because you know the population standard
deviation for 1996, = 1.30, you can use
that value in the z-statistic. With z = 1.088
and a P-value for a right-tailed test of 0.1383,
dont reject the H0. You dont have statistically
significant evidence that the mean grade in
2001 increased over that of 1996.
E53. This sample cannot reasonably be considered
a simple random sample taken from one large
population. With 2 133.052 and a P-value
close to 0, reject the H0. These are not results
like those you would expect if people were
placed on lifeboats without regard to class of
travel. The explanation apparently is that firstclass passengers were the first to be allowed into
lifeboats and third-class passengers were last.
E55. a. B and I b. A and III c. C and II
Chapter 11
Section 11.1
P1. a. 9 calories per gram of fat
b. the number of calories in a serving
containing no grams of fat
c. Calories in pizza come from carbohydrates
and protein as well as from fat; there
22-03-2009 21:30
Lesson
37 de 44
P2. a.
b.
c.
d.
P3. a.
b.
c.
d.
may be other sources of variation in the

measurement process.
y = 0 + x, where x is the arm span and y
is the mean height at that arm span.
= 7.915 + 0.952x
The height increases, on average, by about
0.952 inch for each 1-inch increase in
arm span.
The estimated slope of 0.952 is very close
to Leonardos theoretical slope of 1. The
estimated intercept is 7.915, which is quite far
from Leonardos theoretical intercept of 0.
All of the negative deviations are for larger arm
spans, and the plot shows a downward trend.
The residuals look more random than the
values of with no downward trend.
= b0 + b1 x
opening day = 97.37 + 1.75 swe
If swe = 31.0, opening day = 151.62 or 152
(June 1).
The random variation, should be relatively
large, given all the conditions that can affect
the opening date.
P4. 0.341366
P5. a. 5; greater variability in the conditional
distributions of y results in a larger value
in the numerator of .
b. 3; a smaller spread in the values of x results
in a smaller value in the denominator of .
c. 10; all else being equal, a larger sample
size tends to result in less variability in the
estimates of parameters.
d. The theoretical slope does not matter, all else
being equal.
e. The theoretical intercept does not matter, all
else being equal.
P6. a. 0.1472
b. = 1.71525 + 0.52490x
c. The first value is the slope of the regression
line, b1 . The second value is the estimated
standard error of that estimated slope, as
calculated in part a.
P7. a.
b. s = 3.127, close to the theoretical value of 3

c. = 0.3496 is close to 0.335.
d. yes
P8. V, II, I, IV, III
P9. I, III, II, IV.
E1. a. Yes for (hp, price), but note there is a
tendency for the spread in y to increase as
x increases. In the plot of (hp, mpg), there is
a fairly strong negative trend with a hint of
curvature. Yes for (l, mpg), except for two
potentially influential points on the left.
No for (rpm, mpg), because the variation in
responses is so great at the larger values of
x that no linear trend is apparent.
b. b1 = 0.069. For every increase of 1 unit in
horsepower, the gas mileage tends to drop by
0.069 mile per gallon. = 0.0233.
c. The estimated slope appears under Parameter
Estimates in row HP and column
Estimate. The standard error is found in
column Std Error and row HP. The value
of s is the Root Mean Square Error = 4.7785.
s is the estimate of the common variability in
the miles per gallon for each fixed horsepower.
E3. a.
, or about 3.1416. For every 1 cm increase
in the distance across, the circumference will
increase by 3.1416 cm.
b. The linear model that fits the situation
perfectly is C = d. However, it is difficult to
measure C and d with great accuracy.
E5. a. The soil samples should have the larger
variability in the slope because the distance of
y from the regression line tends to be larger
compared to the spread in x.
=0.3617. As predicted, the standard
b.
error for the soil samples is much larger.
E7. a. The plot shows a linear trend with large but
homogeneous variation in heights across
values of age.
b. The plot shows a linear trend with
homogeneous variation that is smaller than
the variation in part a.
c. The plot shows a linear trend with variability
in height increasing with age.
22-03-2009 21:30
Lesson
38 de 44
E9. Choose a value for x.

Predict the mean value of y by finding through
Generate a value at random from a normal
distribution with mean 0 and standard deviation
approximately equal to s = 0.3414.
Add that value to the calculated above. This is
your estimate of the redness for a rock with sulfate
percentage x.
Complete the above four steps for each value of x
in the study.
Fit a least squares regression line through the
resulting points.
Use the regression line to predict the redness
when x is 2.72, the value for Half Dome.
This model is reasonable given the limited
information we have, but it may not be reasonable
given that this is a sample of only a few rocks and
we cant be sure that (1) the relationship is linear,
(2) the variation in y is constant for all values of x,
or (3) the distribution of the errors is normal.
Section 11.2
P10. t = 0.3688. With df = 4, the P-value is greater
than 2(0.25) = 0.5 from Table B and 0.731 from
a calculator. Do not reject the H0 that the true
slope is 0.
P11. Small (in absolute value); there is no pronounced
linear trend and the variation in the residuals is
large.
P12. t = 5.6982. With 11 2 = 9 degrees of freedom,
the P-value from the calculator is 0.0003. Reject
the hypothesis that there is no linear relationship
between percentage of sulfur and redness.
= 25.232 + 3.291x. The temperature rises,
P13. a.
on average, about 3.291 degrees if the number
of chirps per second increases by 1.
b. The residual plot shows no obvious pattern
and little heterogeneity, so a linear model fits
the data well. The dot plot of the residuals
shows no outliers or any other indications of
non-normality.
c. The P-value is close to 0. Reject the H0 that
there is no linear relationship between rate of
chirping and temperature.
P14. Conditions are not met because this is not
a random sample from any well-defined
population. The plot of these data does not

show a linear relationship. The residuals have a
slightly skewed distribution. Proceeding anyway,
the regression equation is IQ = 0.9969 head
circumference + 45.050.
With a test statistic t 0.5904 and 18 degrees of
freedom, the P-value is 0.5622. Do not reject the
H0. There is no statistically significant evidence of
a linear relationship between head circumference
and IQ.
P15. a. You are 90% confident that the slope of the
true regression line for predicting titanium
dioxide content from silicon dioxide content
for Mars rocks is in the interval 0.0429 to
0.0245.
b. s = 0.0257 is an estimate of , the standard
deviation of the errors from the true regression
line. It can also be interpreted as an estimate
of the variability in the titanium dioxide
percentage at each fixed silicon dioxide
percentage.
P16. You are 90% confident that the slope of the true
regression line for predicting titanium dioxide
content from silicon dioxide content for Mars soil
is in the interval 0.0714 to 0.1251.
Because the confidence interval for the soil
samples entirely overlaps the one for the rocks,
you cant conclude that the slopes are different.
= 0.0233
E11. a.
b. Yes. With t = 2.98 and associated P-value
of 0.0106, reject the H0 at the 5% level of
significance. Conclude that the slope is
significantly different from zero. Note,
however, the slight curvature may mean that
a linear model is inappropriate.
c. The 90% confidence interval estimate based
on 13 degrees of freedom is (0.1107,
0.0281). This interval gives the plausible
values of the expected decrease in miles per
gallon per 1 unit increase in horsepower.
E13. Conditions are met, except that the variation in
y tends to increase with x. With t = 1.40 and
13 degrees of freedom, the P-value is 0.185. Do
not reject H0; there is no evidence of a linear
association between expected gas mileage and
the maximum speed of the engine in rpms.
E15. a. i. You are 95% confident that the true slope
of the regression line for predicting the
22-03-2009 21:30
Lesson
39 de 44
temperature from the chirp rate is in the

interval 1.99 to 4.59.
ii. You are 95% confident that the true slope
of the regression line for predicting chirp
rate from temperature is in the interval
0.128 to 0.296.
b. When you reverse the roles of chirp rate
and temperature, the entire regression line
changes. The unit of the slope changes from
degrees per chirp to chirps per degree.
E17. To test that
= 1, t = 1.454. With 13 degrees
of freedom, the P-value is 2(0.08485) 0.1697.
Do not reject the hypothesis that the slope is 1.
Alternatively, a 95% confidence interval for the
slope is 0.9517 0.0718, which includes 1 as a
plausible value for .
relationship between length and weight, you

might choose the log-log transformation.
b. Conditions are met, assuming this is a
random sample. t = 4.3230. With 7 degrees
of freedom and a two-tailed test, the P-value
is 0.0035. Reject the H0 that the slope of the
true regression line for ln(weight) versus
ln(length) is equal to 3.
c. You are 95% confident that the slope of
the true regression line for (ln(length),
ln(weight)) would be in the interval (2.244,
2.779). The 95% confidence interval for
widemouth bass was (3.301, 3.460). Because
there is no overlap in these confidence
intervals, you can conclude that the power
function that relates length to weight of a
black crappie has a smaller exponent than
that for widemouth bass.
E19. size = 31.83 0.712 acid. For every increase of

1 g/ml in acid concentration, the radius of the
P19. a. The distribution of log(img) should be
fungus colony tends to decrease by 0.712 mm, on
approximately normal with mean around
average.
1.654 and standard deviation around
Assuming the treatments were randomly assigned,
s = 0.2588. About 45.082.
the conditions are met. With t = 19.84 and
b.
You are 95% confident that in the population
10 degrees of freedom, the P-value is essentially 0.
of all countries, the slope of the regression
Reject H0 :
= 0. There is a negative linear
line for (fr, log(img)) is (0.1423, 0.2477). This
relationship between colony radius and acid
indicates that the exponential function of the
concentration.
form y = abx that best fits the relationship
E21. price = 32.3263 + 39.4553 bedrooms
between fr and img has plausible values of b
The statistically significant slope indicates that
between 10 (0.1423) and 10 (0.2477), or 1.388 to
the expected increase in price is about $39,455
1.769. Thus, the img increases from about 39%
per additional bedroom. But there is only one
to 77% for every one-unit increase in fr.
two-bedroom house and there are three
P20.
a.
A linear regression model will not work well
four-bedroom houses. To get a fairer picture
here as the scatterplot and the residual plot
of the true relationship between average selling
both show curvature.
price and number of bedrooms, more two- and
b. Squaring the percent successful straightens
four-bedroom houses should be sampled.
the pattern rather well (a power model does
almost as well): percent successful2 =
Section 11.3
51.40 + 426.632 number of chimps.
P17. Conditions are met. t = 10.7793. With 9 degrees
c. You are 95% confident that the true slope of
the linear relationship between the square
of freedom, the P-value is 1.9 10 6, which is
of the percent of success and the number
very close to 0. Reject the H0 that the slope is 3.
of chimps in the hunting party is between
P18. a. The untransformed data clearly show
366.34 and 486.92.
curvature. Both the log transformation and
E23. a. Yes. The residuals are more evenly scattered.
the log-log transformation straighten the
The slope of the regression line went
pattern considerably. It is difficult to choose
from negative to positive, but is still not
between them on the basis of the residual
significantly different from 0.
plots or dot plots of the residuals. However,
because there tends to be a close-to-cubic
A log-log transformation allows the residuals
to look more randomly scattered and their
22-03-2009 21:30
Lesson
40 de 44
distribution to be more symmetric, but the

slope is not significantly different from 0.
b. The regression line allows you to explore
the relationship between crime rate and
population. A significance test allows
conclusions such as the following: The slope
of the relationship between the crime rate
and total population of the city is no different
from what you would expect if the crime rate
had been assigned at random to the cities.
E25. a. The plot of number of police versus violent
crime rate shows a bit of upward curvature
and shows that variation in y is increasing
with increasing x. The scatterplot of ln(police)
versus (violent crime rate) shows a much
more linear trend. The regression equation is
ln(police) = 1.42 + 0.00334(violent
crime rate).
b. Exponentiating both endpoints of the interval
(0.00126, 0.00542) gives an interval estimate
of (1.0013, 1.0054) for the base of the
corresponding exponential function. Thus,
if City A has a crime rate 1 unit greater than
City B, the number of police officers in City
A is expected to be between 0.13% and 0.54%
greater than the number in City B.
E27. Using a log-log transformation to reduce
curvature, the regression line is ln(flow) =
1.01 + 1.12 ln(static). The slope is significantly
different from 0 with a P-value below 0.0001.
E29. a. No. A log-log transformation will reduce the
curvature.
b. The fitted equation is ln(distance) =
3.7514 + 0.5152 ln(height).
Conditions for inference are not met, as this
is not a randomized experiment, but the
confidence interval is still useful.
You are 95% confident that, for every unit
increase in the natural log of the height, the
natural log of the distance will increase, on
average, by an amount between 0.4681 and
0.5623.
c. These data show much more curvature,
which is not eliminated by the log-log
transformation. A more complex model is
needed. The shelf, by eliminating the initial
downward velocity, clearly made a difference
in the relationship between distance and

height.
E31. a. This is a random sample of countries. The
trend is fairly linear, and the variation in
residuals is quite uniform. The distribution
of the residuals looks quite symmetric with
three outliers. The outliers are a reason for
some caution.
b. Reject the H0 that these data came from
a population where the slope of the true
regression line for predicting fertility rate
from life expectancy for women is 0.
c. No. This was not an experiment where
conditions or treatments were randomly
assigned.
d. The large area of white space in the middle
of the graph suggests two clusters. Each of
these clusters taken separately may tell a very
different story.
Chapter Summary
E33. a. I, III, IV, and V, although IV shows no linear
trend and V has an influential data point
b. IV
c. V, but III is close
d. II: try a log-log transformation. V: remove the
influential data point.
E35. a. If one pizza has 1 g of fat less than another,
you expect to have 4.996 to 9.530 fewer
calories.
b. (24.98, 47.65)
E37. a. Conditions for inference are met. The
regression equation is length =
1037.1 + 0.5771 year.
You are 95% confident that the slope of the
true linear relationship between length of
a film and the year in which it was released
is between 0.1048 and 1.0494. Because 0 is
not in the interval, you have statistically
significant evidence that there is a positive
linear relationship.
b. Conditions for inference are met. The
regression equation is rating =
0.97 + 0.01365 length.
true linear relationship between a films
22-03-2009 21:30
Lesson
41 de 44
rating and its length is between 0.00121 and

0.02609. Because this interval lies entirely
above 0, you have statistically significant
evidence of a positive linear relationship.
c. You would expect a positive linear relationship
between year and rating, but the trend appears
negative.
Conditions for inference are met. The
regression equation is rating =
21.1 0.00959 year.
For testing the H0 of zero slope, t = 1.3381
with 18 degrees of freedom, giving a P-value
of 0.1975. Do not reject the H0 of zero slope.
You do not have statistically significant
evidence that the slope of the relationship
between rating and year for all movies is
different from 0.
d. For the longer movies, conditions for inference
are met. The regression equation is rating =
31.3 0.01458 year.
= 0.100. This is stronger evidence of a
negative linear relationship between a movies
rating and the year in which it was released
than when short and long movies were
considered together.
For the shorter movies, conditions for
inference are met. The regression equation
is rating = 59.7 0.02954 year.
= 0.016. This is much stronger evidence of a
negative linear relationship between a movies
rating and the year in which it was released
than when short and long movies were
considered together.
Ratings tend to go down over the years, but
this overall decreasing trend gets masked by
the fact that more of the longer movies, with
higher ratings, are from recent years.
E39. If standard deviation of the responses is
interpreted as meaning for each fixed x, then
both are descriptions of and so are equal.
If standard deviation of the responses is
interpreted to mean the SD of all values of y,
then that value will be larger than the SD of the
prediction errors.
Chapter 12
Section 12.1
P1. most effective: Treatments 1 and 5, because they
are centered at the lowest values; least effective:
Treatments 3 and 4
P2. The centers vary widely, as do the spreads. As the
mean increases, the spreads tend to decrease.
P3. The treatment with the larger mean tends to have
an SD that is 0.16 cm smaller.
P4. Conditions are met. With t = 0.702 and P-value
= 0.4914, dont reject H0 . There isnt statistically
significant evidence that the treatment means
differ.
P5. The boxplots of Treatments 4 and 5 do not overlap;
they appear to have widely differing means.
(18.4, 30.2) does not include 0, so reject H0 that
the mean growth would have been the same if all
plants could have been given both treatments.
P6. Outliers for Treatment 4: 1 and 4; outlier for
Treatment 5: 5; Treatment 5 is skewed right,
whereas Treatment 4 is fairly symmetric. Plants
given Treatment 5 were slightly taller and more
variable in height to begin with, but there isnt
enough difference to question the conclusion
in P5.
Section 12.2
Conditions are met. With z = 0.7303, the
one-sided P-value is 0.2326. Do not reject the
H0 that the proportion of Florida households in
which the primary language is not English is 0.25
in favor of the alternative that it is less.
P8. Conditions are met. You are 95% confident that
the proportion of all Florida households residing
in an apartment or other attached structure is in
the interval 0.30 0.14, or (0.16, 0.44).
P9. Conditions are nearly met. Any population
proportion in the interval 0.225 0.129,
or (0.096, 0.354), could produce the sample
proportion as a reasonably likely outcome.
P10. For those who own their home with a mortgage
and those who own their home outright, there is
no clear association with building type. Renters
appear more likely to live in an apartment or
other attached structure.
P7.
879
22-03-2009 21:30
Lesson
42 de 44
P11. Conditions are not quite met, as one of the four

cells has an expected frequency less than 5. With
2
= 0.2564 or z = 0.5064, the P-value = 0.6126.
Dont reject H0 that the proportion of households
with older residents is the same for both language
categories.
P12. Conditions are not met, as three of the cells have
expected frequencies below 5. With 2 = 6.8899
and P-value = 0.0319 (perhaps a little larger
because of failed conditions), reject the H0 of
no association. There is a statistically significant
association between the type of building in which
the household resides and the primary language
spoken.
P13. a. The BLD values were randomly paired
with the values of HHL. The frequency of
each type of pair was recorded, expected
frequencies calculated, and 2 statistic
computed. This was repeated 200 times.
b. The observed test statistic 6.8899 was met
or exceeded 7 times out of 200 runs, making
the approximate P-value 0.035.
P14. Conditions are met. You are 90% confident that
the mean number of people per household in
Florida is in the interval 2.175 0.301, or
(1.874, 2.476).
P15. a. There is wide spread but the distribution is
symmetric with no outliers, so with a sample
size of 24, the standard method based on
the t-distribution should work without a
transformation.
b. With t = 0.8105 and P-value = 0.213, do
not reject H0 . You do not have statistically
significant evidence that the mean mortgage
payment in Florida is less than $840.
P16. a. The distribution is highly skewed with three
outliers. A log transformation works well.
b. Conditions are met reasonably well for the
transformed data. You are 95% confident
that the mean of the natural log of the cost of
hazard insurance, for all households that pay
it in Florida, is in the interval 6.510 0.334,
or (6.176, 6.844).
P17. Conditions are met, except that rooms for
apartments is a bit skewed. You are 95%
confident that the difference in the mean number
of rooms in detached houses and the mean
number of rooms in apartments is in the interval
1.82 0.97, or about (0.85, 2.79).
P18. If one house has one more room than another

house, you would expect an increase in the
electricity bill of about $17.10 per month.
P19. Yes. Conditions are met reasonably well. With
t = 3.120, the (one-sided) P-value is about
0.0017. A one-sided test can be justified because
additional rooms should raise the electric bill.
P20. The scatterplot shows a statistically significant
linear trend (P-value = 0.00076), but the
correlation is not high. This is strong evidence of
a fairly weak association.
P21. a. Both scatterplots indicate a weak positive
trend, but the number of rooms has slightly
higher correlation with the number of vehicles.
b. Conditions are met for both tests.
People: With t = 1.97 and P-value = 0.0564,
the observed slope is not quite significantly
different from 0 at the 5% level.
Rooms: With t = 2.85 and P-value = 0.0069,
the observed slope is significantly different
from 0 at the 5% level. Number of rooms is
the better predictor. This could be because
wealth (a lurking variable) is more closely
related to number of rooms than number of
people in a household.
P22. a. INSP = 618.41 + 289.662 RMS. For every
additional room in the house, the annual
insurance payment tends to increase by
about $290.
b. The main problem is the lack of homogeneity.
c. A log transformation on INSP fixes this
quite nicely. ln(INSP) = 4.68 + 0.322 RMS.
linear relationship between the natural log
of a households annual hazard insurance
payment and the number of rooms in the
household for all Florida households is in the
interval 0.322 0.1933, or (0.129, 0.515).
Section 12.3
P23. a. A statistically significant positive linear
relationship (t = 4.56; df = 28; P-value <
0.0001). On the average, there is an $22,700
increase in payroll for each increase of 1,000
in attendance.
b. The American League mirrors essentially the
same pattern as both leagues together, with
a statistically significant (t = 3.313, df = 12,
22-03-2009 21:30
Lesson
43 de 44
P-value = 0.0062) increasing trend of about

$26,700 in increased payroll for each increase
of 1,000 people in attendance.
c. The scatterplot indicates serious curvature,
which isnt straightened by our usual
transformations. A more complex model is
needed.
P24. For all three groups of teams, there appears to
be little association between payroll and batting
average. P-values are
a. 0.3378
b. 0.4760
c. 0.6208
P25. a. A statistically significant positive linear
relationship; if the two teams with unusually
high batting averages (Colorado and
Seattle) are removed, the estimate of the
slope increases from 3.72 to 4.5 tenths of a
percentage point average gain in percent wins
per 1-point increase in team batting average
and the P-value remains about the same.
b. The average increase in percent wins (in
tenths) is 7.2 per 1-point increase in batting
average. However, if Seattle is dropped from
the analysis, the estimate of the slope, 5.3,
is not quite statistically significant at P-value
= 0.071.
c. With Colorado, there is no significant linear
relationship. Without Colorado, the slope
jumps to an estimated 5.8 increase in percent
wins (in tenths) for each 1-point increase in
batting average and the P-value drops to 0.006.
P26. a. No; t = 0.5263 and P-value = 0.6. The
difference in mean attendance is not
b. No; t = 1.447 and P-value = 0.16. The
difference in the mean batting average is not
P27. No; z 0.4503 and P-value 0.6525. A
difference this small in winning percentages can
reasonably be attributed to chance.
P28. Use a chi-square test of homogeneity. The
outcomes are not strictly independent; however,
of the 162 games played, only a small number are
played among each other, so the dependence is
not too serious.
a. No; 2 0.351 with df = 2, P-value 0.84.
The observed differences could very well be
produced by chance alone.
b. Yes; 2 9.203 with df = 2, P-value 0.01.

The differences in the percentages of wins
cannot reasonably be attributed to chance
alone.
Section 12.4
P29. Youth Enterprises; the variance of the difference
is minimized when sample sizes are equal.
P30. a. Min = 22, Q1 = 37, Median = 53, Q3 = 56,
Max = 69
b. Boxplot shows that there are no outliers.
Stemplot shows the distribution is bimodal,
with one peak in the low 30s and a second,
higher peak in the upper 50s. Both show
skewness.
P31. 72%; 56%
P32.
P33. With z = 1.9055 the one-sided P-value = 0.0284.

Reject the H0 that the difference in the proportion
of workers laid off who were under 50 and the
proportion of workers laid off who were 50 or older
is no larger than you would expect if Westvaco
were picking 28 people at random for layoff.
P34. Using 50 as a cutoff leads to much stronger
evidence in support of a claim of discrimination.
P35. a. Neither test so far meets the criterion
established by the Supreme Court, although
the second test (in P33) comes very close
with a one-sided P-value of 0.0284.
b. I
c. a Type I error, if the null hypothesis is true
d. A Type II error would be deciding that a
company does not discriminate on the basis
of age when, in fact, it does.
P36. The distributions are somewhat skewed toward
the smaller ages. The average age of those
terminated is higher than the average age of
those retained. The spread of the ages of those
terminated is slightly greater than for those
retained.
22-03-2009 21:30
Lesson
44 de 44
P37. Here are the five-number summaries needed to

make the boxplots.
P38. With t = 1.066 and P-value = 0.146, there is

insufficient evidence to reject H0 . If 28 workers
were selected totally at random to be laid off,
then it is reasonably likely to get a difference
of 3.68 years or more in the average age of
employees laid off minus the average age of the
employees retained.
P39. a. The hypotheses are the same. In both cases
the null hypothesis is that the actual result
can reasonably be attributed to chance, and
the alternative hypothesis is that it cannot.
b. For a t-test, you must check that both samples

look like they could reasonably have come
from a normally distributed population. This
is not necessary with a randomization test.
c. For a t-test, if the conditions are met,
the sampling distribution should have a
t-distribution. The sampling distribution for
a randomization test is difficult to predict,
but often looks somewhat normal.
d. If the conditions for a t-test were met, the
two tests should reach the same conclusion.
The P-values would be very close. If
the conditions have not been met, the
randomization test is more reliable.
22-03-2009 21:30
Lesson
1 de 11
Index
A
absolute value, spread and, 65
Addition Rule, 319320
for Disjoint Events, 316318
for random variables, 372375
adjacent values, 62
Adrain, Robert, 123
age-neutral simulation, 1315
dAlembert, Jean, 291292
alternate hypothesis
for difference of two means, 625, 647
for difference of two proportions,
530, 532, 544
for experiments, 544
forms of, 498
for means, 589
one-sided, 594595
for one-sided tests, 507, 508
for surveys, 530, 532
See also hypotheses; null hypothesis
American Community Survey (ACS)
case study, 803808
anecdotal evidence, 249250
Anscombe data, 166167
approximate (simulated) sampling
distribution, 410411, 417,
428429
area, standard deviation and, 32
associated variables, 711, 722
association. See strength of association
average. See mean
averages, line of. See least squares
regression line
averages, point of, 125
incorrect response or measurement,

226227
judgment sample, 224
of method vs. sample, 222223
nonresponse bias, 225
point estimators and, 415, 416
questionnaire bias, 225226
response bias, 225227
sample selection bias, 223225
size bias, 223
voluntary response bias, 224, 246
See also confounding; random
sampling
bimodal distributions, 3536, 37
binomial distributions, 382390
calculator technique, 385, 389
defined, 385
expected value of, 387388
independence and, 383, 385, 386
shape of, 388
standard deviation of, 388389
bins, 44
bivariate categorical data, 805807
bivariate quantitative data, 107110
heteroscedasticity and, 774775, 778
inference with, 807808
outliers and, 107, 108, 162
regression and, 739
blind experiment, 251
blocking, 269, 273
blocks, 267, 268270, 273, 274
box-and-whiskers plots. See boxplots
boxplots, 61, 6263
calculator technique, 63
back-to-back graphs, 48, 55

balance point. See mean
balanced design, 266
bar graphs, 5051
histograms compared to, 50
stacked or segmented, 693, 713714
Barry, Dave, 226, 348349
bars
of bar graphs, 50
of histograms, 4446
baseball salaries case study, 810816
Bernoulli, Jacob, 293
bias, 220227
convenience sample, 224
defined, 222
frame of sample and, 224225
Calculator Notes references

binomial distributions, 385386, 389
central intervals for normal
distributions, 92
chi-square probability density
function, 678
chi-square test of homogeneity, 699
conditions for significance test, 757
confidence interval for a mean, 571
confidence intervals, 476, 477
correlation, 141
critical values, 497
cumulative relative frequency plots,
79
deviation, random selection of, 740
dot plots, 4
frequency tables, 68, 290

geometric distributions, 394, 396
graphing by specifying mean and
standard deviation, 33
least squares regression line, 124,
126, 131, 141
normal distributions, 33, 92
numbers of combinations, 20
P-values, 496, 585, 680, 718
pairs of random integers, generating,
305
predicted values, 122
probability distribution, 367
random digit string, generating,
301, 449
random number distribution, 30
random number generation, 234,
305, 563
residuals, 122, 131
sampling distributions, 412
scatterplot, 107, 121, 131
simulation, 14, 306
spinning coins, 293
sum of the squared errors (SSE), 124
t-distributions, 584
t*-values, 566
test statistic, 718
transformations, 192, 605
two-sample t-interval, 622, 626
unit changing, effects of, 76
z-scores, 84, 85, 89
calculators
histograms and, 45, 46
least squares lines on, 127128
percentiles on, 85
standard deviation and, 65
standard normal distribution,
proportion of values in, 84
summary statistics and, 66
See also Calculator Notes references;
software, statistical
capture rate, 477479
for confidence intervals for the
mean, 571572, 604
defened, 477
estimating the standard deviation
and, 563
skewness and, 604
case studies
American Community Survey
(ACS), 803808
baseball salaries, 810816
833
22-03-2009 21:31
Lesson
2 de 11
flower production, 800803

Westvaco case, 38, 1116, 817822
cases
of bivariate data, 107
defined, 4
categorical data, 692693, 805807
categorical variables
defined, 44
graphs of. See bar graphs
inference involving. See chi-square;
proportion
strength of association of, 722723
cause and effect
correlation not implying, 147
experiments establishing. See
experiments
lurking variables and, 147, 243244
observational studies and, 247, 279
Celsius, conversion to, 76
census, 219, 358
center
defined, 27
of normal distributions, 31
recentering. See recentering
skewed distributions and, 34
center, measures of
computation of, 5658
mean. See mean; mean of a
population; sample mean
median. See median
midrange, 73, 421
central intervals, 9092
Central Limit Theorem, 430431
chance. See random sampling
chi-square
critical values, table of, 827
degrees of freedom and, 679680,
695696, 703705
distribution of, 677678, 684
expected frequency of, 684,
694695, 715
goodness-of-fit test, 674, 680683,
684686
history of, 674
homogeneity test. See homogeneity,
chi-square test of
independence. See independence,
chi-square test of
sample size and, 678, 722723
stacked bar graph, 693, 713714
, 675
symbol
table of values for, 679680, 829
test statistic, 674675, 681
two-way table and, 692693, 715
z-test compared to, 685, 702705
See also significance tests
884 Index
clinical trials, 247, 251, 542543,

545546
See also experiments
cluster samples, 237238, 240
clusters, 3637, 107
coefficient of correlation. See
correlation
coeffcient of determination, 148151
column chart, 714
column marginal frequency, 713
common ratio, 401
comparative studies. See experiments;
observational studies
comparing, as characteristic of
experiment design, 252
comparison group, 250251
complement, 288
complete list, 291
completely randomized design, 264,
265266, 274, 643
conditional distribution of y given x,
741
conditional probability, 325333
definition of, 330
inference and, 332333
medical tests and, 330331
Multiplication Rule and, 327329,
330
from the sample space, 326
symbol for, 326
conditional relative frequencies, 693,
713
conditions assigned. See factors;
treatments
conditions of significance tests
checking, 500
531532, 544
fixed-level test for mean, 589
goodness-of-fit, 681, 682
homogeneity, 697, 699
independence, 716, 718
for means, 589, 625
for one-sided tests, 508
for proportions, 498, 500, 509
for slope, 757, 758, 759760, 765
t-test, 820821
confidence intervals, 470473,
482483
95% confidence interval, 473, 572
capture rate, 571572, 604
chart for reasonably likely events,
470472
conditions required for, 476
correct language use of, 479

for difference in two proportions,
518523, 541543
for difference of two means,
618623, 629
for difference of two means from
paired comparisons, 645646
error sources, 430
estimation of standard deviation
and, 563568
formulas for, 473477, 519520
margin of error. See margin of error
for mean of population, 562573
for mean of samples. See sample
means, confidence intervals for
for observational studies, 547548
outliers and, 604609
for proportions, 429430, 475477
skewness and, 602609
for slope, 762763, 765
for surveys, 518523
confidence level, t-values and, 566
confounding
in comparative studies, 245247,
248249
See also bias; random sampling
constant strength, 107
constant values, 76
control group, 250251
convenience sample, defined, 224
correlation, ecological, 815816
correlation (r), 140151
and appropriateness of linear model,
144145
cause and effect not implied by, 147
and coefficient of variation (r2),
148151
estimating, 140141
formula for, 142144
interpretation of, 148151
negative, 140
positive, 140
slope, relationship with, 146, 152
steps for determining, 107110
strength of, 140
as term, 140
transformations vs. use of, 779
units absent from, 143
See also strength of association
count. See frequency, histograms and
Counting, Fundamental Principle of,
294295
critical values
chi-square distribution, table of,
679680, 829
22-03-2009 21:31
Lesson
3 de 11
of confidence interval for a mean,

475476
defined, 496
and level of significance, 496497
t-distribution, table of, 828
test statistic and, 496497
See also t*-values
cumulative percentage plots, 7879
cumulative probabilities, 385386, 396
cumulative relative frequency plots,
7879
D
data
bivariate. See bivariate categorical
data; bivariate quantitative data
categorical, 692693, 805807
elliptical, 154
independence and. See
independence, chi-square test of
independent events with, 343345
inference requirements and, 651,
652, 664
probability distributions from,
358360, 361
production of, 640
quantitative (measurement), 807808
raw, 27
summaries. See summary statistics
symmetry and, 292
time-series, 765
univariate, 805807
See also paired comparisons
decay, exponential, 179180
degrees of freedom
chi-square and, 679680, 695696,
703705
difference of two means and, 621
slope estimation and, 756, 764765
t-values and, 566, 584, 621
dependent events
defined, 339
matched pairs design and, 644
time as, 765
See also independent events
dependent variables, 711
design. See experiments and
experiment design
determination, coefficient of, 148151
deviations, 6465
See also standard deviation
diagnostics
outliers and, 162167, 172
residual plots, 168, 172
vs. summaries, 162, 167, 172
diagrams, tree, 294295, 327328
difference of two means
confidence intervals and, 618623,
629, 645646
increasing power in, 629630

significance testing and, 624628,
629630, 646647, 820821
difference of two proportions
confidence interval for, 518523,
541543
sampling distribution of, 527529,
539541
significance test for, 526535,
544546, 819
difference, sampling distribution of,
445, 527529
discrete distributions, chi-square, 684
disjoint events (mutually exclusive
events), 315316
Addition Rule and, 316318
defined, 291
property of, 317
distributions
centers of. See center, measures of
chi-square. See chi-square
conditional, of y given x, 741
defined, 4
modes of, 3536
probability. See probability
distributions
rescaling. See rescaling
sampling. See sampling distributions
shapes of. See bimodal distributions;
normal distributions; skewed
distributions; uniform
(rectangular) distributions
summaries of, 37
See also graphs and graphing
dot plots, 44
defined, 4
mean, estimation of, 57
rounding rules and, 44
software for statistics and, 44
usefulness of, 44
double-blind experiment, 251
E
ecological correlations, 815816
ecological fallacy, 816
economics, sampling and, 219, 237,
480481
educational research, units and, 252
elliptical data, 154
errors
degrees of freedom and, 703705
hypotheses and, 501506, 593594
measurement error problems, 592
prediction, 120
sum of squared, 123125, 128
See also margin of error; standard
error; standard error of the mean
estimation
of mean, 57
of median, 34
of parameter, 220
point estimators, 415417
of slope, 756, 764765
of standard deviation, 32, 563568
units of measurement and, 50
ethics, observational studies and, 547
even numbers, and quartiles, finding,
59
events
defined, 288
dependent. See dependent events
disjoint. See disjoint events
independent. See independent
events
rare, 412, 468, 593, 761
See also reasonably likely events;
sample space
exact sampling distributions, 413414
expected frequency, of chi-square tests,
684, 694695, 715
expected value
of binomial distributions, 387388
defined, 365
in everyday situations, 369370
formula for, 366
of geometric distributions, 397398
linear transformations on, 371375
symbol for, 365
experiments and experiment design
balanced, 266
blind, 251
cause inferred through, 243245,
256, 279
characteristics of, 252256
clinical trials, 247, 251
completely randomized design, 264,
265266, 274, 643
confidence intervals of, 541543
confounding in, 245247, 248249
control groups, 250251
double-blind, 251
evidence, strength of, 617618
flower production case study,
800803
lurking variables in, 243244
matched pairs design, 264265,
643644, 645
observational studies contrasted to,
247, 279
placebo effect, 250251
protocol, 270
random assignment, 539541
randomized block design, 268270,
274
Index 885
22-03-2009 21:31
Lesson
4 de 11
randomized comparative
experiment, 251
design, 264265, 266268
randomizing. See randomizing
repeated measures design, 265,
644645
replication, 252, 253
response in, 244
significance tests for, 544546
surveys contrasted to, 278279, 539
as term, 245
treatments. See treatments
units. See units, experimental
variability, management of, 262263,
269, 270274
See also observational studies;
random sampling; surveys
experts. See judgment sample
explanation of pattern, 107, 109
explanatory (predictor) variable, 120
exploration of data, defined, 3
exponential functions
growth and decay, 179186
and log transformations, 180187,
193
regression lines and, 774, 781
extrapolation, 120, 129
F
factors, 248
Fahrenheit, conversion to Celsius, 76
fallacy, ecological, 816
false negatives, 331
false positives, 331
first quartile. See lower quartile
Fisher, R. A., 15, 269
Fishers exact test, 819
five-number summary, 6061
fixed-level testing, 588593
ower production case study, 800803
frame, 224225
frequency tables
chi-squares and, 692693, 712713
summaries from, 6768
See also relative frequency
histograms; relative frequency
plots
Fundamental Principle of Counting,
294295
G
Galton, Francis, 152
gaps, 3637
Gauss, Carl Friedrich, 123
generalization of pattern, 107, 108
genes, 490
886 Index
geometric distributions, 393399

defined, 396
expected value of, 397398
formula for, 395396
memoryless property of, 404
shape of, 398
standard deviation of, 398
as waiting-time problem, 393
geometric sequence, 401
glossary, 829838
goodness-of-fit test, 674, 680683,
684686
graphs and graphing
bar graphs. See bar graphs
boxplots, 61, 6263
choice of, summarized, 5152
column chart, 714
cumulative frequency plots, 7879
dot plots. See dot plots
histograms. See histograms
independence tests and, 712714
questions raised by, 50
scatterplots. See scatterplots
stemplots. See stemplots
growth, exponential, 179, 182186
H
heteroscedasticity, 107, 774775, 778
histograms, 4447
back-to-back, 55
bar graphs compared to, 50
bars of, 4446
bins of, 44
mean, estimation of, 57
relative frequency, 4647
software for statistics and, 45, 46
usefulness of, 46
homogeneity, chi-square test of,
692705
categorical data with two variables
and, 692693
degrees of freedom, 696, 703705
expected frequencies, 694695
independence test vs., 720722
multiple z-tests vs. one chi-square
test, 702703
procedure for, 697702
test statistic, 695696, 697, 699
horizontal axis, 83
hypotheses
alternate. See alternate hypothesis
errors and, 501506, 593594
for goodness-of-fit test, 681, 682683
for homogeneity test, 697, 698699
for independence test, 716, 718
for means, 585587, 589, 625
null. See null hypothesis

for slope, 758, 760
for t-tests, 589
hypothesis tests. See significance tests
I
incorrect response bias, 226227
independence, chi-square test of,
711724
expected frequencies in, 715
graphical display of data, 712714
homogeneity test vs., 720722
procedure for, 716719
sample size and, 722723
strength of association and, 722723
independent, as term, 345
independent events, 339345
binomial distributions and, 383,
385, 386
confidence interval for difference of
means and, 621
data and, 343345
definition of, 339340, 345
geometric distributions and, 396
Multiplication Rule for, 341343
random variables, 372373
sampling with and without
replacement and, 345
variance and, 372373, 445
See also independence, chi-square
test of
inference, 217218
with categorical data, 805807
chi-squares and. See chi-square
conditional probability and, 332333
confidence intervals and. See
data requirements for, 651, 652, 664
defined, 3
with quantitative data, 807808
randomization and, 249
significance testing and. See
significance tests
simulation and, 1216
See also cause and effect;
experiments; observational
studies; random sampling
inflection points, 3132
informed consent, 542
interpolation, 120, 129
interquartile range (IQR), 5960, 6162
IQR. See interquartile range
J
joint frequency, 693, 713
judgment sample, 224
22-03-2009 21:31
Lesson
5 de 11
L
Law of Large Numbers, 293
least squares regression line, 123
calculator technique, 124, 126, 131,
141
computer output for, 127128
equation, 125
finding, 123124, 125126
as line of means, 152153
linear models and, 739741,
772773
point of averages of, 125
properties of, 125
slope of, 125, 126, 128
sum of squared errors (SSE),
123125, 128
as term, 152
transformations and, 189, 774779,
781
true regression line equation, 739
y-intercept of, 125, 126, 128, 739
See also slope
leaves, 48
Legendre, Adrien-Marie, 123
Leonardo da Vinci, 141
level of confidence, 566
level of significance, 496497, 500, 588
levels of factors, 248
line of averages. See least squares
regression line
line of means. See least squares
regression line
line of symmetry, 31, 32
linear equations
finding, from two points, 118119
slope-intercept form, 116119
transformations to find. See
transformations
linear models
checking fit of, 771773
correlation and appropriateness of,
144145
selection of, 188
true regression line and, 739741,
772773
linear transformations, 7576, 371375
linearity, bivariate data and, 107, 108
lines
least squares. See least squares
regression line
for prediction, 120122
regression. See least squares
regression line
as summaries, generally, 116119,
129130
Lister, Joseph, 707
log-log transformations, 187189,
774779
logarithmic transformations, 193

exponential functions and, 180187,
193
power transformations and, 13,
187189, 774779
See also exponential functions
logistic functions, 779
lower quartile
defined, 34
in five-number summary, 61
interquartile range and, 59
lurking variables, 107, 109, 147,
243244
M
margin of error
confidence interval of a mean and,
572573
defined, 476477
sample size and, 479481, 572
marginal frequencies, 693
marginal relative frequency, 693
matched pairs design, 264265,
643644, 645
maximum
adjustment of sample maximum,
426
range and, 61
Mayo, Charles, 246, 247
mean ( x )
computation of, 56
confidence intervals for. See under
mean of a population; sample
mean
defined, 32, 56
deviations from, 6465
estimation of, 57
of frequency tables, 6768
midrange and, 73
of a population. See mean of a
population
of a probability distribution. See
expected value
regression toward, 152153
of a sample. See sample mean
of a sampling distribution, 430, 435,
459
as sensitive to outliers, 79
significance tests for. See sample
means, significance tests and;
t-tests
standard error of. See standard error
of the mean
as summary statistic, choice of, 74
trimmed, 73
mean of a population ()
capture rate for, 571572
confidence intervals for, 562573
margin of error for, 572573
significance testing of, 580596. See
also ttests
symbol for, 430
test statistic for, 581582, 589, 625
means, line of. See least squares
regression line
measurement
inaccurate, bias and, 227
units of. See units of measurement
measurement data, 807808
measurement error problems, 592
measures of center. See center,
measures of
median
computation of, 5758
defined, 57
estimation of, 34
as resistant to outliers, 79
of skewed distribution, 34
spread around (interquartile range),
5960, 6162
medical experiments. See experiments
medical tests, 330331
midrange, 73, 421
minimum
range and, 61
mode
defined, 31
two or more, 3536
models. See linear models; probability
models
modified boxplot, 6263
Multiplication Rule, 327329, 330
for independent events, 341343
mutually exclusive outcomes. See
disjoint events
N
negative trend, 107
nonresponse bias, 225
normal distributions, 3033
binomial distributions and, 388
center of, 31
central intervals for, 9092
common sources of, 33
568, 569
defined, 31
difference of two means from paired
observations and, 645, 647
Index 887
22-03-2009 21:31
Lesson
6 de 11
infeection points of, 31, 32

mean of, 32
mode of, 31
rare events, 412, 468, 593, 761
reasonably likely values. See
reasonably likely events
sample size and use of. See sample
size
sampling distributions and. See
sampling distributions, normal
distributions and
shape of, 31, 83
significance test for a mean and, 589
spread of, 31
standard. See standard normal
distributions
summary of, 37, 74, 92
symmetry line of, 31, 32
t-distributions and, 584
table of probabilities, 826827
unknown percentage problems and,
84
unknown value problems and, 8485
usefulness of, 79
x-axis of, stretching, 83
not A(complement), 288
null hypothesis, 492
defined, 498
530, 532, 544, 545
errors and, 501506, 593594
for experiments, 544, 545
for means, 585587, 589, 625
P-value and, 494495, 585589
power of test and, 503506, 594,
629630
randomness not involved and, 554
rejection of, 587589, 593594, 761
rejection of, language for, 510
symbol for, 492
t-distributions and, 582583
See also hypotheses
O
observational studies
baseball salaries case study, 810816
confidence interval for, 547548
confounding in, 245247, 248249
defined, 246
ethics and, 547
experiments contrasted to, 247, 279
factors in, 248
inference and, 547, 812
levels of factors, 248
significance test for, 548549
surveys contrasted to, 279
888 Index
units and, 252254

See also experiments; surveys
observed significance level. See
P-values
odd numbers, and quartiles, Finding,
59
one-sided (one-tailed) tests of
significance, 507508, 594595
opinion polls
methods of, 468
questions to ask about, 482
See also surveys
or, as term, 315
outcomes
disjoint. See disjoint events
equally likely, 288
Fundamental Principle of Counting,
294295
list of. See sample space
See also events
outliers, 3637
bivariate data and, 107, 108, 162
boxplots and, 6263
confidence intervals of a mean and,
604609
defined, 36, 61
influence of, 77, 162167, 172
and interquartile range, value of,
6162
scatterplots and, 162167
P
P-values
calculator technique, 496, 585, 680
computing, 495496
defined, 494
for difference of two means, 625,
647, 650
531, 532, 544545
for homogeneity test, 696, 697698
interpretation of, 494495, 496
for means, 585587, 589590,
594595
for one-sided tests, 507, 508
one-tailed, 594595
simulation of, 814
slope and, 761
t-statistic and, 593594
two-tailed, 586
paired comparisons, 651652
chance mechanisms behind, 651,
652
confidence interval and, 644,
645646
designs of experiment and, 540545
meaningful comparisons, 650651

significance tests and, 646647
transforming non-normal data,
648649
parameter, 220
Pearson, Karl, 674
percentiles, 7879
permutation test, 821
placebo, 250251
placebo effect, 250
plausible percentages, 471472
plausible values, 562563
point of averages, 125
pooled estimate, 530, 629630
population
census of, 219, 358
defined, 219
frame contrasted to, 224225
mean of (). See mean of a
population
parameter of, 220
sampling of. See random sampling
size of (N). See population size
standard deviation of. See
population standard deviation ( )
unknown, 358
population pyramid, 55
population size (N)
defined, 219
replacement and, 326, 433
representativeness of sample and,
433434, 479481
sampling distribution of the mean
and, 434
symbol for, 430
population standard deviation ( )
defined, 412
estimating, 563568
reasonably likely events and, 412
symbol for, 430
See also standard deviation; standard
error
positive trend, 107
potentially influential outliers, 163167
power functions, equation of, 188
power of tests, 503506, 594, 629630
power transformations, 190192, 193
logarithms and, 187189, 774779
regression lines and, 774, 781
predicted ( response ) variable, 120
predicted value, 120, 130
prediction
error of, 120
extrapolation, 120, 129
interpolation, 120, 129
lines used for, 120122
22-03-2009 21:31
Lesson
7 de 11
prediction error, 120

predictor (explanatory) variable, 120
probability
Addition Rule. See Addition Rule
assignment of, 289291
complement event, 288
conditional. See conditional
probability
defined, 288
distributions of. See probability
distributions
estimated, simulation used for,
301309
events and. See events; independent
events
Law of Large Numbers and, 293
Multiplication Rule. See
Multiplication Rule
of number of successes, 446448
observed data and, 289
P-values as, 587
of proportions, 452453
for sample means, 431433
sample space. See sample space
for sample sums, 434436
sampling distinguished from, 358
significance test P-values. See
P-values
simulation used to estimate,
301309
subjective estimates of, 289290
symmetry and, 289, 290, 292
probability distributions, 357399
binomial distributions. See binomial
distributions
data as basis of, 358360, 361
defined, 289
geometric. See geometric
distributions
linear transformations of, 371375
mean of. See expected value
random variables, 361, 372373
standard deviation of, 364367,
371375
table of standard normal, 826827
theory as basis of, 361364
variance of, 365367, 371375
probability models
appropriateness of, 358
construction of, 288296
data and, 292, 343344
defined, 96
simulation and, 301309
symmetry and, 292
wrong model, 593, 761
probability sample, 231
proportion
chi-square tests of. See chi-square
confidence intervals and. See
difference of two. See difference of
two proportions
of samples. See sample proportion
significance tests and, 498500,
509510
significance tests and difference of,
529534, 544546, 819
test statistic for, 493494, 499500
test statistic for difference of, 531,
532, 544
protocol, 270
Q
quantitative data, 807808
quantitative variables
bivariate. See bivariate quantitative
data
defined, 44
graphs of. See dot plots; histograms;
scatterplots; stemplots
inference involving. See least squares
regression line
quartiles
defined, 34
nding of, 5960
interquartile range (IQR), 5960,
6162
lower, 34, 59, 61
odd number of cases, middle value
in, 59
percentiles, 7879
as resistant to outliers, 79
upper, 34, 59, 61
questionnaire bias, 225226
R
r. See correlation
random digit table, 301309, 359360,
828
random number generator, 30, 232, 234
random number table, 232
random sampling, 231240
bias as overcome by, 231
cluster sample, 237238, 240
568, 569
confounding as overcome by,
248249
control groups, 250251
independent selection and, 345
lack of, experiments and, 542543,
554
normal distribution and, 3233
placebo effect, 250251
random digit table, 301309,

359360, 828
random number generator or table,
30, 232, 234
replacement and. See replacement,
sampling with and without
simple random sample, 231232, 240
stratified random sample, 234237,
240
systematic sample with random
start, 238239, 240
two-stage sample, 238, 240
See also randomizing
random variables, 361, 372373
random variation, 739
randomization test, 821
randomized block design, 268270,
274
randomized comparative experiment,
251
randomized paired comparison design,
264265, 266268
randomizing
assignment in an experiment,
539541
bias avoidance and, 231
as characteristic of experiment
design, 252, 255, 278279
importance of, 248249
See also random sampling
range, defined, 61
rare events, 412, 468, 593, 761
raw data, 27
Rayleigh, Lord, 3637
reasonably likely events, 468469
chart for, 470472
defined, 412
recentering, 7576, 79
of probability distributions, 371375
standardizing with, 83, 8587
See also transformations
reciprocal transformations, 190,
605607
rectangular (uniform) distributions,
2830, 37
regression effect (regression toward the
mean), 152153
regression lines. See least squares
regression line
relative frequency histograms, 4647
relative frequency plots, 7879
repeated measures design, 265,
644645
replacement, sampling with and
without
independent events and, 345
Index 889
22-03-2009 21:31
Lesson
8 de 11
sample size vs. population size and,

326, 433
sampling distribution properties
and, 433, 444445
replication, 252, 253
rescaling, 7576, 79
of probability distributions, 371375
standardizing with, 83, 8587
See also transformations
residual plots, 167172
defined, 168
as diagnostic plot, 168, 172
time and, 184, 193
types of, 171172
residuals
defined, 120
finding, 120122
homogeneity of, in power model,
188
regression slope and, 772773
as sum of squared errors, 123125,
128
See also residual plots
resistance to outliers, 77
response, 244
response bias, 225227
response variable, 120
robustness, 602, 607
row marginal frequency, 713
S
sample mean ( x )
comparing, usefulness of, 650651
confidence intervals. See sample
means, confidence intervals and
importance of, 426
nding probabilities of, 431433
reasonably likely outcomes for, 412
sample proportion as form of, 450
sampling distribution of. See
sampling distribution of the
sample mean
significance tests for. See sample
means, significance tests and
symbol for, 430
use of, in text, 426
sample means, con dence intervals
and, 573
capture rate and, 571572
for difference between two means,
618623, 629
differences, paired comparisons,
645646
890 Index
estimation of standard deviation

and, 565568
sample size, 568, 569, 572
skewness and outliers, effect of,
602609
sample means, significance tests and,
580596
components of, 589590
for difference between two means,
624628, 629630
differences, based on paired
samples, 546547
fixed-level testing, 588593
one-tailed tests, 594595
P-value and, 585587, 594595
rejection of null hypothesis,
meaning of, 593594
sample proportion (p )
assigning variables, 521
mean and, 450
pooled estimate, 530, 629630
probabilities of, finding, 452453
reasonably likely, 468469
sampling distribution of, 448453
symbol for, 448
testing. See significance tests
sample selection bias, 223225
sample size
binomial distributions and, 386
Central Limit Theorem and, 431
chi-square and, 678, 722723
568, 569, 572
difference of two means and, 621, 647
economics and, 219, 480481
estimation formula, 480481
15/40 guideline for inferences,
607608
margin of error and, 479481, 572
power and Type II errors and,
503506, 594, 629
replacement and, 326, 433
replications as equivalent to, 252
representativeness of, population
size and, 433434, 479481
sample proportion and, 450
sampling distribution and, 564
sampling distribution of the sample
mean and, 433434, 437
sampling distribution of the sum of
a sample and, 435
significance test for the difference of
two proportions, 531
skewness and outliers and, 607608
standard error of the mean and, 431
stratification and, 236237

strength of association and, 722723
symbol for, 430
t-procedures and, 568, 607608, 629
sample space
data and symmetry and, 292
defined, 291
Fundamental Principle of Counting
and, 294295
samples
defined, 219
economics and, 219, 237, 480481
generalizing from, 217218
mean of. See sample mean
probability, 231
reasons for, 219, 227
size of. See sample size
standard deviation of. See standard
deviation (SD)
sums of, probabilities of, 434436
variation in, 491
See also bias; random sampling
sampling bias. See sample selection bias
sampling distinguished from
probability, 58
See also random sampling
sampling distributions, 410453,
458459
Central Limit Theorem, 430431
defined, 410
of a difference, 445
of a difference of two proportions
for a survey, 527529
of a difference of two proportions
for an experiment, 539541
exact, 413414
Fishers exact test, 819
listing samples, generation by, 417
mean of, 430, 435, 459
of number of successes, 446448,
453
of the sample mean. See sampling
distribution of the sample mean
of sample proportion, 448453
samples generated for. See random
sampling
shape, center, and spread, 411412
simulation, generation by, 410411,
417, 428429
skewed distributions and, 428, 430,
431, 433434, 564
standard deviation of. See standard
error
of the sum and difference, 445
usefulness of, 410, 459
sampling distributions, normal
distributions and, 411412
22-03-2009 21:31
Lesson
9 de 11
Central Limit Theorem and,

430431
determination of shape, 453
importance of, 426
number of successes and, 447
sample proportion and, 450
sample size and, 433434, 435, 437
sum and difference and, 445
sampling distribution of the sample
mean, 427, 436437
and probabilities for sample means,
431433
and probabilities for sample totals,
434436
properties of, 430431, 433434
shape, center, and spread of,
427431, 433434
simulation of, 428429
and sum of a sample, 434436
sampling frame, 224225
scale. See rescaling
scatterplots, 106110
outliers on, 162167
patterns in, steps for describing,
107110
residual plots. See residual plots
shape-changing transformations. See
transformations
screening tests, 330331
SD. See standard deviation
SE. See standard error
second quartile. See median
segmented or stacked bar graph, 693,
713714
sensitivity, 331
sensitivity to outliers, 77
shape, 27
of binomial distribution, 388
of bivariate data, 107, 108
See also bimodal distributions;
normal distributions; skewed
distributions; uniform
(rectangular) distributions
significance tests, 509510
components of, generally, 500,
509510
conditions for use. See conditions of
significance tests
defined, 490
for difference of two means,
624628, 629630, 820821
for difference of two means from
paired observations, 646647
529534, 544546, 819
errors, power of tests to avoid,

503506, 594, 629630
errors, types of, 501506
xed-level, 588593
formal language of, 498500
goodness-of-fit, 674, 680683,
684686
homogeneity test, 697702, 720722
hypotheses of. See hypotheses
independence test, 612616,
720722
informal, 491493
level of significance, 496497, 500,
588
for mean of population, 580596.
See also ttests
for mean of sample. See sample
means, significance tests and
for observational studies, 548549
one-sided (one-tailed), 507508,
594595
P-values. See P-values
power of tests, 503506, 594,
629630
for slope, 757761, 765
statistical significance, 490, 492, 501
for surveys, 529534
test statistic. See test statistic
See also confidence intervals
simple random sample (SRS), 231232,
240
See also replacement, sampling with
and without
simulation
age-neutral, 1315
defined, 13
key steps in, 16
probability distributions using,
358360, 361
probability estimation using,
301309
sampling distributions generated by,
410411, 417, 428429
Siri, W. E., 214
size
of population. See population size
of sample. See sample size
size bias, 223
skewed distributions, 3335
confidence intervals and, 564,
602609
defined, 33
geometric distributions and, 398
median of, 34
quartiles and, 34
sampling distributions and, 428,

430, 431, 433434, 564
skewed left, 33
skewed right, 33
summary of, 37, 74
transformations of. See
transformations
slope
confidence interval for, 762763,
765
correlation coefficient r and, 146,
152
defined, 116
degrees of freedom and, 756,
764765
of least squares regression line, 125,
126, 128
significance test for, 757761, 765
standard error of, 741748
symbols for (b 1, ), 125, 739
test statistic for, 755756, 758,
760761, 790
variability in estimated, 739749
slope-intercept form, 116119
software, statistical
dot plots, 44
histograms, 45, 46
least squares line, 127128
quartiles, finding, 59
residual plots, 171
stemplots, 49
See also Calculator Notes references;
calculators
specificity, 331
split stems, 48
spread
around mean. See standard
deviation
around median (interquartile
range), 5960
defined, 27
square root transformation, 649650
SRS (simple random sample), 231232,
240
SSE (sum of squared errors), 123125,
128
stacked or segmented bar graph, 693,
713714
standard deviation (SD)
of binomial distributions, 387388
computation of, 66
defined, 32
estimation of, 32, 563568
formula for, 65
Index 891
22-03-2009 21:31
Lesson
10 de 11
of frequency table, 6768

of geometric distributions, 398
model checking and, 772773
of populations. See population
standard deviation ( )
of probability distributions,
364367, 371375
of sample proportion, 450451
of sampling distributions. See
standard error (SE)
as sensitive to outliers, 79
as summary statistic, choice of, 74
symbols for, 430
variance and. See variance
standard error of the mean
defined, 430
sample size and, 431
symbol for ( ), 430
standard error (SE)
defined, 412
of difference of two proportions,
519520
of estimated slope, 741748
of the number of successes, 447
point estimators and, 415
of the sample proportion, 450
of sampling distributions, 435, 459
of the sum of a sample, 435
standard normal distributions, 8385
defined, 83
solving unknown percentage and
value problem, 8790
standardizing, 83, 8587
table of probabilities, 826827
unstandardizing, 86
See also z-score
standard normal probabilities, table of,
824825
standard units. See z-score
standardizing, 8587
statistic, as estimate from sample, 220
statistical significance, 490, 492, 501
stem-and-leaf plot. See stemplots
stemplots, 4750
back-to-back, 48
rounding rules and, 49
software for statistics and, 49
split stems, 48
truncation of values, 49
usefulness of, 49
stems, 48
strata, defined, 235
stratification, defined, 235
stratified random sample, 234237,
240
strength of association
correlation and. See correlation
defined, 722
892 Index
independence tests and, 722723

scatterplots and, 107, 108
strength of evidence, 617618
Strutt, John William, 3637
sum ( ), as symbol, 56
sum of squared errors (SSE), 123125,
128
summary statistics
bias and, 415, 416
of bivariate data, 107109, 116119
choice of, generally, 45, 28, 7475
defined, 12
diagnostics vs., 162
distribution of. See sampling
distributions
of distributions, 37
five-number, 6061, 63
outliers, influence of, 77
precise, 415
proportion. See sample proportion
recentering of, 7576, 79
rescaling of, 7576, 79
See also center, measures of;
diagnostics; spread
summary tables, 6
surveys
American Community Survey
(ACS) case study, 803808
bias and. See bias
causal inferences and, 278, 279
experiments contrasted to, 278279,
539
parameters and, 220
random sampling for. See random
sampling
samples used in, reasons for,
219220
significance tests for, 529534
See also experiments; observational
studies
symbols, glossary of, 831832
symmetry
defined, 33
line of, 31, 32
of normal distribution, 31
probability and, 289, 290, 292
systematic sample with random start,
238239, 240
t-distribution, table of critical values,

826
t-distributions, 582584, 826
t-procedures
pooled vs. unpooled, 629630
as robust, 602, 607
sample size and, 568, 607608, 629
skewness and outliers and, 602609

slope and, 756
transformations and, 605607
two samples vs. paired design,
640647
t-tables, 565566, 828
t-tests, 589593
conditions for, 589, 820821
paired data vs. single sample in,
640641
use of, 640
t*-values
confidence intervals based on,
565573, 618623, 762763
confidence level and, 566
defined, 565
degrees of freedom and, 566, 584,
621
outliers and, 604
P-values and, 593594
significance testing and, 624628,
760761
t-distributions, 582584, 828
t-tables, 565566, 828
z-values compared to, 583
tables
chi-square values, 679680, 829
frequency. See frequency tables
probability distribution, 365366,
375, 824825
random digits, 301309, 359360,
830
random number, 232
reading vertically, 4
summary, 6
t-tables, 565566, 828
two-way, 692693, 715
z-table, 8485, 826827
tail probability, 586
test statistic (t)
chi-square tests, 674675, 681
critical values and, 496497
defined, 493494
for difference of two means, 625,
647, 650
531, 532, 544
error types and, 501502, 593594
of experiments, 544
goodness-of-fit test, 681, 683
homogeneity test, 695696, 697, 699
for mean of a population, 581582,
589, 625
normal distribution of, 531
paired differences, 647
22-03-2009 21:31
Lesson
11 de 11

for slope, 755756, 758, 760761,
790
symbol for, 494
as t-statistic, 631
use of, 499
as z-score, 494
See also t-procedures; z-score
testing, samples and, 219
tests of significance. See significance
tests
theoretical probability, probability
distributions from, 361364
third quartile. See upper quartile
time
as dependent data, 765
and patterns of data, 179, 193
plots of, 822
transformations and, 184, 193
transformations, 179, 193
exponential functions and. See
exponential functions
heteroscedasticity and, 774775, 778
linear vs. nonlinear, 179
log-log, 187189, 774779
logarithmic. See logarithmic
transformations
of paired data, 648649
power. See power transformations
reciprocal transformations, 190,
605607
regression lines and, 189, 774779,
781
rescaling. See rescaling
of skewness and outliers, data with,
605607, 648649
square root transformations,
649650
treatments
balanced design and, 266
defined, 244
variability between vs. within,
262263, 270272
tree diagrams, 294295, 327328
trend, 107, 108
trimmed mean, 73
truthfulness, surveys and, 226227
two-stage sample, 238, 240
two-way tables, 692693, 715
Type I error, 501503, 506, 593
Type II error, 501506, 594
U
uniform (rectangular) distributions,
2830, 37
unimodal distributions, 35
units, experimental, 252255
blocks, 267, 268270, 273, 274
defined, 252
units of measurement
calculator technique exploring
changes in, 76
correlation as quantity without, 143
estimation and, 50
standardizing. See z-score
summary statistics and changes in,
7576
units of population, 219
univariate data
categorical, 805
measurement, 807
unknown percentage problems, 84
recentering and rescaling and, 83
solving, 8790
unknown values problems, 8485
recentering and rescaling and, 83
solving, 8790
upper quartile
defined, 34
V
variability
as challenge in data analysis, 4
linear models and, 741
management of, 262263, 269,
270274
point estimators and, 415, 417
slope and, 739749
variables
bivariate. See bivariate quantitative
data
categorical. See categorical variables
defined, 4
identification of, 107
lurking, 107, 109, 147, 243244
predictor (explanatory), 120
quantitative. See quantitative
variables
random, 361, 372373
response (predicted), 120
variance
defined, 65
independent selection and, 372373,

445
pooling and, 630
of a probability distribution,
365367, 371375
sampling distributions of the sum
and difference and, 445
variation
coefficient of determination and,
148151
random, 739
in sampling, 491
varying strength, 107
vertical axis, of histograms, 45
voluntary response bias, 224, 246
W
waiting-time problems, 393
weak association, 107, 108
See also correlation; strength of
association
Westvaco case study, 38, 1116,
817822
whiskers, 61, 62
X
x-axis, 83
Y
y-axis, histograms and, 45
y-intercept, symbols for (b0, ), 125,
739
See also slope-intercept form
Z
z-score, 83
calculator technique, 84, 85, 89
chi-square compared to, 685,
702705
comparisons using, 87
computation of, 86
confidence intervals and, 476477
correlation and, 142143
defined, 86
nding value with known, 86
significance testing and, 819
t-values compared to, 583
test statistic as, 494
unstandardizing, 86
using z-table, 8485
z*. See critical values
Index 893
22-03-2009 21:31
Photo Credits Page 894
1 de 1
http://acr.keypress.com/KeyPressPortalV3.0/ImportingCourses/SIA2/ph...
Photo Credits
Chapter 1
2: John Olson/Corbis. 4: Zigy Kaluzny/Getty Images. 10: Jeff Greenberg/The Image Works.
Chapter 2
26: Romeo Gacad/Stringer/Getty Images. 31: Cheryl Fenton. 34: Kennan Ward/Corbis. 36: Emilio Segr Visual Archives/
American Institute of Physics. 39: Steve Dipaola/Reuters/Corbis. 41: Louie Psihoyos/Corbis. 43: TIM FITZHARRIS/Minden
Pictures. 46: B.S.P.I./Corbis. 52: Frans Lanting/Corbis. 53: A. Ramey/PhotoEdit. 64: Cheryl Fenton. 82: Alan Schein
Photography/Corbis. 88: Royalty-Free/Corbis. 89: Anna Peisl/zefa/Corbis.
Chapter 3
104: Getty Images. 111: Getty Images. 112: David Cooper/Toronto Star/ZUMA/Corbis. 113: Royalty-Free/Corbis. 117: Cheryl
Fenton. 125: Frank Cezus/Getty Images. 131: Getty Images. 132: James A. Sugar/Corbis. 138: Charles ORear/Corbis. 149:
Cheryl Fenton. 152: Royalty-Free/Corbis. 157: Reuters/Corbis. 159: Scott Smith/Corbis. 160: Stephen Mallon/Getty Images.
163: Laura Murray. 173: Frank Trapper/Corbis. 175: Galen Rowell/Corbis. 180: Cheryl Fenton. 187: Tony Anderson/Getty
Images. 193: Alan Kearney/Getty Images. 194: Alan Schein Photography/Corbis. 195: Sean Garnsworthy/Getty Images.
Chapter 4
216: Fredrik D. Bodin/Stock Boston/PictureQuest. 219: Cheryl Fenton. 220: Tom McCarthy/PictureQuest. 224: CALVIN AND
HOBBES 1995 Watterson. Reprinted with permission of UNIVERSAL PRESS SYNDICATE. All rights reserved. 226: The
New Yorker Collection 1989 Mick Stevens from cartoonbank.com. All rights reserved. 227: CALVIN AND HOBBES 1995
Watterson. Reprinted with permission of UNIVERSAL PRESS SYNDICATE. All rights reserved. 232: Larry Gonick. 250: Scott
Adams/Dist. by United Feature Syndicate, Inc. 258: Cheryl Fenton. 260: Matt Perry. 261: Michael Melford/Getty Images.
264: Cheryl Fenton. 270: Cheryl Fenton. 271: Cheryl Fenton. 276: Vince Streano/Corbis. 277: David Young Wolff/Getty Images.
Chapter 5
286: Hulton Deutsch Collection/Corbis. 292: Cheryl Fenton. 295: Cheryl Fenton. 297: Leonard de Selva/Corbis. 299: Cheryl
Fenton. 318: Richard Cummins/ Corbis. 325: Matthew Polak/Corbis. 335: MLB Photos via Getty Images. 344: Cheryl Fenton.
Chapter 6
356: Paul A. Souders/Corbis. 366: Jose Luis Pelaez, Inc./Corbis. 386: Hulton Archive/Getty Images. 387: Cheryl Fenton.
392: Natalie Fobes/Corbis. 393: AFP/Corbis. 400: Pierre Vauthey/Corbis. 403: Antnio Rafael C. Paiva. 404: Tom Stewart/
Corbis.
Chapter 7
408: The Image Works. 416: NASA. 418: NASA. 420: Scott Smith/Corbis. 423: Richard Hamilton Smith/Corbis. 438: Bettmann/
Corbis. 439: Royalty-Free/Corbis. 448: Martyn Goddard/Corbis. 455: Brooks Kraft/Corbis. 456: Chuck Savage/Corbis.
460: Keren Su/ Corbis.
Chapter 8
466: Getty Images. 470 (top): Getty Images; (bottom) Cheryl Fenton. 473: Nancy Kaszerman/ZUMA/Corbis. 485: James W.
Porter/Corbis. 491: Cheryl Fenton. 494: From: Causal knowledge and imitation/emulation switching in chimpanzees (Pan troglodytes)
and children (Homo sapiens), Animal Cognition; Victoria Horner and Andrew Whiten, Springer Berlin/Heidelberg; Volume, 8,
Number 3/July 2005, pp. 164 181. 512: Michael Newman/PhotoEdit, Inc. 513: Eric OConnell/Getty Images. 517: Peter Horree/
Alamy. 525: Paul A. Souders/Corbis. 527: Cheryl Fenton. 538: Engel & Gielen/Getty Images. 553: Bettmann/Corbis.
Chapter 9
560: Robert Essel NYC/Corbis. 569: Colin Cuthbert/Photo Researchers, Inc. 591: Roger Ressmeyer/Corbis. 611: NASA.
614: Michael Newman/PhotoEdit, Inc. 617: Raymond Gehman/Corbis. 621: Jose Luis Pelaez, Inc./Corbis. 636: Dann Coffey/
Getty Images. 640: Cheryl Fenton. 658: Bonnie Kamen/PhotoEdit, Inc. 661: Bettmann/Corbis.
Chapter 10
632: Robert Maass/Corbis. 678: Andres Liivamagi/Alamy. 691: Connie Ricca/Corbis. 692: Cheryl Fenton. 698: Reuters/Corbis.
700: Getty Images. 708: Kate Powers/Getty Images. 717: Peter Turnley/Corbis.
Chapter 11
736: AP/Wide World Photos. 737: NASA. 745: Getty Images. 754: Alan Schein Photography/Corbis. 756: Peter Menzel/Stock Boston/
PictureQuest. 766: NASA. 769: Per Eriksson/Getty Images. 775: Steve Maslowski/Photo Researchers, Inc. 777: Liba Taylor/Corbis.
Chapter 12
798: Roger Ball/Corbis. 800: Michelle Garrett/Corbis. 810: MLB Photos via Getty Images. 812: Getty Images.
894 Photo Credits
22-03-2009 21:31

Statistics in Action

Uploaded by

Copyright:

Available Formats

Statistics in Action

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Statistics in Action

Uploaded by

Copyright:

Available Formats

cover

2008 Key Curriculum Press

Editorial Production Manager

Joshua Zucker, Castilleja School, Palo Alto, California

Textbook Product Manager

2008 by Key Curriculum Press. All rights reserved.

2008 Key Curriculum Press

About The Author

About the Authors

Florida, where he served as chairman of the Department of Statistics for 12 years.

About The Author

2008 Key Curriculum Press

A Note to Students from the Authors

Statistical Reasoning: Investigating a

1.1 Discrimination in the Workplace: Data Exploration

2.1 Visualizing Distributions: Shape, Center, and Spread

Relationships Between Two Quantitative Variables

Sample Surveys and Experiments

2008 Key Curriculum Press

5.1 Constructing Models of Random Behavior

6.1 Random Variables and Expected Value

7.1 Generating Sampling Distributions

Inference for Proportions

8.1 Estimating a Proportion with Confidence

Inference for Means

Inference for Regression