Lecture 8 T Test New

Medical Statistics
t - test
Jing Wang
aydgwwangjing@163.com
24/2/7 安医大公卫学院流统系王静制作
Review some key concepts of the last lessons
1. Statistical inference
draw conclusions of population from
sampled data.
Statistical Inference
Population Sample
Statistical Inference
Population characteristics Sample information

（ Parameter ） (Statistic)
2 Sampling error of mean
The difference between the sample mean and
population mean, or among different sample
means, caused by individual variation and
sampling.
It is expressed by two aspects:
① X ≠ 
② Xi ≠Xj
3 Standard error of mean, SEM
 S
X  SX 
n n
SEM reflects the variation among a group

of X from a same population (Xi ≠Xj ).
4 t-distribution
X  X 
t  ,   n 1
SX S n
It is a statistical distribution law of
sampling error of mean and a theory
foundation of statistical inference.
t-test
t-test is a type of hypothesis test based
on t-distribution.
Hypothesis test is one aspect of statistical inference.
What is the basic principle of hypothesis test?

How to perform it?
• Example1: Hemoglobin of 25 lead workers of a sample
were measured, the mean was 130g/L and standard
deviation is 25.6g/L. It is known that the average
hemoglobin of general adult is 140g/L.
n  25 ，X  130.0 g / L , s  25.6 g / L
 1  140 g/L
Question: if there had a difference on the

hemoglobin of lead workers and that of general
adults? (study objective) according to population
μ2 = μ1 μ2 ≠ μ1
X ≠ μ1
Population 1
Population 1 Population 2
Sample
μ1=140g/L μ1=140g/L μ2
n=25
x=130.0g/L
n=25
s=25.6 g/L
x=130.0g/L
s=25.6 g/L
due to sampling error due to different populations
Question: Which is the truth?
-- with the help of hypothesis test
Basic logic of a hypothesis test:

Proof by contradiction
Two probable conclusions are denoted as H0 and H1 (two
hypotheses) respectively.
H0 —— Null hypothesis
H0: μ2 = μ1= 140 ， X≠μ1 is due to sampling error
H1 —— Alternative hypothesis
H1: μ2 ≠μ1 ， X≠μ1 is due to the two different populations
Assume that the null hypothesis H0 holds;
Then to calculate a probability ( P -value)

P value: the possibility to occur the current sample
situation (such as t value) and even more unfavorable (or
extremely) situation under H0.
To obtain the P value is based on a statistical distribution law of

sampling error (such as “t-distribution”)
P<0.05
If it is less possible to occur the current sample
situation and even more unfavorable situation under
,then reject
H0 ; otherwise,
H 0 not reject . H0
P>0.05
(  is called the significance level of test, or testing criterion
 =0.05 )
small probability (event) principle
Three steps of a hypothesis test
Step1: Set hypotheses and the significance level of test
H0: μ2 = μ1= 140
H1: μ2 ≠μ1
 = 0.05.
Step2: Select an appropriate test method and calculate
the test statistic (such as t value)
It is a random quantitative variable (hemoglobin) with

the unit (g/L) of the measure value Xi
If Xi follows a normal distribution N (  ,  ) ,

2
X 
Then t ~ t distribution
S
n
Step3: Determine P value and make a conclusion
P-value in the hypothesis test: the probability of
the current sample situation (such as t value) and
even more unfavorable (or extremely) situation
under H0 .
.4
.3
=24

f(t)
.2
.1 P
P
0.0
-5.0 -4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0 5.0
Conclusion:
① conclusion in statistics: If the P ≤  (=0.05) , we could say
that the difference is statistically significant at level , then H0 is
rejected. Otherwise, H0 isn’t rejected. (the writing of these
conclusions is same for all the hypothesis tests)
② conclusion in major: to explain the content of H0 or H1 from

the viewpoint of major. (the writing of these conclusions is
different for different data)
Design types of t-test
One sample t- test
Paired / matched t-test
Two samples t-test
Conditions of t-test:
① Sample size n is small ( usually n<50 ).

② quantitative variable in the population follows a
normal distribution.
③ The variances are equal (  12   22)
One sample t- test
 Aim: comparison between one sample mean
(X) and a population mean ().
• Example1: Hemoglobin of 25 lead workers of a sample
were measured, the mean was 130g/L and standard
deviation is 25.6g/L. It is known that the average
hemoglobin of general adult is 140g/L.
n  25 ，X  130.0 g / L , s  25.6 g / L
 1  140 g/L
Question: if there had a difference on the hemoglobin

of lead workers and that of general adults?
Set hypotheses ：
H 0 :  2   1  140
H 1 :  2  140
  0.05
 If H 0 :  2   1  140 holds,
 Based on the current sample, the current sample

situation is t-value
130  140
t  1.953 ~ t distribution
25.6
25
  25  1  24 P/2=0.03129 P/2=0.03129
t
-1.953 1.953
Conclusion ：
Two-sided , P=0.06258 >0.05. On the level of  =
0.05 ， H0 isn’t rejected ； we could say that there
has no difference on hemoglobin between the lead
workers and general adults, the population mean of
hemoglobin of lead workers 2 = 140 g/L.
Paired / matched t-test
Two types of paired designs
① paired design on a same individual
Such as:
 Serum IgG measured before and after taking
medicine for measles→ if the medicine is effective
 fat contents of a beverage product measured by two
methods→ if the measurements of these two
methods are different
Example i Before After
(1) (2) (3)
1 1206.44 1678.44
 12 boys with measles
2 921.69 1293.36
treated by a medicine, 3 1294.08 1711.66
the IgG(mg/dl) was 4 945.36 1416.70
measured before and 5 721.36 1204.55
after the treatment as 6 692.32 1147.30
follows. 7 980.01 1379.59

8 691.01 1091.46
Is the medicine effective? 9 910.39 1360.34
10 568.56 1091.83
11 1105.52 1728.03
12 757.43 1398.86
Example
10 beverage products were randomly sampled, two methods were
used to determine the fat content. Has there difference
between the measurements of two methods?
② paired design on two different individuals
Such as:
 20 pigs are divided into 10 pairs matched on
birthweight, two pigs in each pair have same or
nearly same birthweight, then, two pigs in each
pair are all randomly treated by two treatments.
→ if these two treatments are different
（ detail data was omitted ）
Example 2 i Before After Difference di
(1) (2) (3) (4)=(3)-(2)
1 1206.44 1678.44 472.00
 12 boys with measles
2 921.69 1293.36 371.67
treated by a 3 1294.08 1711.66 417.58
medicine, the 4 945.36 1416.70 471.34
IgG(mg/dl) was 5 721.36 1204.55 483.19
measured before and 6 692.32 1147.30 454.97
after the treatment as 7 980.01 1379.59 399.58
8 691.01 1091.46 400.45
follows.
9 910.39 1360.34 449.95
Is the medicine effective? 10 568.56 1091.83 523.27
11 1105.52 1728.03 622.51
12 757.43 1398.86 641.44
Assume that if the medicine has no effect on IgG, all
the differences (di) should follow a normal distribution
whose population mean is “0”.
Paired t-test is aim to compare the new sample
mean (d, mean of the differences) with the known
population mean “0”.
Paired t-test is equivalent to One-sample t-test.
Population 1: hemoglobin of
general adult men
μ0=0 μd ?
n=12
d=475.66
s=84.27
μd≠μ0 due to “not a same population” d≠ 0
Sample
μd= μ 0 =0
n=12
d=475.66
s=84.27
μd= μ0 due to “Sampling error” d≠ 0
Hypothesis testing of paired t-test
Step 1: Set hypotheses and the level of test
H 0： d  0 H 1 :  d  0   0.05
Step 2 : Select an appropriate test and
calculate the test statistics (t value)
n  12,  d  5707.95, d   d n  5707.95 / 12  475.66

5707.95 2
2793182.166 
 d  2793182.166, S d 
2 12  84.2747
12  1
475.66  0
t  19.552,   12  1  11
84.2747 / 12
Step 3 ： Determine P value, and make a conclusion
 Two-sided , P=0.00000000068 <0.05. On the level of  =
0.05 ， H0 is rejected;
 we could say that there has difference between the IgG
before and after treatment on the measles boys;
 combined with the detail information of data, it could be
concluded that “after treatment, IgG of boys with measles
could be increased”, then “this medicine has effect on
measles”.
Example 3
10 beverage products were randomly sampled, two
methods were used to determine the fat content. Is
there difference between the two methods?
(1) Set hypotheses and the significance
level of test
H0 ： d ＝ 0
H1 ： d≠0
=0.05
(2) calculate the test statistic
n=10 ， d=2.724 ， d2=0.8483 ，
d  d / n  2.724 /10  0.2724
(  d ) 2
(2.724) 2
d 2  0.8483 
Sd  n  10  0.1087
n 1 10  1
0.2724
t  7.925,   10  1  9
0.1087 10
(3) Determine P value, and make a conclusion
P = 0.000024 <0.05, on the level of =0.05, reject

H0 ， accept H1 ；
there has difference between the measurements of
two methods ；
combined with the data, it is concluded that “the
measurement of method1 is higher than that of
method2”.
Independent two samples / groups t-test
 Two independent samples taken from normal
populations are compared their means.
 X 1 ≠ X 2 , 1= 2 ?
Conditions:
① quantitative variable in the population follows a
normal distribution.
② 
The variances are equal (   2)
2
1
2
Example 4
Two groups of rats were fed by different food, one group (n1=7)
fed by higher protein and the other group (n2=12) fed by
lower protein. Comparing the effects of different food on
increasing weight.
The increasing weight of rats in different groups
Groups 1 2 3 4 5 6 7 8 9 10 11 12 Mean SD
Lower
70 118 101 85 107 132 94 101 20.62
protein
Higher
134 146 104 119 124 161 107 83 113 129 97 123 120 20.48
protein
 Two independent samples taken from normal
populations are compared their means.
μ1 μ2
n1=7 n2=12
X1 X2
s1 s2
 X1 ≠ X2, 1= 2 ?
 We should think about “whether the variances are
equal between the two normal populations” firstly.
 And, we only know it through performing a
hypothesis test.
 12   22 
 2
2
1
2
First: to perform homogeneity of variance test
(F-test)
H 0： 12 = 22
H1： 12   22
 =0.05
S12
F  2 ~ F  1 , 2 ，  1  n1  1, 2  n2  1
S2
F - distribution
24/2/7 安医大公卫学院流统系
安徽医科大学流行病与卫生统计学
王静制作
系王静
if F  F  2，（ 1，
 2）
 P   , reject H 0 ，variances are not equal.
if F  F  2，（ 1，
 2）
 P   , not reject H 0 ，variances are equal.
 When the two population variances are
equal, then do Pooled t-test.
 Sc2 is the variance by pooled estimation
with S12 and S22. (confederate)
X1  X 2
t ,   n1  n2  2
SX1X 2
1 1 (n1  1) S12  (n2  1) S 22 1 1

SX1X 2  SC2 (  )  (  )
n1 n2 n1  n2  2 n1 n2
 When the two population variances are
not equal to each other, then do t’-test.
 (t’ : t apostrophe )
t 
X1  X 2
,  
S 2X1  2
S X2 2
S 12 S 22 S 4X1 S 4X 2
 
n1 n2 n1  1 n2  1
Two groups of rats were fed by different food, one group
(n1=7) fed by higher protein and the other group
(n2=12) fed by lower protein. Comparing the effects
of different food on increasing weight.
The increasing weight of rats in different groups
Groups 1 2 3 4 5 6 7 8 9 10 11 12 Mean SD
Lower
70 118 101 85 107 132 94 101 20.62
protein
Higher
134 146 104 119 124 161 107 83 113 129 97 123 120 20.48
protein
First to test if two variances are equal
H 0： 12 = 22
H1： 12   22
 =0.05
2
20.62
F 2
 1.014 ~ F（ 1，
 ），  1  7  1, 2  12  1
20.48 2
P=0.928
F 0.05 / 2， 11 )  3.88  1.014
( 6，
 P  0.05， reject
H 0 isn’t H 1
，
Variances are equal
rejected
Then to perform the pooled t-test
H 0 : 1   2
H1 : 1   2
  0.05
 ( X  X ) 2
  ( X  X ) 2
sc2  1 1 2 2
n1  n2  2
 X 1  (  X 1 ) / n1   X 2  (  X 2 ) / n2
2 2

n1  n2  2
177832  1440 2
/ 12  73959  707 2
/7
sc 
2
 446.12
12  7  2
2 1 1 1 1
s X 1  X x  sc (  )  446.12(  )  10.05
n1 n2 12 7
X 1  X 2 120  101
t     1   2  6  11  17
 1.891 ，
s X1  X 2 10.05
 Two-sided , P =0.0758 >0.05. On the level of 
= 0.05 ， H0 isn’t rejected ；
 conclusion could be that there is no difference
on increasing weight between the higher
protein and lower protein groups, which is
equal to “higher or lower protein in food has no
effect on increasing weight”.
Questions for hypothesis testing
 What does P-value mean?
 What are the prerequisites of t-test?
 Type I error & Type II error
 Attentions for hypothesis testing
Q1: What does P-value mean?
 P-value is the area of the tails in the distribution

of the test statistic (such as “t-distribution”)
beyond the values (such as “t”) calculated from
the sample under H0 .
-t t
Q2: What are the prerequisites of t-test?
 The variable distributed a normal distribution;

 Sample size is small;
 The variances are equal.
Q3: What are Type I error & Type II error?
The truth Result of hypothesis test

not reject H0 reject H0
H0 holds Correct Wrong
True negative False positive
1 －  (specificity)  (Type I error)
Wrong Correct
H1 holds False negative True positive
 (Type Ⅱ error) 1 －  (sensitivity)
H 0 : 1   2 H1 : 1   2

① Under a certain sample size (n), if  is
decreased ， will be increased.
“the better the smallest  ?”
② only increasing sample size n,  and 
would be decreased at the same time.
Q4: What are the attentions for hypothesis
testing?
 Research design should be scientific and rigorous;
 How to choose a certain hypothesis test should be
based on the different data.
 Conclusion could not be absolute.
 Conclusion should be reached from the point of
view in major ultimately.
 Hypothesis test and confidence interval are equally
important and often used simultaneously.

Lecture 8 T Test New

Uploaded by

Copyright:

Available Formats

Lecture 8 T Test New

Uploaded by

Document Information

Original Title

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

Lecture 8 T Test New

Uploaded by

Copyright:

Available Formats

Medical Statistics

Population characteristics Sample information

SEM reflects the variation among a group

Hypothesis test is one aspect of statistical inference.

What is the basic principle of hypothesis test?

Question: if there had a difference on the

due to sampling error due to different populations

-- with the help of hypothesis test

Basic logic of a hypothesis test:

Then to calculate a probability ( P -value)

To obtain the P value is based on a statistical distribution law of

small probability (event) principle

H0: μ2 = μ1= 140

It is a random quantitative variable (hemoglobin) with

If Xi follows a normal distribution N (  ,  ) ,

② conclusion in major: to explain the content of H0 or H1 from

① Sample size n is small ( usually n<50 ).

Question: if there had a difference on the hemoglobin

 Based on the current sample, the current sample

follows. 7 980.01 1379.59

μd≠μ0 due to “not a same population” d≠ 0

μd= μ0 due to “Sampling error” d≠ 0

Step 1: Set hypotheses and the level of test

n  12,  d  5707.95, d   d n  5707.95 / 12  475.66

P = 0.000024 <0.05, on the level of =0.05, reject

The increasing weight of rats in different groups

1 1 (n1  1) S12  (n2  1) S 22 1 1

The increasing weight of rats in different groups

 P-value is the area of the tails in the distribution

 The variable distributed a normal distribution;

The truth Result of hypothesis test

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.