Scribd
Upload
55 views30 pages

Design of Experiments

This document provides an overview of analysis of variance (ANOVA) techniques for experimental design. It discusses: 1) Partitioning total variability in a response variable into treatment and error components using ANOVA; 2) Calculating mean squares and using an F-test to determine if treatment means differ significantly; 3) Extending the basic single-factor ANOVA model to account for multiple factors and interaction effects. The key idea is that ANOVA allows experimenters to test hypotheses about differences in treatment means and identify optimal factor levels.

Uploaded by

snakefangs
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
55 views30 pages

Design of Experiments

This document provides an overview of analysis of variance (ANOVA) techniques for experimental design. It discusses: 1) Partitioning total variability in a response variable into treatment and error components using ANOVA; 2) Calculating mean squares and using an F-test to determine if treatment means differ significantly; 3) Extending the basic single-factor ANOVA model to account for multiple factors and interaction effects. The key idea is that ANOVA allows experimenters to test hypotheses about differences in treatment means and identify optimal factor levels.

Uploaded by

snakefangs
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 30

DESIGN OF EXPERIMENTS

ANALYSIS OF VARIANCE

Normal Distribution
Gaussian Equation:
m

f(y) =

Measurement:
yi = + ei

m
b

P(a y b) =

f ( y )dy

f ( y)dy 1

Estimation of Parameters
Measurement:
yi = + ei

The average (mean) :

1 n
y yi estimates the population mean m
n i n1
1
y 2
1 yi n estimates2 the population mean m 2 2
Deviation)
:
SThe
nSample
( yi y ) (Standard
estimates the
variance s
i 1 Variance
n 1 i n1
1
2
2
2
S
( yi y ) estimates the variance s

n 1 i 1

The Standard Normal Distribution

One way to simplify calculating probabilities is to use a


Normal Deviate (z)
Subtract m from all y observations and the new
observations will have a mean of 0
Divide these new observations by s and the new standard
deviation becomes 1.

z
1

f ( z)
exp
2
2
2

ym

Result: f(z) = a normal standard distribution

Areas under the standard normal distribution curve

m z= 0
s z= 1
Y

f ( z )dz

Y
Y

rea

rea

rea

rea

rea

rea

rea

rea

-4,00

0,0000

-3,00

0,0013

-2,00

0,0228

-1,00

0,1587

0,00

0,5000

1,00

0,8413

2,00

0,9772

3,00

0,9987

-3,99

0,0000

-2,99

0,0014

-1,99

0,0233

-0,99

0,1611

0,01

0,5040

1,01

0,8438

2,01

0,9778

3,01

0,9987

-3,98

0,0000

-2,98

0,0014

-1,98

0,0239

-0,98

0,1635

0,02

0,5080

1,02

0,8461

2,02

0,9783

3,02

0,9987

-3,97

0,0000

-2,97

0,0015

-1,97

0,0244

-0,97

0,1660

0,03

0,5120

1,03

0,8485

2,03

0,9788

3,03

0,9988

-3,96

0,0000

-2,96

0,0015

-1,96

0,0250

-0,96

0,1685

0,04

0,5160

1,04

0,8508

2,04

0,9793

3,04

0,9988

-3,95

0,0000

-2,95

0,0016

-1,95

0,0256

-0,95

0,1711

0,05

0,5199

1,05

0,8531

2,05

0,9798

3,05

0,9989

-3,94

0,0000

-2,94

0,0016

-1,94

0,0262

-0,94

0,1736

0,06

0,5239

1,06

0,8554

2,06

0,9803

3,06

0,9989

-3,93

0,0000

-2,93

0,0017

-1,93

0,0268

-0,93

0,1762

0,07

0,5279

1,07

0,8577

2,07

0,9808

3,07

0,9989

-3,92

0,0000

-2,92

0,0018

-1,92

0,0274

-0,92

0,1788

0,08

0,5319

1,08

0,8599

2,08

0,9812

3,08

0,9990

-3,91

0,0000

-2,91

0,0018

-1,91

0,0281

-0,91

0,1814

0,09

0,5359

1,09

0,8621

2,09

0,9817

3,09

0,9990

Portland Cement Formulation Summary Statistics


Observation Formula 1
(Sample), j
y1,j

Formula 2
y2,j

Formulation 1

y1 16.76

16.85

17.50

16.40

17.63

S12 0.100

17.21

18.25

16.35

18.00

S1 0.316

16.52

17.86

17.04

17.75

16.96

18.22

Formulation 2

17.15

17.90

16.59

17.96

y2 17.92

10

16.57

18.15

Average:

16.76

17.92

yij = i + eij (i=1,2, j=1-10)

n1 10

S 0.061
2
2

S 2 0.247
n2 10

Frequency

Cement Formulation Data


y1 16.76

y2 17.92

S12 0.100

S 22 0.061

S1 0.316

S 2 0.247

n1 10

n2 10
0.247
0.316

16.76

14

15

16

17

17.92

18

19

Observation

Do these formulations differ?

20

21

The Two-Sample t-Test


k degrees
of freedom

y1 y 2
t0
1 1
Sp

n1 n2

2
2
(
n

1
)
S

(
n

1
)
S
1
2
2
S p2 1
n1 n 2 2

t0 follows a
t-distribution
with n1+n2-2 degrees
of freedom

The Two-Sample t-Test


Cement
Formulation:
t0 = -9.13

t distribution
with 18 DOF
2.5%

2.5%

m1 = m2 is considered
to be a rare event if
| t0 | is 2.101.

The P-value (3.6810-8) is the risk of


wrongly rejecting the null hypothesis of
equal means

What If There Are More


Than Two Factor Levels?
The t-test does not directly apply
There are lots of practical situations where
there are either more than two levels of
interest, or there are several factors of
simultaneous interest
The analysis of variance (ANOVA) is the
appropriate analysis engine for these types
of experiments
Used extensively today for industrial
experiments

Cotton Fiber Example

New synthetic fiber to make cloth for shirts


Response variable: tensile strength
Cotton content vary between 15 and 35%
Each experiments replicated 5 times
What is the best weight % of cotton to use?

Does changing the


cotton weight
percent change the
mean tensile
strength?
Is there an optimum
level for cotton
content?

In general, there will be a levels of the factor, or a


treatments, and n replicates of the experiment, run in
random order, i.e. a completely randomized design

N = an total runs

Objective is to test for differences between the a means

Models for the Data


Consider the normal (means) model:
yij = mi + eij (mi: mean for the i:th treatment)

If we define:

mi = m + ti

We will get an effects model:

yij = m + ti + eij

where ti is the i:th treatment effect

Analysis of Variance (ANOVA)


The name analysis of variance is derived from
a partitioning of the total variability in the
response variable into its components parts
The basic single-factor ANOVA model is

i 1, 2,..., a
yij m t i e ij ,
j 1, 2,..., n

Treatments
Replications

i 1, 2,..., a
yij m t i e ij ,
j 1, 2,..., n

ANOVA

Definitions:
Observation Mean:
n

yi .

y
j 1

m an overall mean, t i ith tr


Overall Mean:

e ij experimental
error, NID(0
a
n

ij

(within treatments)

y..

y
i 1 j 1

ij

(all measurements)

The Analysis of Variance


Total Corrected Sum of Squares:
a

SS T ( y ij y .. ) 2

a measure of the
total variability

i 1 j 1

The basic ANOVA partitioning is:

SS T

2
(
y

y
)
ij ..
i 1 j 1

2
[(
y

y
)

(
y

y
)]
i . .. ij i .
i 1 j 1

Treatment
Average

The Analysis of Variance


a

2
[(
y

y
)

(
y

y
)]
i. .. ij i.
i 1 j 1

SS T

2
2
(
y

y
)

(
y

y
)
i . .. ij i . 2 ( y i . y .. )( y ij y i. )
i 1 j 1

i 1 j 1

n ( y i . y .. )
i 1

i 1 j 1

(y
j 1

ij

yi. ) 0

The Analysis of Variance


a

SS T n ( y i . y .. ) 2 ( y ij y i . ) 2 SS Trea tm en ts SS E rro rs
i 1

i 1 j 1

The Analysis of Variance

SST SSTreatments SS E

A large value of SSTreatments reflects large


differences in treatment means
A small value of SSTreatments likely indicates no
differences in treatment means
Formal statistical hypotheses are:

H 0 : m1 m 2

ma

H1 : At least one mean is different

The Analysis of Variance

While sums of squares cannot be directly compared to


test the hypothesis of equal means, mean squares can.
A mean square (MS) is a sum of squares divided by its
a
degrees of freedom:
n ( y i . y.. ) 2
SS Treatments
MS Treatments
i 1
a 1
a 1
a

MS Error

SS Error

N a

2
(
y

y
)
ij i.
i 1 j 1

N a

Recall the equation for the


sample variance:

S2

2
(
y

y
)
i
i 1

n 1

The Analysis of Variance

If the treatment means are equal, the treatment and


error mean squares will be (theoretically) equal.
m1 = m2 = = ma
MSTreatment ~ MSError

If treatment means differ, the treatment mean


square will be larger than the error mean square.
mi mk
MSTreatment > MSError

If the error mean squares are larger than the


treatment mean squares, there is a PROBLEM

The Analysis of Variance


It turns out that the ratio of MSTreatments and
MSError follows the F distribution with (a-1)
and (N-a) degrees of freedom

So the test statistic for the hypothesis of no


difference in mean is:

MS Treatments
F0
MS Error
Reject the null hypothesis
(equal treatment means) if

F0 F , a 1, a ( n 1)

Cotton Fiber Example

ANOVA
Source of
Variation
Between Groups
Within Groups

Total

SS

df

MS

475.76

161.2

20

636.96

24

P-value

F crit

118.94 14.75682 9.13E-06 2.866081


8.06

MS Treatments
F0
14.76
MS Error

The F Distribution
With a-1 (5-1=4) and N-a (25-5=20)
Degrees of Freedom (F4,20)
F0.05,4,20 = 2.87 (5% probability)
F0.01,4,20 = 4.43 (1% probability)

F0 = 14.76 >> F0.01,4,20

MORE FACTORS?
The Statistical Effects Model
i 1, 2,..., a

yijk m t i j (t )ij e ijk j 1, 2,..., b


k 1, 2,..., n

ti = effect of treatment A
j = effect of treatment B
(t)ij = synergistic effect of
treatments A and B (interaction term)

Extension of ANOVA to Factorials


a

i 1

j 1

2
2
2
(
y

y
)

bn
(
y

y
)

an
(
y

y
)
ijk ...
i.. ...
. j. ...
i 1 j 1 k 1

n ( yij . yi.. y. j . y... ) ( yijk yij . ) 2


2

i 1 j 1

Degrees of Freedom:
SST = abn-1
SSA = a-1
SSB = b-1
SSAB = (a-1)(b-1)
SSE = ab(n-1)

i 1 j 1 k 1

Can calculate mean


squares as before to
determine if the
variables have effect

ANOVA Table Fixed Effects Case

Available computer softwares can


easily perform these calculations.

Factorials with More Than


Two Factors
Basic procedure is similar to the two-factor
case; all a,b,c,,k treatment combinations
are run in random order
ANOVA identity is also similar:

SST SS A SS B
SS ABC

SS AB SS AC

SS AB

SS E

Computers are great, are they not?

Minitab Statistical Software


Download 30-day trial version from
web site:
http://www.minitab.com/
1) Click on Products tab
2) MiniTab 17: Click: Learn More
3) Click: try Minitab 17,
then 30-day Free trial

You might also like

pFad - Phonifier reborn

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

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy