Design of Engineering Experiments
Hussam Alshraideh
Chapter 3: Analysis of Variance
October 4, 2015
Hussam Alshraideh (JUST) ANOVA October 4, 2015 1 / 44
Overview
1 ANOVAIntroductionFixed effects caseModel Adequacy CheckingComparison of MeansSample Size Determination
2 Other Examples of Single-Factor Experiments
3 The Random Effects Model
Hussam Alshraideh (JUST) ANOVA October 4, 2015 2 / 44
ANOVA Introduction
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 thantwo levels of interest, or there are several factors of simultaneousinterest
The analysis of variance (ANOVA) is the appropriate analysis“engine” for these types of experiments
The ANOVA was developed by Fisher in the early 1920s, and initiallyapplied to agricultural experiments.
Used extensively today for industrial experiments.
Hussam Alshraideh (JUST) ANOVA October 4, 2015 3 / 44
ANOVA Fixed effects case
An Example (See pg. 66)
An engineer is interested in investigating the relationship between theRF power setting and the etch rate for this tool. The objective of anexperiment like this is to model the relationship between etch rateand RF power, and to specify the power setting that will give adesired target etch rate.
The response variable is etch rate.
She is interested in a particular gas (C2F6) and gap (0.80 cm), andwants to test four levels of RF power: 160W, 180W, 200W, and220W. She decided to test five wafers at each level of RF power.
The experimenter chooses 4 levels of RF power 160W, 180W, 200W,and 220W
The experiment is replicated 5 times, runs made in random order
Hussam Alshraideh (JUST) ANOVA October 4, 2015 4 / 44
ANOVA Fixed effects case
An Example (See pg. 66)
Hussam Alshraideh (JUST) ANOVA October 4, 2015 5 / 44
ANOVA Fixed effects case
An Example (See pg. 66)
Hussam Alshraideh (JUST) ANOVA October 4, 2015 6 / 44
ANOVA Fixed effects case
An Example (See pg. 66)
Does changing the power change the mean etch rate?
Is there an optimum level for power?
We would like to have an objective way to answer these questions
The t-test really doesnt apply here more than two factor levels
Hussam Alshraideh (JUST) ANOVA October 4, 2015 7 / 44
ANOVA Fixed effects case
The Analysis of Variance (Sec. 3.2, pg. 68)
In general, there will be a levels of the factor, or a treatments, and nreplicates of the experiment, run in random order: a completelyrandomized design (CRD)N = an total runsWe consider the fixed effects case, the random effects case will bediscussed later.Objective is to test hypotheses about the equality of the a treatmentmeans
Hussam Alshraideh (JUST) ANOVA October 4, 2015 8 / 44
ANOVA Fixed effects case
The Analysis of Variance
The name analysis of variance stems from a partitioning of the totalvariability in the response variable into components that areconsistent with a model for the experiment.
The basic single-factor ANOVA model is:
yij = µ+ τi + εij
{i = 1, 2, · · · , aj = 1, 2, · · · n
where:
µ an overall meanτi= i th treatment effectεij=experimental error ∼ N(0, σ2)
Hussam Alshraideh (JUST) ANOVA October 4, 2015 9 / 44
ANOVA Fixed effects case
Models for the Data
There are several ways to write a model for the data:
yij = µ+ τi + εij is called the effects model
Let µi = µ+ τi , then
yij = µi + εij is called the means model
Hussam Alshraideh (JUST) ANOVA October 4, 2015 10 / 44
ANOVA Fixed effects case
The Analysis of Variance
Total variability is measured by the total sum os squares:
SST =a∑
i=1
n∑j=1
(yij − y ··)2
The basic ANOVA partitioning is:
a∑i=1
n∑j=1
(yij − y ··)2 =
a∑i=1
n∑j=1
[(y i· − y ··) + (yij − y i·)]2
na∑
i=1
(y i· − y ··)2 +a∑
i=1
n∑j=1
(yij − y i·)2
SST = SSTreatment + SSE
Hussam Alshraideh (JUST) ANOVA October 4, 2015 11 / 44
ANOVA Fixed effects case
The Analysis of Variance
SST = SSTreatment + SSE
A large value of SSTreatments reflects large differences in treatmentmeans
A small value of SSTreatments likely indicates no differences intreatment means
Formal statistical hypotheses are:
H0 : µ1 = µ2 = · · · = µa
H1 : At least one mean is different
Hussam Alshraideh (JUST) ANOVA October 4, 2015 12 / 44
ANOVA Fixed effects case
The Analysis of Variance
While sums of squares cannot be directly compared to test thehypothesis of equal means, mean squares can be compared.
A mean square is a sum of squares divided by its degrees of freedom:
dfTotal = dfTreatment + dfError
an − 1 = (a− 1) + a(n − 1)
MSTreatment =SSTreatment
a− 1, MSE =
SSEa(n − 1)
If the treatment means are equal, the treatment and error meansquares will be (theoretically) equal.
If treatment means differ, the treatment mean square will be largerthan the error mean square.
Hussam Alshraideh (JUST) ANOVA October 4, 2015 13 / 44
ANOVA Fixed effects case
The Analysis of Variance: Summary
Computing, see text, pp 69
The reference distribution for F0 is the Fa−1,a(n−1) distribution
Reject the null hypothesis (equal treatment means) if
F0 > Fα,a−1,a(n−1)
Hussam Alshraideh (JUST) ANOVA October 4, 2015 14 / 44
ANOVA Fixed effects case
The Analysis of Variance: Computing formulas
SST =a∑
i=1
n∑j=1
y2ij −
y2··N
SSTreatment =1
n
a∑i=1
y2i· −
y2··N
SSE = SST − SSTreatment
Hussam Alshraideh (JUST) ANOVA October 4, 2015 15 / 44
ANOVA Fixed effects case
ANOVA table: Example 3.1
Hussam Alshraideh (JUST) ANOVA October 4, 2015 16 / 44
ANOVA Fixed effects case
ANOVA table: Example 3.1
Hussam Alshraideh (JUST) ANOVA October 4, 2015 17 / 44
ANOVA Fixed effects case
ANOVA computing using software
Design-Expert,
JMP
Minitab
See pages 102-105 for discussion on summary statistics from thesepackages.
Hussam Alshraideh (JUST) ANOVA October 4, 2015 18 / 44
ANOVA Model Adequacy Checking
Model Adequacy Checking in the ANOVA
Checking assumptions is important
NormalityConstant varianceIndependence
Have we fit the right model?
Later we will talk about what to do if some of these assumptions areviolated
Hussam Alshraideh (JUST) ANOVA October 4, 2015 19 / 44
ANOVA Model Adequacy Checking
Model Adequacy Checking in the ANOVA
Examination of Residuals(sec 3.4):
eij = yij − yij
= yij − y i·
Computer softwaregenerates the residuals
Residual plots are veryuseful
Normal probability plotof residuals
Hussam Alshraideh (JUST) ANOVA October 4, 2015 20 / 44
ANOVA Model Adequacy Checking
Other Important Residual Plots
Residuals vs order → Independence
Residuals vs fitted values → Constant variance
Hussam Alshraideh (JUST) ANOVA October 4, 2015 21 / 44
ANOVA Comparison of Means
Comparison of Means
The analysis of variance tests the hypothesis of equal treatmentmeans
If that hypothesis is rejected, we dont know which specific means aredifferent
Determining which specific means differ following an ANOVA is calledthe multiple comparisons problem
There are lots of ways to do this. → see text, Section 3.5
We will use pairwise t-tests on means sometimes called Fishers LeastSignificant Difference (or Fishers LSD) Method
Hussam Alshraideh (JUST) ANOVA October 4, 2015 22 / 44
ANOVA Comparison of Means
Graphical Comparison of Means: sliding t-distribution
Hussam Alshraideh (JUST) ANOVA October 4, 2015 23 / 44
ANOVA Comparison of Means
Fishers’s LSD Method
Controls error rate (α) for each test, but not the rate for the tests as awhole.
H0 : µi = µj
H1 : µi 6= µj
Reject H0 if:
|t0| =
∣∣∣∣∣∣ y i· − y j·√MSE ( 1
ni+ 1
nj)
∣∣∣∣∣∣ > tα/2,N−a
or:
LSD =∣∣y i· − y j·
∣∣ > tα/2,N−a
√MSE (
1
ni+
1
nj)
Hussam Alshraideh (JUST) ANOVA October 4, 2015 24 / 44
ANOVA Comparison of Means
Tukey’s Method
Based on the studentized range statistic.
Controls the family-wise error rate.
Hussam Alshraideh (JUST) ANOVA October 4, 2015 25 / 44
ANOVA Comparison of Means
Why Does the ANOVA Work?
We are sampling from Normal populations so:
SSTreatment
σ2∼ χ2
a−1 if H0 is true, andSSEσ2∼ χ2
a(n−1)
so:
F0 =SSTreatment/(a− 1)
SSE/(a(n − 1))=
χ2a−1/(a− 1)
χ2a(n−1)/(a(n − 1))
∼ Fa−1,a(n−1)
Finally,
E [MSTreatment ] = σ2 +
na∑
i=1
τ2i
a− 1, and E [MSE ] = σ2
Hussam Alshraideh (JUST) ANOVA October 4, 2015 26 / 44
ANOVA Sample Size Determination
Sample Size Determination
See text section 3.7.
FAQ in designed experiments. Answer depends on lots of things;including
what type of experiment is being contemplated,how it will be conducted,resources, anddesired sensitivity
Sensitivity refers to the difference in means that the experimenterwishes to detect
Generally, increasing the number of replications increases thesensitivity or it makes it easier to detect small differences in means
Hussam Alshraideh (JUST) ANOVA October 4, 2015 27 / 44
ANOVA Sample Size Determination
Sample Size Determination: Fixed Effects Case
Can choose the sample size to detect a specific difference in meansand achieve desired values of type I and type II errors
Type I error : reject H0 when it is true (α)
Type II error : fail to reject H0 when it is false (β)
Power = 1− βOperating characteristic curves plot β against a parameter Φ where
Φ2 =
na∑
i=1
τ2i
aσ2
Hussam Alshraideh (JUST) ANOVA October 4, 2015 28 / 44
ANOVA Sample Size Determination
Sample Size Determination: Fixed Effects Case
The OC curves for the fixed effects model are in the Appendix, TableV
A very common way to use these charts is to define a difference intwo means D of interest, then the minimum value of Φ2 is
Φ2 =nD2
2aσ2
Typically work in term of the ratio of D/σ and try values of n untilthe desired power is achieved
Most statistics software packages will perform power and sample sizecalculations see page 108
There are some other methods discussed in the text
Hussam Alshraideh (JUST) ANOVA October 4, 2015 29 / 44
ANOVA Sample Size Determination
Sample Size Determination: Fixed Effects Case, Table V
Hussam Alshraideh (JUST) ANOVA October 4, 2015 30 / 44
ANOVA Sample Size Determination
Sample Size Determination: Fixed Effects Case, Minitaboutput
Hussam Alshraideh (JUST) ANOVA October 4, 2015 31 / 44
Other Examples of Single-Factor Experiments
3.8 Other Examples of Single-Factor Experiments
Hussam Alshraideh (JUST) ANOVA October 4, 2015 32 / 44
Other Examples of Single-Factor Experiments
3.8 Other Examples of Single-Factor Experiments
Hussam Alshraideh (JUST) ANOVA October 4, 2015 33 / 44
Other Examples of Single-Factor Experiments
3.8 Other Examples of Single-Factor Experiments
Hussam Alshraideh (JUST) ANOVA October 4, 2015 34 / 44
Other Examples of Single-Factor Experiments
3.8 Other Examples: Marketing Example
Hussam Alshraideh (JUST) ANOVA October 4, 2015 35 / 44
Other Examples of Single-Factor Experiments
3.8 Other Examples: Marketing Example
Hussam Alshraideh (JUST) ANOVA October 4, 2015 36 / 44
The Random Effects Model
3.9 The Random Effects Model
There are a large number of possible levels for the factor(theoretically an infinite number)
The experimenter chooses a of these levels at random
Inference will be to the entire population of levels
Hussam Alshraideh (JUST) ANOVA October 4, 2015 37 / 44
The Random Effects Model
3.9 The Random Effects Model
Hussam Alshraideh (JUST) ANOVA October 4, 2015 38 / 44
The Random Effects Model
3.9 The Random Effects Model: covariance structure
Hussam Alshraideh (JUST) ANOVA October 4, 2015 39 / 44
The Random Effects Model
3.9 The Random Effects Model: covariance structure
For a = 3 and n = 2
Hussam Alshraideh (JUST) ANOVA October 4, 2015 40 / 44
The Random Effects Model
3.9 The Random Effects Model
Hussam Alshraideh (JUST) ANOVA October 4, 2015 41 / 44
The Random Effects Model
3.9 The Random Effects Model
E (MSTreatment) = σ2 + nσ2τ
E (MSE ) = σ2
ANOVA F-test is identical to the fixed-effects case.
σ2 = MSE
σ2τ =
MSTreatment −MSEn
Hussam Alshraideh (JUST) ANOVA October 4, 2015 42 / 44
The Random Effects Model
3.9 The Random Effects Model: Example
Hussam Alshraideh (JUST) ANOVA October 4, 2015 43 / 44
The Random Effects Model
Homework, due on Monday October 20th 2014
Solve the following end of chapter problem form the textbook. Solve bothmanually and using Minitab.
Problem 3.7
Problem 3.11
Problem 3.25
Problem 3.31 parts (a) and (b).
Hussam Alshraideh (JUST) ANOVA October 4, 2015 44 / 44