Analysis Of Variance

( ANOVA )
 Analysis Of Variance

Learning Objectives
 Basic Concept of ANOVA
– Hypothesis Testing
– Assumptions
– Models

 ANOVA Models
One-Way ANOVA:
– Fixed Effects
– Random Effects

Two-Way ANOVA/Multi-Factor ANOVA:

– Fixed Effects
– Random Effects
– Mixed Effects

Seagate Confidential 2 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

ANOVA is another hypothesis test method used:
 For comparing means of different populations
 When there are 3 or more populations in the study

Example:
(a) 3 lines set up to plate the same product. We want
to compare if the mean plating thickness of
products from the 3 lines are the same or not.

Seagate Confidential 3 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

ANOVA is a statistical analysis technique in
which the total variation is divided into
meaningful sources of variation.

 Total
2
  Component
2
1   2
Component 2     2
Component k

Recall from Measurement System Analysis,

 Total
2
  Gage
2
  2Parts
  2Re peatability   2Re producibility   2Parts
  2Inherent Error   Operators
2
  Operators
2
 Parts   2
Parts

Seagate Confidential 4 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

Variation
Inputs between
subgroups
Process 1 1
Y1
Inputs
Process 2 2 1= 2= 3?? Total
Variation
Y2
Inputs
Process 3 3 Y3
Variation
within
subgroups

Seagate Confidential 5 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

 To appreciate ANOVA, we need to revisit
sample variance again.
 x  x
n 2
i
s 2
 i 1
n1

Why must we use xi  x  ? Why not xi  x ?

2


4
6
2

x6 x3 x5 x1 x2 x7 x4 x
x
Seagate Confidential 6 Supplier Six Sigma Modular Training
 Analysis Of Variance

Basic Concept of ANOVA

Sample Variance - Moment of Inertia
 Imagine each data as a unit mass rotating about
a central axis ( x ) with a radial distance of i.
4
6
2

x6 x3 x5 x1 x2 x7 x4 x
x
 In mechanics, the effect of these rotating masses
is a total pulling force (moment of inertia or
second moment) on the central axis.
Seagate Confidential 7 Supplier Six Sigma Modular Training
 Analysis Of Variance

Basic Concept of ANOVA

Sample Variance - Moment of Inertia
 The variance is equivalent (in concept) to the
total moment of inertia divided by the degrees of
freedom, to give an “average” moment of inertia
per data.
The numerator directly
represents the amount
n

i 1 xi  x 2
of variation, and is
s2  known as the
n1 Sum of Squares

 The variance is sometimes known as the second

moment of data.
Seagate Confidential 8 Supplier Six Sigma Modular Training
 Analysis Of Variance

Basic Concept of ANOVA

Total Variation (Total Sum of Squares)
 The first step in ANOVA is to compute the total variation contributed by all the
data.
 This is done by swinging every point about the grand average Y to get the
Total Sum of Squares.

90

80 Sample 2
70 Y31
60 Y11
50 
40
Y21
30 Sample 1
20 Sample 3

Seagate Confidential 9 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

Total Sum of Squares

SSTotal   
2
 
2
 y 11  y  y 12  y ..  y 21  y  .. y ij  y 
2
2

a n

 y 
2
SSTotal  ij -y Total Variation
i 1 j 1

Seagate Confidential 10 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

Between Variation
 The next step is to break down the total variation into respective sources of
variations. One of them is the variation between subgroups .
 This is obtained by swinging the Yi (with all the local data points attached)
about the grand average Y. If all the y’s are close to each other ,
what will be the between variation ?
90 Sample 2
80
n2=7
70
Y1 Y3
60 Y1 Y3
50 
40 Y2
30 Sample 1
n3= 7
20
n1=6 Y2 Sample 3

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

SSBetween  n1 y 1

2
 2
 y  n2 y 2  y ..  ..n i y i  y  2

a Variation
 y  2
SSFactors  ni 1
y between
i 1 subgroups

Seagate Confidential 12 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

Within Variation
 The third component is to get the variation within subgroups.
 This is done by with swinging the data points about its local mean Y i.

90

80 Sample 2
e Y3
Y1
31
70

60 e 12 e 11

50 
40 e 21

30 Sample 1
20 Y2 Sample 3

Seagate Confidential 13 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

SS Error  y 11
 y1   ..  y
2
21
 y2   ..y ij  y i 
2 2

a n Variation
SSError   y
i 1 j 1
ij -y 2
within
subgroups

Seagate Confidential 14 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

ANOVA Identity
 Putting back all the different components of
variation together, we have the ANOVA identity:

SSTotal = SSFactor + SSError

a n a a n

i=1 j=1
( yij – y )2 = n 
i
i=1
( y i – y ) 2

+ i=1 j=1 ij i
( y - y ) 2

Seagate Confidential 15 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

ANOVA Table
 The information from the variation decomposition is typically
organized into an ANOVA table:
a n a
SSError =  ( yij – yi )2 SS Factors = n i
i=1
( y i – y ) 2
i=1 j=1

Source Degree of Freedom Sum of Square Mean Square

Factor SSFactor

Error SSError

Total SSTotal

a n
SS Total =  ( y ij – y ) 2
i=1 j=1
Seagate Confidential 16 Supplier Six Sigma Modular Training
 Analysis Of Variance

Basic Concept of ANOVA

Degree of Freedom
Source Degree of Freedom Sum of Square Mean Square
Factor f = k–1 SSFactor
Error e = N – k SSError

Total  = N–1 SSTotal

To appreciate degree of freedom , let’s review the

sample variance calculation again .

Seagate Confidential 17 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

4
6
2

x6 x3 x5 x1 x2 x7 x4 x
x

 The n deviations x 1  x , x 2  x , …, x n  x always

sum to zero.
 So specifying the values of any (n - 1) of these
quantities automatically determines the last one.
 Only (n - 1) of the n deviations are truly free to
take on any value. The last deviation is fixed.
Seagate Confidential 18 Supplier Six Sigma Modular Training
 Analysis Of Variance

Basic Concept of ANOVA

Total Degree of Freedom (  for the SS Total )

90

80 Sample 2 n =7
2
70
Y1 Y31 Y3
60 Y11
50 
40 Y21
30 Sample 1
n3= 7
20 Y2 Sample 3
n1=6

The total degrees of freedom for the example above will be 19.
(N-1) deviations freely to move about the grand average Y
Seagate Confidential 19 Supplier Six Sigma Modular Training
 Analysis Of Variance

Basic Concept of ANOVA

Degree of Freedom
Source Degree of Freedom Sum of Square Mean Square
Factor f = k–1 SSFactor
Error e = N – k SSError

Total  = N–1 SSTotal

Here , k is the factor levels and the degrees of freedom for

factor is k - 1.

Seagate Confidential 20 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

Degree of Freedom for SS factor ( f )

90

Sample 2 n =7 Y31
80 2
70
Y1 Y3
60 Y11
50 
40 Y21
30 Sample 1
n3= 7
20 Y2 Sample 3
n1=6

The degrees of freedom for the factor will be 2 ( K-1), which

are total deviations freely to move about the grand average Y.

Seagate Confidential 21 Supplier Six Sigma Modular Training

 Analysis Of Variance

Basic Concept of ANOVA

Degree of Freedom
Source Degree of Freedom Sum of Square Mean Square
Factor f = k–1 SSFactor
Error e = N – k SSError

Total  = N–1 SSTotal

Lastly , the degree of freedom (Error) is:

Total Degree of Freedom - Degree of freedom for factor
f = N - k
Seagate Confidential 22 Supplier Six Sigma Modular Training
 Analysis Of Variance

Basic Concept of ANOVA

Degree of Freedom for SS Error (e)

90

80 Sample 2 n =7
2
70
Y1 Y3
60 Y11 Y31
50 
40 Y21
30 Sample 1
n3= 7
20 Y2 Sample 3
n1=6

The degrees of Freedom for the Error (e) will be 17 ( N - k)

deviations freely to move about the grand average Y.
Seagate Confidential 23 Supplier Six Sigma Modular Training
 Analysis Of Variance

Basic Concept of ANOVA

Mean Square
Source Degree of Freedom Sum of Square Mean Square
Factor f = k–1 SSFactor MSFactor = SSFactor  f
Error e = N – k SSError MSError = SSError  e
Total  = N–1 SSTotal

Mean Square ~ Sample variance (s2)

2 Sum of Squares
S = Degree of Freedom


i1 i
n
x  x 2

s2 
n 1
Seagate Confidential 24 Supplier Six Sigma Modular Training
 Analysis Of Variance

Basic Concept of ANOVA

Recall we want to find out 1=2==k ??

SS Factor is an indication of whether the above is true or not .

SS Factor small  1 = 2 =  = k

SS Factor Big  i  j for some i  j

at least one pair of group means is not equal

How do we determine whether SS Factor is big or small ?

 Compare against its natural variation represented by SS Error

 MS Factor

S2 Factor
~
MS Error S2 Error
Seagate Confidential 25 Supplier Six Sigma Modular Training
 Analysis Of Variance

Basic Concept of ANOVA

F-Test
S
2

Recall is actually an F statistic

2
S
1

H0 : 1 = 2 =  = k
all group means are equal
Ha : i  j for some i  j
at least one pair of group means is not equal
2
Variation between groups S between
Fcalculated   2
Variation within group S within
Fcrit

Compare Fcal MS Factor to Fcrit = F ( ,1,2)
MS Error 
If F cal > F crit, then reject Ho
Seagate Confidential 26 Supplier Six Sigma Modular Training
 Analysis Of Variance

One-Way ANOVA- Example 1

 Four lapping machines are being investigated to
see if they all provide the same amount of
roughness.
 Eighteen lapped bars are taken from production
for each machine and measured for roughness.
 Data is in ANOVA.mtw (columns C1-C3)

How many factors?

How many factor levels?

Seagate Confidential 27 Supplier Six Sigma Modular Training

 Analysis Of Variance

One-Way ANOVA- Example 1

 Construct the ANOVA Table (using Minitab):
Stat > ANOVA > One-way ...
Analysis of Variance for Roughness
Source DF SS MS F P
Lapper 3 0.008512 0.002837 3.39 0.023
Error 68 0.056834 0.000836
Total 71 0.065346
Individual 95% CIs For Mean
Based on Pooled StDev
Level N Mean StDev -----+---------+---------+---------+-
Lap_08 18 0.06249 0.02591 (--------*--------)
Lap_10 18 0.07484 0.02898 (--------*--------)
Lap_13 18 0.07929 0.03365 (--------*--------)
Lap_16 18 0.05151 0.02646 (--------*--------)
-----+---------+---------+---------+-
Pooled StDev = 0.02891 0.045 0.060 0.075 0.090

Seagate Confidential 28 Supplier Six Sigma Modular Training

 Analysis Of Variance

One-Way ANOVA- Example 1

 Interpret the P-Value (or the F-statistic) for the factor effect
– Assuming the residual assumptions are satisfied:
– If P-Value < 0.05, then REJECT Ho
– Otherwise, operate as if null hypothesis is true
If P is less than 5% then
at least one group mean
Analysis of Variance for Roughnes is different. In this case,
Source DF SS MS F P we reject the hypothesis
Lapper 3 .008512 0.002837 3.39 0.023 that all the group means
Error 68 .056834 0.000836 are equal. At least one
Total 71 .065346 Lapper mean is different.

F-test is close to 1.00 An F-test this large

2 2 2 2 when group means are
2    
   could
 Pooled
 1 2
4
3 4
similar. In this case, happen by chance, but in
In this case, it is calculated using the the F-test seems much less than one time out of
actual equation for pooling variances. greater. 40 chances.

Seagate Confidential 29 Supplier Six Sigma Modular Training

 Analysis Of Variance

ANOVA — Assumptions
 The observations are mutually independent.
– Stat  Nonparametrics  Runs Test

 The k groups exhibit homogeneity of variance.

i.e. 1 ² = 2 ² =  = k²
– Stat  ANOVA  Test for Equal Variances

 The data from each of the k groups is normally

distributed, implying all error terms are normally
distributed.
i.e. Factor Level i ~ NID(i, ²) or e ~ NID(0, ²)
– Stat  Basic Statistics  Normality Test
Seagate Confidential 30 Supplier Six Sigma Modular Training
 Analysis Of Variance

ANOVA -General Method

Step 1: State the Practical Problem
Step 2: Do the assumptions for the model hold?
 Response means are independent and normally distributed
 Population variances are equal across all levels of the factor
 Run a homogeneity of variance analysis - by factor level - first!
Step 3: State the Null and Alternate Hypotheses
Step 4: Construct the ANOVA Table
Step 5: Do the assumptions for the errors hold (residual analysis)?
 Errors of the model are independent and normally distributed

Step 6: Interpret the P-Value (or the F-statistic) for the factor effect
 P-Value < 0.05, then REJECT Ho
 Otherwise, operate as if the null hypothesis is true
Step 7: Translate the statistical conclusion into process terms

Seagate Confidential 31 Supplier Six Sigma Modular Training

 Analysis Of Variance

One Way
ANOVA-Example 1
Using the previous Example -
ANOVA.mtw (columns C1-C3)

Step 1
 The practical problem: Do the machines contribute to an overall increase
in roughness variation because their means are different ?

Seagate Confidential 32 Supplier Six Sigma Modular Training

 Analysis Of Variance

One Way ANOVA-Example 1

Step 2
 Do the assumptions for the model hold?
 Population by level are normally distributed .Won’t show
significance for small # of samples
 Variances are equal across all levels of the factor
– Stat > ANOVA > Test for Equal Variances

What if variances are

not equal, P<0.05?

Seagate Confidential 33 Supplier Six Sigma Modular Training

 Analysis Of Variance

One Way ANOVA-Example 1

Step 3
 State the Null and Alternate Hypotheses
 Ho: µ Lap_08 = µ Lap_10 = µ Lap_13 = µ Lap_16 (or) Ho: ’s = 0
 Ha: at least two lappers differ from each other (or) Ha: ’s  0

 Interpretation of the null hypothesis:

hypothesis the average
roughness is the same for each machine [or whichever
machine you use will NOT affect roughness].
 Interpretation of the alternate hypothesis:
hypothesis at least one of
the lapping machines gives a different average roughness
[or which machine you use WILL affect roughness].

Seagate Confidential 34 Supplier Six Sigma Modular Training

 Analysis Of Variance

One Way ANOVA-Example 1

Step 4
 Construct the ANOVA Table (using Minitab):
Stat > ANOVA > One-way ... Store Residuals & Fits
for later use
Analysis of Variance for Roughness
Source DF SS MS F P
Lapper 3 0.008512 0.002837 3.39 0.023
Error 68 0.056834 0.000836
Total 71 0.065346
Individual 95% CIs For Mean
Based on Pooled StDev
Level N Mean StDev -----+---------+---------+---------+-
Lap_08 18 0.06249 0.02591 (--------*--------)
Lap_10 18 0.07484 0.02898 (--------*--------)
Lap_13 18 0.07929 0.03365 (--------*--------)
Lap_16 18 0.05151 0.02646 (--------*--------)
-----+---------+---------+---------+-
Pooled StDev = 0.02891 0.045 0.060 0.075 0.090

Seagate Confidential 35 Supplier Six Sigma Modular Training

 Analysis Of Variance

One Way ANOVA-Example 1

Step 5 Stat > Regression > Residual Plots Trends? Outliers?

Normal?

If points are
ordered in time
sequence,
shows possible
time effects

Bell shaped? Same sigma?

Seagate Confidential 36 Supplier Six Sigma Modular Training
 Analysis Of Variance

One Way ANOVA-Example 1

Step 6
 Interpret the P-Value (or the F-statistic) for the factor effect
– Assuming the residual assumptions are satisfied:
– If P-Value < 0.05, then REJECT Ho
– Otherwise, operate as if null hypothesis is true
If P is less than 5% then
at least one group mean
Analysis of Variance for Roughnes is different. In this case,
Source DF SS MS F P we reject the hypothesis
Lapper 3 .008512 0.002837 3.39 0.023 that all the group means
Error 68 .056834 0.000836 are equal. At least one
Total 71 .065346 Lapper mean is different.

F-test is close to 1.00 An F-test this large

2 2 2 2 when group means are
2    
   could
Pooled  1 2
4
3 4
similar. happen by chance, but in
In this case, the F-test less than one time out of
When group sizes are equal is much greater. 40 chances.

Seagate Confidential 37 Supplier Six Sigma Modular Training

 Analysis Of Variance

One Way ANOVA-Example 1

Step 7
Translate the statistical conclusion to practical terms
The choice of lapping machine does affect roughness.
 But which ones give lower roughness, which ones give higher roughness,
and which ones are the same?

Individual 95% CIs For Mean

Based on Pooled StDev
Level -----+---------+---------+---------+-
Lap_08 (--------*--------)
Lap_10 (--------*--------)
Lap_13 (--------*--------)
Lap_16 (--------*--------)
-----+---------+---------+---------+-
Lower?
.045 0.060 0.075 0.090
Seagate Confidential 38 Supplier Six Sigma Modular Training
 Analysis Of Variance

ANOVA — Models
Fixed Effects (ANOVA Fixed) Model
 factor levels in the experiment are specifically chosen
 the conclusion is relevant only for the chosen levels and may not be
extended to similar treatments that were not considered

Random Effects (ANOVA Random ) Model

 factor levels are randomly drawn from a large population of treatments
 the conclusion is valid for the entire population of treatments

Mixed Effects (ANOVA Mixed ) Model

 one or more factors are fixed factors while others are random factors

Seagate Confidential 39 Supplier Six Sigma Modular Training

 Analysis Of Variance

ANOVA — Models
A k–Way ANOVA model involves the study of k
factors.
k–Way ANOVA Fixed Model : k factors;all
factors of fixed effect
k–Way ANOVA Random Model: k factors;all
factors of random effect
k–Way ANOVA Mixed Model : k factors;some
factors of fixed effect others of random effect

Seagate Confidential 40 Supplier Six Sigma Modular Training

 Analysis Of Variance

ANOVA Model
i = 1, 2, …, 5
Yij = μ + τj + εij j = 1, 2, 3, 4

A parameter common Random error components

to all materials called A parameter associated all assumed to be normally
the overall mean with the jth material and independently distributed
called the jth material effect with a mean of zero and
common variance σ2
Adding gives the
mean of the
jth material
We calculate estimates
of the parameters with
the use of data!
yij = y + yMj + rij
Seagate Confidential 41 Supplier Six Sigma Modular Training
 Analysis Of Variance

One-Way ANOVA-Fixed Model

Yij  .  i   ij

where Yij is the value of the response variable in

the jth trial for the ith factor level or treatment
. is the overall mean for all observations
i=i–. is the effect of the ith treatment s.t.
ii  0
ij is the random error component s.t.
 ij ~ N0,  2 
The hypothesis may now be rephrased:
H0 : 1 = 2 =  = r = 0
Ha : not all i equal 0

Seagate Confidential 42 Supplier Six Sigma Modular Training

 Analysis Of Variance

One-Way ANOVA-Fixed
Model
Source SS DF MS F-Statistic

Factor A SSA a–1 MSA MSA

MSE

Error SSE a(n–1) MSE

Total SSTO an–1

Seagate Confidential 43 Supplier Six Sigma Modular Training

 Analysis Of Variance

One-Way ANOVA-Random
Model
Yij  .  i   ij
where Yij is the value of the response variable in the
jth trial for the ith factor level or treatment
. is the overall mean for all observations
i=i–.is the effect of the ith factor level s.t.i ~ N0,  2 
ij is the random error component 
s.t. ij ~ N 0,  2

The hypothesis may now be rephrased:
H0 : ² = 0
Ha : ² > 0

Seagate Confidential 44 Supplier Six Sigma Modular Training

 Analysis Of Variance

One-Way ANOVA-Random
Model
Source SS DF MS E(MS) F-Statistic

MSA
Factor A SSA a–1 MSA  2  n 2A
MSE

Error SSE a(n–1) MSE 2

Total SSTO an–1

There is no difference in calculation of the F-Statistic between a

One-Way ANOVA-Fixed Model and a One-Way ANOVA-
Random Model . The difference lies in the conclusion.

Seagate Confidential 45 Supplier Six Sigma Modular Training

 Analysis Of Variance

One-Way ANOVA-Random
Example 2
A textile manufacturing company is interested in
loom-to-loom variability in tensile strength. The
engineer selects four looms at random and
determines the strength of four fabric samples
chosen at random from each loom.

The results may be found in Analysis Of

Variance.MTW.
Seagate Confidential 46 Supplier Six Sigma Modular Training
 Analysis Of Variance

One-Way ANOVA-Random
Model — Example 2
Stat  ANOVA  General Linear Model

Seagate Confidential 47 Supplier Six Sigma Modular Training

 Analysis Of Variance

One-Way ANOVA-Random
Model — Example 2
Stat  Regression  Residual Plots

Seagate Confidential 48 Supplier Six Sigma Modular Training

 Analysis Of Variance

One-Way ANOVA-Random
Model — Example 2
Residual Plots for One-Way ANOVA-II Model
Normal Plot of Residuals I Chart of Residuals
5
2 UCL=4.167
4
3
1 2

Residual
Residual

1
0 0 Mean=0
-1
-1 -2
-3
-4 LCL=-4.167
-2 -5
-2 -1 0 1 2 0 5 10 15
Normal Score Observation Number

Histogram of Residuals Residuals vs. Fits

4 2

3 1
Frequency

Residual

2 0

1 -1

0 -2
-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 91 92 93 94 95 96 97 98
Residual Fit

Seagate Confidential 49 Supplier Six Sigma Modular Training

 Analysis Of Variance

One-Way ANOVA-Random
Model — Example 2
Session Window

General Linear Model: Tensile versus Loom

Factor Type Levels Values

Loom random 4 1 2 3 4

Analysis of Variance for Tensile, using Adjusted SS for Tests

Source DF Seq SS Adj SS Adj MS F P

Loom 3 89.187 89.187 29.729 15.68 0.000
Error 12 22.750 22.750 1.896
Total 15 111.937

Seagate Confidential 50 Supplier Six Sigma Modular Training

 Analysis Of Variance

One-Way ANOVA-Random
Model — Example 2
Stat  Basic Statistics  Display Descriptive
Statistics

Seagate Confidential 51 Supplier Six Sigma Modular Training

 Analysis Of Variance

One-Way ANOVA-Random
Model — Example 2
Dotplots of Tensile by Loom

99

98

97

96
Tensile

95

94

93

92

91

90

Loom
1

Seagate Confidential 52 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed Model

Yijk  ..   i   j   ij   ijk
where Yijk is the observed value of the response variable
.. is the overall mean for all observations
i is the main effect of the ith level of fixed factor A
j is the main effect of the jth level of fixed factor B
ij is the interaction of the ith level of factor A with the jth level of
factor B
ijk is the random error component

such that and

  0
i i for j = 1,, 0
j jb

 j ij  0 for i = 1, , a
and

iij  0
 ij k ~ N0,  2 
Seagate Confidential 53 Supplier Six Sigma Modular Training
 Analysis Of Variance

Two-Way ANOVA-Fixed Model

The hypothesis under test are:
a)H0 : 1 = 2 =  = a = 0
Ha : not all i equal 0

b)H0 : 1 = 2 =  = b = 0
Ha : not all j equal 0

c) H0 : all ij = 0
Ha : not all ij equal 0

Seagate Confidential 54 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed Model

ANOVA Table

Source DF Sum of Sq Mean Square

Factor A a–1 SSA MSA = SSA (a – 1)

Factor B b–1 SSB MSB = SSB (b – 1)
AB-Interaction (a–1)(b–1) SSAB MSAB = SSAB [(a – 1) (b – 1)]
Error (n–1)ab SSE MSE = SSE (N–a)
Total N–1 SSTO

Seagate Confidential 55 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed Model

MS DF E(Mean Square) Under H0 F*

MSA a–1  2
 nb
 i
2

nb
 2
i
0
MSA
a 1 a 1 MSE

MSB b–1  2
 na
 2
j
na
 2j 0
MSB
b 1 b 1 MSE

MSAB (a–1)(b–1)   n
2   2
ij
n
  2
ij
0
MSAB
a  1b  1 a  1b  1 MSE

MSE (n–1)ab 2

Seagate Confidential 56 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed
Model — Example 3
A supplier of Italian bread to supermarkets
studied the effects of height of the shelf display
(factor A:bottom, middle, top) and width of the
shelf display (factor B: regular, wide) on the
sales of their bread. Twelve supermarkets were
assigned at random to obtain two replicates per
treatment. The results may be found in Analysis
Of Variance.MTW.
Seagate Confidential 57 Supplier Six Sigma Modular Training
 Analysis Of Variance

Two-Way ANOVA-Fixed
Model — Example 3
Stat  ANOVA  Two-way

Seagate Confidential 58 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed
Model — Example 3
Stat  Regression  Residual Plots

Seagate Confidential 59 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed
Model — Example 3
Residual Plots for Two-Way ANOVA-I Model
Normal Plot of Residuals I Chart of Residuals
3 UCL=5.319
5
2
8
1

Residual
Residual

0 0 Mean=0

-1
6
-2
-5 LCL=-5.319
-3
-1 0 1 0 5 10
Normal Score Observation Number

Histogram of Residuals Residuals vs. Fits

3 3
2
Frequency

2 1
Residual

0
1 -1
-2
0 -3
-3 -2 -1 0 1 2 3 40 50 60 70
Residual Fit

Seagate Confidential 60 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed
Model — Example 3
Session Window
Two-way ANOVA: Sales versus Height, Width

Analysis of Variance for Sales

Source DF SS MS F P
Height 2 1544.0 772.0 74.71 0.000
Width 1 12.0 12.0 1.16 0.323
Interaction 2 24.0 12.0 1.16 0.375
Error 6 62.0 10.3
Total 11 1642.0

Seagate Confidential 61 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed
Model — Example 3
Individual 95% CI
Height Mean ---+---------+---------+---------+--------
Bottom 44.0 (----*----)
Middle 67.0 (----*----)
Top 42.0 (---*----)
---+---------+---------+---------+--------
40.0 48.0 56.0 64.0

Individual 95% CI
Width Mean ---+---------+---------+---------+--------
Regular 50.0 (------------*------------)
Wide 52.0 (------------*------------)
---+---------+---------+---------+--------
47.5 50.0 52.5 55.0

Seagate Confidential 62 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed
Model — Example 3
Stat  ANOVA  Main Effects Plot

Seagate Confidential 63 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed
Model — Example 3
Main Effects Plot for Sales of Italian Bread

Height Width
71

63
Sales

55

47

39

Seagate Confidential 64 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed
Model — Example 3
Stat  ANOVA  Interactions Plot

Seagate Confidential 65 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Fixed
Model — Example 3
Interactions Plot for Sales of Italian Bread

Height

70 Bottom
Middle
Top

60
Mean

50

40

Regular Wide
Width

Seagate Confidential 66 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Random Model

Yijk  ..   i   j   ij   ijk

where Yijk is the observed value of the response variable

.. is the overall mean for all observations
i is the main effect of the ith level of random factor A
j is the main effect of the jth level of random factor B
ij is the interaction of the ith level of factor A with the
jth level of factor B
ijk is the random error component

 ~ N0,  2   ~ N0, 2   ~ N0,  

2
  ~ N0,  2 
such that , , and .

Seagate Confidential 67 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Random Model

The hypothesis under test are:
a)H0 : ² = 0
Ha : ² > 0
b)H0 : ² = 0
Ha : ² > 0
c) H0 : ² = 0
Ha : ² > 0

Seagate Confidential 68 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Random Model

ANOVA Table
Source DF Sum of Sq Mean Square

Factor A a–1 SSA MSA = SSA(a – 1)

Factor B b–1 SSB MSB = SSB(b – 1)
AB-Interaction (a–1)(b–1) SSAB MSAB = SSAB[(a – 1) (b – 1)]
Error (n–1)ab SSE MSE = SSE(N–a)

Total N–1 SSTO

Seagate Confidential 69 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Random Model

MS DF E(Mean Square) Under H0 F*

MSA
MSA a–1  2  nb 2  n 
2
nb 2  0
MSAB
MSB
MSB b–1  2  na2  n 
2
na2  0
MSAB
MSAB
MSAB (a–1)(b–1)   n 
2 2
n 
2
0
MSE

MSE (n–1)ab 2

Seagate Confidential 70 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Random
Model — Example 4
An automobile manufacturer wants to study the
effects of drivers (factor A) and cars (factor B) on
gasoline consumption. Four drivers were selected
at random; also five cars of the same model were
randomly selected from the assembly line.Each
driver drove the car twice over a certain test
course. The results may be found in Analysis Of
Variance.MTW.

Seagate Confidential 71 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Random
Model — Example 4
Stat  ANOVA  General Linear Model

Seagate Confidential 72 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Random
Model — Example 4
Stat  Regression  Residual Plots

Seagate Confidential 73 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Random
Model — Example 4
Residual Plots for Two-Way ANOVA-II Model
Normal Plot of Residuals I Chart of Residuals
4 UCL=6.528
3 5
2
1

Residual
Residual

0
0 Mean=0
-1
-2
-3
-4 -5
LCL=-6.528
-5
-2 -1 0 1 2 0 5 10
Normal Score Observation Number

Histogram of Residuals Residuals vs. Fits

2 4
3
2
Frequency

1
Residual

0
1
-1
-2
-3
-4
0 -5
-5 -4 -3 -2 -1 0 1 2 3 4 40 50 60 70
Residual Fit

Seagate Confidential 74 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Random
Model — Example 4
Session Window
General Linear Model: Gasoline versus Driver, Car

Factor Type Levels Values

Driver random 41234
Car random 512345

Analysis of Variance for Gasoline, using Adjusted SS for Tests

Source DF Seq SS Adj SS Adj MS F P

Driver 3 280.285 280.285 93.428 458.26 0.000
Car 4 94.714 94.714 23.678 116.14 0.000
Driver*Car 12 2.447 2.447 0.204 1.16 0.371
Error 20 3.515 3.515 0.176
Total 39 380.960

Seagate Confidential 75 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Random
Model — Example 4
Main Effects Plot for Gasoline Consumption

Driver Car
35.0

32.5
Gasoline

30.0

27.5

25.0

Seagate Confidential 76 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Random
Model — Example 4
Interactions Plot for Gasoline Consumption

Driver
1
2
35 3
4
Mean

30

25

1 2 3 4 5
Car

Seagate Confidential 77 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed Model

There are two forms of this model
 restricted model
– requires the crossed, mixed terms to sum to zero
– adopted by textbooks
 unrestricted model
– does not require the crossed, mixed terms to sum to zero
– adopted by statistical software (MiniTab, SAS, SPSS, JMP, etc)
The reasons to choose one form over the other have not been clearly
defined.
“ … that question really has no definitive, universally acceptable
answer …” – S.R. Searle, G Casella and C.E. McCulloch
Variance Components, Wiley & Sons (92)

Seagate Confidential 78 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model
(Unrestricted)
If factor A is a fixed effect and factor B is a random effect.
Source SS DF MS E(MS) F-Statistic


 2
MSA
Factor A SSA a–1 MSA  2 i
 n 
2

a 1 MSAB
MSB
Factor B SSB b–1 MSB  2  an2  n 
2

MSAB
AB Interaction SSAB (a–1)(b–1) MSAB MSAB
 2  n 
2

MSE
Error SSE ab(n–1) MSE 2

Total SSTO abn–1

Seagate Confidential 79 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model
(Restricted)
If factor A is a fixed effect and factor B is a random effect.
Source SS DF MS E(MS) F-Statistic

Factor A SSA a–1 MSA  2

 bn   i2
 n 2
MSA

a 1 MSAB
MSB
Factor B SSB b–1 MSB  2  an2
MSE
AB Interaction SSAB (a–1)(b–1) MSAB MSAB
 2  n 
2

MSE
Error SSE ab(n–1) MSE 2

Total SSTO abn–1

Seagate Confidential 80 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
A manufacturer of imitation pearls entailed studying the
effect of the number of coats of a special lacquer (factor
A) applied to a opalescent plastic bead on the market
value of the pearl. Four batches of 12 beads (factor B)
were used in the study. The three levels of factor A (6, 8
and 10 coats) were fixed in advance, while the four
batches were randomly selected from the bead
production process.
The results may be found in Analysis Of Variance.MTW.

Seagate Confidential 81 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
Stat  ANOVA  General Linear Model

Seagate Confidential 82 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
Stat  ANOVA  General Linear Model

Seagate Confidential 83 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
Stat  ANOVA  Residual Plots

Seagate Confidential 84 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
Residual Plots for Two-Way ANOVA-III Model
Normal Plot of Residuals I Chart of Residuals
3 10
2 UCL=7.100

Residual
Residual

0
-1 0 Mean=-8.9E-16

-2
-3
-4 LCL=-7.100

-5 -10
-2 -1 0 1 2 0 10 20 30 40 50
Normal Score Observation Number

Histogram of Residuals Residuals vs. Fits

10 3
2
1
Frequency

Residual

0
5 -1
-2
-3
-4
0 -5
-5 -4 -3 -2 -1 0 1 2 3 71 72 73 74 75 76 77 78 79
Residual Fit

Seagate Confidential 85 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA- Mixed

Model — Example 5
Session Window
General Linear Model: Value versus Coats, Batch

Factor Type Levels Values

Coats fixed 3 6 8 10
Batch random 41234

Analysis of Variance for Value, using Adjusted SS for Tests

Source DF Seq SS Adj SS Adj MS F P

Coats 2 150.388 150.388 75.194 243.60 0.000
Batch 3 152.852 152.852 50.951 165.06 0.000
Coats*Batch 6 1.852 1.852 0.309 0.06 0.999
Error 36 173.625 173.625 4.823
Total 47 478.717

Seagate Confidential 86 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
Stat  ANOVA  Balanced ANOVA

Seagate Confidential 87 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
Stat  ANOVA  Residual Plots

Seagate Confidential 88 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
Residual Plots for Two-Way ANOVA-III Model
Normal Plot of Residuals I Chart of Residuals
3 10
2 UCL=7.100

Residual
Residual

0
-1 0 Mean=-8.9E-16

-2
-3
-4 LCL=-7.100

-5 -10
-2 -1 0 1 2 0 10 20 30 40 50
Normal Score Observation Number

Histogram of Residuals Residuals vs. Fits

10 3
2
1
Frequency

Residual

0
5 -1
-2
-3
-4
0 -5
-5 -4 -3 -2 -1 0 1 2 3 71 72 73 74 75 76 77 78 79
Residual Fit

Seagate Confidential 89 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed Model —

Example 5
Session Window
ANOVA: Value versus Coats, Batch

Factor Type Levels Values

Coats fixed 3 6 8 10
Batch random 4 1 2 3 4

Analysis of Variance for Value

Source DF SS MS F P
Coats 2 150.388 75.194 243.60 0.000
Batch 3 152.852 50.951 10.56 0.000
Coats*Batch 6 1.852 0.309 0.06 0.999
Error 36 173.625 4.823
Difference in F*
Total 47 478.717
No difference in p-Value

Seagate Confidential 90 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
Stat  ANOVA  Main Effects Plot

Seagate Confidential 91 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
Main Effects Plot for Pearl Value

Coats Batch
80.0

77.5
Value

75.0

72.5

70.0

Seagate Confidential 92 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
Stat  ANOVA  Interactions Plot

Seagate Confidential 93 Supplier Six Sigma Modular Training

 Analysis Of Variance

Two-Way ANOVA-Mixed
Model — Example 5
Interactions Plot for Pearl Value

Coats

80 6
8
10
Mean

75

70

1 2 3 4
Batch

Seagate Confidential 94 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Fixed Model

Yijkm  ...   i   j   k   ij   ik   jk   ijk   ijkm

Y is the observed value of the response variable

... is the overall mean for all observations
 is the main effect of the fixed factor A   0
s.t.
 is the main effect of the fixed factor B   0
s.t
 is the main effect of the fixed factor C   0
s.t.
 is the interaction between factor A and factor B s.t.
 is the interaction between factor A and factor C s.t.
 is the interaction between factor B and factor C s.t.
 is the interaction between factors A, B and C s.t.
 is the random error component  k ~ N0,  
s.t.
ij
2

Seagate Confidential 95 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Fixed Models

MS DF E(Mean Square)

MSA a–1
 2  nbc   i2 a  1
MSB b–1
 2  nac   2j b  1
MSC c–1
 2  nab   2k c  1
MSAB (a–1)(b–1)

MSAC (a–1)(c–1)
 2  nc   ij2 a  1b  1

MSBC (b–1)(c–1)  2  nb   ik2 a  1c  1

MSABC (a–1)(b–1)(c–1)  2  na   2jk b  1c  1
MSE (n–1)abc
 2  n   ijk
2
a  1b  1c  1

2

Seagate Confidential 96 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Fixed Models

H0 Ha Test Statistic

all i=0 not all i=0 F*  MSA

MSE

all j=0 not all j=0 F*  MSB

MSE
all k=0 not all k=0 F*  MSC
MSE
all ij=0 not all ij=0
F*  MSAB
MSE
all ik=0 not all ik=0
F*  MSAC
MSE
all jk=0 not all jk=0
F*  MSBC
MSE
all iik=0 not all iik=0
F*  MSABC
MSE

Seagate Confidential 97 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Fixed
Model — Example 6
Assemblers in an electronics firm will attach 12
components to a new printed circuit board. A
study was performed to assess the effects of
gender (factor A), method of sequence (factor B)
and experience of operator (factor C) on the time
required to assemble 50 boards.
The results may be found in Analysis Of
Variance.MTW.
Seagate Confidential 98 Supplier Six Sigma Modular Training
 Analysis Of Variance

Three-Way ANOVA-Fixed
Model — Example 6
Stat  ANOVA  General Linear Model

Seagate Confidential 99 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Fixed
Model — Example 6
Stat  ANOVA  General Linear Model

Seagate Confidential 10 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Fixed
Model — Example 6
Stat  Regression  Residual Plots

Seagate Confidential 10 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Fixed
Model — Example 6
Residual Plots for Three-Way ANOVA-I Model
Normal Plot of Residuals I Chart of Residuals
50 100 UCL=90.63

Residual
Residual

0 0 Mean=-2.7E-14

-50 LCL=-90.63
-100
-2 -1 0 1 2 0 10 20 30 40 50 60
Normal Score Observation Number

Histogram of Residuals Residuals vs. Fits

50
10
Frequency

Residual

0
5

0 -50

-50 0 50 900 1000 1100 1200 1300

Residual Fit

Seagate Confidential 10 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Fixed
Model — Example 6
Session Window
General Linear Model: Time versus Gender, Sequence, Operator
Factor Type Levels Values
Gender fixed 2 Female Male
Sequence fixed 3 1 2 3
Operator fixed 2 Experienced On Training

Analysis of Variance for Time, using Adjusted SS for Tests

Source DF Seq SS Adj SS Adj MS F P

Gender 1 540361 540361 540361 629.76 0.000
Sequence 2 49320 49320 24660 28.74 0.000
Operator 1 382402 382402 382402 445.67 0.000
Gender*Sequence 2 542 542 271 0.32 0.730
Gender*Operator 1 91 91 91 0.11 0.746
Sequence*Operator 2 911 911 456 0.53 0.591
Gender*Sequence*Operator 2 19 19 10 0.01 0.989
Error 48 41186 41186 858
Total 59 1014832

Seagate Confidential 10 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Fixed
Model — Example 6
Main Effects Plot for Assembly Time

Gender Sequence Operator

1150

1100
Time

1050

1000

950

Seagate Confidential 10 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Fixed
Model — Example 6
Interactions Plot for Assembly Time

Gender 1200
Male

1050

Female 900

Sequence 1200
3

1050
2

1 900

Operator

Seagate Confidential 10 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Random
Model
Yijkm  ...   i   j   k   ij   ik   jk   ijk   ijkm
Y is the observed value of the response variable
... is the overall mean for all observations
 is the main effect of the random factor A s.t.  ~ N0,  2 
 is the main effect of the random factor B s.t  ~ N0,  
2

 is the main effect of the random factor C s.t.  ~ N0,  2 

 is the interaction between factor A and factor B s.t.  ~ N0,  2 
 is the interaction between factor A and factor C s.t.  ~ N0,  2 
 is the interaction between factor B and factor C s.t.  ~ N0,  2 

 is the interaction between factors A, B and C s.t.
 ~ N0,  2

 is the random error component s.t.
 ~ N0,  2


Seagate Confidential 10 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Random Models

MS DF E(Mean Square)

MSA a–1  2  nbc 2  nc 

2
 nb 
2
 n 
2

MSB b–1  2  nac2  nc 

2
 na2  n 
2

MSC c–1
 2  nab 2  nb 
2
 na2  n 
2

MSAB (a–1)(b–1)
 2  nc 
2
 n 
2

MSAC (a–1)(c–1)
 2  nb 
2
 n 
2

MSBC (b–1)(c–1)
 2  na2  n 
2

MSABC (a–1)(b–1)(c–1)

MSE (n–1)abc  2  n 

2

2

Seagate Confidential 10 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Random
Model
H0 Ha Test Statistic

² = 0 ² > 0 F*  MSA

MSAB  MSAC  MSABC
² = 0 ² > 0 F*  MSB
MSAB  MSAC  MSABC
² = 0 ² > 0 F*  MSC
MSAB  MSAC  MSABC
² = 0 ² > 0
F*  MSAB
MSABC
² = 0 ² > 0
F*  MSAC
MSABC
² = 0 ² > 0
F*  MSBC
MSABC
²= 0² > 0
F*  MSABC
MSE

Seagate Confidential 10 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Random
Model — Example 7
An engineer plans to assess the variation in lead
tip co-planarity for PQFPs.
The study involved 2 projectors, 3 inspectors and
10 specimens — all factors were randomly
selected. The results may be found in Analysis Of
Variance.MTW.

Seagate Confidential 10 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Random
Model — Example 7
Stat  ANOVA  General Linear Model

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Random
Model — Example 7
Stat  Regression  Residual Plots

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Random
Model — Example 7
Residual Plots for Three-Way ANOVA-II Model
Normal Plot of Residuals I Chart of Residuals
0.2
0.1
UCL=0.1542

0.1
777 77777 777 77

Residual
Residual

777 7
0.0 0.0 7 7 7777
Mean=1.04E-16

7 7 77
-0.1
LCL=-0.1542
-0.1 -0.2
-2 -1 0 1 2 0 50 100
Normal Score Observation Number

Histogram of Residuals Residuals vs. Fits

40 0.1

30
Frequency

Residual

20 0.0

10

0 -0.1
-0.10
-0.08
-0.06
-0.04
-0.02
0.000.020.040.060.080.10 0 1 2 3 4 5
Residual Fit

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Random
Model — Example 7
Session Window

General Linear Model: Coplanarity versus Projector, Inspector, Part

Analysis of Variance for Coplanar, using Adjusted SS for Tests

Source DF Seq SS Adj SS Adj MS F P

Projector 1 6.5801 6.5801 6.5801 4.65 0.060 x
Inspector 2 0.0302 0.0302 0.0151 **
Part 9 156.2651 156.2651 17.3628 12.22 0.001 x
Projector*Inspector 2 0.0082 0.0082 0.0041 0.25 0.778
Projector*Part 9 12.8558 12.8558 1.4284 89.12 0.000
Inspector*Part 18 0.1532 0.1532 0.0085 0.53 0.906
Projector*Inspector*Part 18 0.2885 0.2885 0.0160 2.63 0.003
Error 60 0.3650 0.3650 0.0061
Total 119 176.5459

x Not an exact F-test.

** Denominator of F-test is zero.

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Random
Model — Example 7
Main Effects Plot for Lead Tip Coplanarity

Projector Inspector Part

6.0

4.5
Coplanarity

3.0

1.5

0.0

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Random
Model — Example 7
Interactions Plot for Lead Tip Coplanarity

Projector 5.0
2

2.5

1 0.0

Inspector 5.0
3

2.5
2

1 0.0

Part

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
The effect of certain factors on the thickness of a
certain coating under qualification was assessed.
The factors were
a)Shift – morning and night
b) Machine Setting – 35, 44 and 52 sec
c) Operator – 3 operators from a large pool
The results may be found in Analysis Of
Variance.MTW.

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
Stat  ANOVA  Balanced ANOVA
(Unrestricted Model)

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
Stat  Regression  Residual Plots

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
Residual Plots for Three-Way ANOVA-III Model
Normal Plot of Residuals I Chart of Residuals
3 5 UCL=4.825
2
1

Residual
Residual

0 0 Mean=0

-1
-2
-3 -5 LCL=-4.825

-2 -1 0 1 2 0 10 20 30 40
Normal Score Observation Number

Histogram of Residuals Residuals vs. Fits

6 3
5 2
Frequency

4 1
Residual

3 0
2 -1
1 -2
0 -3
-2.5-2.0-1.5-1.0-0.50.0 0.5 1.0 1.5 2.0 2.5 40 50 60 70 80 90 100 110
Residual Fit

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
Session Window
ANOVA: Thickness versus Shift, Setting, Prod Operator
Analysis of Variance for Thickness

Source DF SS MS F P
Shift 1 9.0 9.0 0.29 0.644
Setting 2 15676.4 7838.2 73.18 0.001
Prod Operator 2 1120.9 560.4 4.91 0.090 x
Shift*Setting 2 114.5 57.3 2.39 0.208
Shift*Prod Operator 2 62.0 31.0 1.29 0.369
Setting*Prod Operator 4 428.4 107.1 4.46 0.088
Shift*Setting*Prod Operator 4 96.0 24.0 7.08 0.001
Error 18 61.0 3.4
Total 35 17568.2

x Not an exact F-test.

Seagate Confidential 12 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
ANOVA: Thickness versus Shift, Setting, Prod Operator
Source Variance Error Expected Mean Square for Each Term
component term (using unrestricted model)
1 Shift 5 (8) + 2(7) + 6(5) + Q[1,4]
2 Setting 6 (8) + 2(7) + 4(6) + Q[2,4]
3 Prod Operator 37.194 * (8) + 2(7) + 4(6) + 6(5) + 12(3)
4 Shift*Setting 7 (8) + 2(7) + Q[4]
5 Shift*Prod Operator 1.167 7 (8) + 2(7) + 6(5)
6 Setting*Prod Operator 20.778 7 (8) + 2(7) + 4(6)
7 Shift*Setting*Prod Operator 10.306 8 (8) + 2(7)
8 Error 3.389 (8)

Error Terms for Synthesized Tests

Source Error DF Error MS Synthesis of Error MS
3 Prod Operator 3.73 114.1 (5) + (6) - (7)

Seagate Confidential 12 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
H0 Ha Test Statistic

all i = 0 not all i = 0 F*  MSA

MSAC

all j = 0 not all j = 0 F*  MSB

MSBC

² = 0 ² > 0 F*  MSC  MSC

MSE Syn MSAC  MSBC  MSABC
all ij = 0 not all ij = 0
F*  MSAB
MSABC
² = 0 ² > 0
F*  MSAC
MSABC
² = 0 ² > 0
F*  MSBC
MSABC
² = 0 ² > 0
F*  MSABC
MSE

Seagate Confidential 12 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
Stat  ANOVA  Balanced ANOVA
(Restricted Model)

Seagate Confidential 12 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
Session Window

ANOVA: Thickness versus Shift, Setting, Prod Operator

Analysis of Variance for Thickness

Source DF SS MS F P
Shift 1 9.0 9.0 0.29 0.644
Setting 2 15676.4 7838.2 73.18 0.001
Prod Operator 2 1120.9 560.4 165.38 0.000
Shift*Setting 2 114.5 57.3 2.39 0.208
Shift*Prod Operator 2 62.0 31.0 9.15 0.002
Setting*Prod Operator 4 428.4 107.1 31.61 0.000
Shift*Setting*Prod Operator 4 96.0 24.0 7.08 0.001
Error 18 61.0 3.4
Total 35 17568.2

Seagate Confidential 12 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
ANOVA: Thickness versus Shift, Setting, Prod Operator

Source Variance Error Expected Mean Square for

Each Term
component term (using restricted model)
1 Shift 5 (8) + 6(5) + 18Q[1]
2 Setting 6 (8) + 4(6) + 12Q[2]
3 Prod Operator 46.421 8 (8) + 12(3)
4 Shift*Setting 7 (8) + 2(7) + 6Q[4]
5 Shift*Prod Operator 4.602 8 (8) + 6(5)
6 Setting*Prod Operator 25.931 8 (8) + 4(6)
7 Shift*Setting*Prod Operator 10.306 8 (8) + 2(7)
8 Error 3.389 (8)

Seagate Confidential 12 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
H0 Ha Test Statistic

all i = 0 not all i = 0 F*  MSA

MSAC
all j = 0 not all j = 0 F*  MSB
MSBC
² = 0 ² > 0
F*  MSC
MSE
all ij = 0 not all ij = 0
F*  MSAB
MSABC
² = 0 ² > 0
F*  MSAC
MSE
² = 0 ² > 0
F*  MSBC
MSE
² = 0 ² > 0
F*  MSABC
MSE

Seagate Confidential 12 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
Session Window

Analysis of Variance (Balanced Designs)

Analysis of Variance for Thickness

Restricted Model Unrestricted Model

Source F P F P
Shift 0.29 0.644 0.29 0.644
Setting 73.18 0.001 73.18 0.001
Prod Operator 165.38 0.000 4.91 0.090
Shift*Setting 2.39 0.208 2.39 0.208
Shift*Prod Operator 9.15 0.002 1.29 0.369
Setting*Prod Operator 31.61 0.000 4.46 0.088
Shift*Setting*Prod Operator 7.08 0.001 7.08 0.001
Error 3.4

Seagate Confidential 12 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
Stat  ANOVA  Main Effects Plot

Seagate Confidential 12 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
Main Effects Plot for Coating Thickness

Shift Setting Prod Oper

110

90
Thickness

70

50

30

Seagate Confidential 12 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
Stat  ANOVA  Interactions Plot

Seagate Confidential 13 Supplier Six Sigma Modular Training

 Analysis Of Variance

Three-Way ANOVA-Mixed
Model — Example 8
Interactions Plot for Coating Thickness

Shift 100
Night
75

50

Morning

Setting 100
52
75

44
50

35

Prod Oper 100

3
75

2
50

Seagate Confidential 13 Supplier Six Sigma Modular Training

 Analysis Of Variance

End of Topic
What question do you have?

Quote:
Statistical Significance does not
always mean Practical Significance
Seagate Confidential 13 Supplier Six Sigma Modular Training