Design and Analysis
of Experiments
Chapter 10
Experiments with
Random Factors
Dr. Tran Thanh Hung
Department of Automation Technology,
College of Engineering, Can Tho University
Email: tthung@[Link]
Chapter objectives
At the end of this chapter, students can:
• Understand the concept of random effect
• Be familiar with random effect models
and components of variance in each
model
• Learn how to deal with models containing
two random factors
• Know how to analyze experiments where
one of the factors is fixed and the other
one is random
Experiments with Random Factors:
Introduction
• Text reference, Chapter 13, p. 573
• Previous chapters have considered fixed
factors
– A specific set of factor levels is chosen for the
experiment
– Inference confined to those levels
– Often quantitative factors are fixed
• When factor levels are chosen at random from a
larger population of potential levels, the factor is
random
– Inference is about the entire population of levels
– Industrial applications include measurement system
studies 3
Random Effects Models,
example 3.11: weaving fabric on looms
• Response variable is strength
• Interest focuses on determining if there is difference in
strength due to the different looms
• However, the weave room contains many (100s) looms
• Solution – select a (random) sample of the looms, obtain
fabric from each
• Consequently, “looms” is a random factor
It looks like standard single-factor
experiment with a = 4 & n = 4
4
Random Effects Models
• The usual single factor ANOVA model is
i 1, 2,..., a (random levels)
yij i ij
j 1, 2,..., n (replications)
• Now both the error term and the treatment
effects are random variables:
ij NID(0, 2 ) , i NID(0, 2 )
(NID: Normally and Independently Distribution)
• Variance of any observation:
V ( yij ) , where
2
2
2 V ( ij ), 2 V ( i ) :variance components 5
Relevant Hypotheses in the Random
Effects
• In the fixed effects model we test equality of
treatment means
• This is no longer appropriate because the
treatments are randomly selected
– the individual ones we happen to have are not
of specific interest
– we are interested in the population of treatments
• The appropriate hypotheses are
H 0 : 2 0
H1 : 2 0
6
Testing Hypotheses –
Random Effects Model
• The standard ANOVA partition of the total
sum of squares still works; leads to usual
ANOVA display
• Form of the hypothesis test depends on
the expected mean squares
E ( MS E ) 2 , E ( MSTreatments ) 2 n 2
• Therefore, the appropriate test statistic is
F0 MSTreatments / MS E
• The reference distribution for F0 is the
Fa-1, a(n-1) 7
Estimating the Variance
Components
• Use the ANOVA method; equate expected
mean squares to their observed values:
ˆ 2 MS E , ˆ 2 nˆ2 MSTreatments
2 MSTreatments MS E
ˆ
n
ˆ 2 MS E
• Potential problems with these estimators
– Negative estimates
– They are moment estimators & don’t have best
statistical properties
8
Example 3.11: Minitab Solution
(Balanced ANOVA)
Factor Type Levels Values
Loom random 4 1 2 3 4
Analysis of Variance for y
Source DF SS MS F P
Loom 3 89.188 29.729 15.68 0.000
Error 12 22.750 1.896
Total 15 111.938
Source Variance Error Expected Mean Square for Each Term
component term (using unrestricted model)
1 Loom 6.958 2 (2) + 4(1)
2 Error 1.896 (2)
9
Extension to Factorial Treatment
Structure
• Suppose that we have two factors, A and B, and both factors
have a large number of levels that are of interest
• We will choose at random a levels of factor A and b levels of
factor B and arrange these factor levels in a factorial
experimental design and n observations are made at each
treatment combination
• We may represent the observations by the linear model:
i 1, 2,..., a
yijk i j ( )ij ijk j 1, 2,..., b
k 1, 2,..., n
V ( i ) 2 , V ( j ) 2 , V [( )ij ] 2 , V ( ijk ) 2
V ( yijk ) 2 2 2 2
The model parameters are normal distribution random variables
10
Testing Hypotheses - Random Effects
Model
• Once again, the standard ANOVA partition is
appropriate
• Relevant hypotheses:
H 0 : 2 0 H 0 : 2 0 H 0 : 2 0
H1 : 2 0 H1 : 2 0 H1 : 2 0
• Form of the test statistics depend on the expected
mean squares:
MS A
E ( MS A ) 2 n 2 bn 2 F0 , Fa -1,( a -1)(b -1)
MS AB
MS B
E ( MS B ) 2 n 2 an 2 F0 , Fb-1,( a -1)(b-1)
MS AB
MS AB
E ( MS AB ) 2 n 2 F0 , F( a -1)(b -1),ab ( n 1)
MS E
11
E ( MS E ) 2
Estimating the Variance Components
– Two Factor Random model
• As before, use the ANOVA method; equate
expected mean squares to their observed values:
2 MS A MS AB
ˆ
bn
MS A ˆ 2 nˆ2 bnˆ2
ˆ 2 MS B MS AB
MS B ˆ 2 nˆ2 anˆ 2
an
MS AB ˆ 2 nˆ2 2 MS AB MS E
ˆ
MS E ˆ 2 n
ˆ 2 MS E
• Potential problems with these estimators
12
Example 13.1
A Measurement Systems Capability Study
• Gauge capability (or Repeatability &
Reproducibility(R&R)) is of interest
• The gauge is used by an operator to measure a critical
dimension on a part
• Repeatability is a measure of the variability due only
to the gauge
• Reproducibility is a measure of the variability due to
the operator
• See experimental layout, Table 13.1. This is a two-
factor factorial (completely randomized) with both
factors (operators, parts) random – a random effects
model
13
Example 13.1: data
Variance components
• Linear model and total variability:
yijk i j ( )ij ijk
V ( yijk ) y2 2 2 2 2
2 is called the gauge repeatability
2 2 is called the reproducibility of the gauge
Example 13.1
Minitab Solution – Using Balanced ANOVA
16
Example 13.1
Minitab Solution – Balanced ANOVA
• There is a large effect of parts (not
unexpected)
• Small operator effect
• No Part – Operator interaction
• Negative estimate of the Part – Operator
interaction variance component
• Fit a reduced model with the Part –
Operator interaction deleted
yijk i j ijk
17
Example 13.1
Minitab Solution – Reduced Model
18
Example 13.1
Minitab Solution – Reduced Model
• Estimating gauge capability:
ˆ gauge
2
ˆ 2 ˆ 2
0.88 0.01
0.89
• This is relatively small comparing to the
variability of the product.
19
The Two-Factor Mixed Model
• Consider the case that one of the factors is fixed, for
example A, and the other one (B) is a random factor.
i 1, 2,..., a
yijk i j ( )ij ijk j 1, 2,..., b
k 1, 2,..., n
V ( j ) 2 , V [( )ij ] [(a 1) / a] 2 , V ( ijk ) 2
a a
i 1
i 0, ( )
i 1
ij 0
• The model parameters and are NID random
variables, the interaction jeffect isijknormal, but not
independent
• This is called the restricted model (giới hạn) 20
Testing Hypotheses - Mixed Model
• Once again, the standard ANOVA partition is appropriate
• Relevant hypotheses:
H0 : i 0 H 0 : 2 0 H 0 : 2 0
H1 : i 0 H1 : 2 0 H1 : 2 0
• Test statistics depend on the expected mean squares:
a
bn i2
MS A
E ( MS A ) 2 n 2 i 1
F0 , Fa -1,( a -1)(b -1)
a 1 MS AB
MS B
E ( MS B ) 2 an 2 F0 , Fb -1,ab ( n-1)
MS E
MS AB
E ( MS AB ) 2 n 2 F0 F( a -1)(b-1),ab ( n -1)
MS E 21
E ( MS ) 2
Estimating the Variance Components
– Two Factor Mixed model
• It is possible to estimate the fixed factor effects as
ˆ y... , ˆi yi .. y... , i 1, 2,..., a
• Use the ANOVA method; equate expected mean
squares to their observed values:
2 MS B MS E
ˆ
MS B ˆ 2 anˆ 2
an
2 MS AB MS E
MS AB ˆ 2 nˆ2 ˆ
n
MS E ˆ 2 ˆ 2 MS E
• Estimate the fixed effects (treatment means) as
usual 22
Example 13.2
The Measurement Systems Capability Study
Revisited
• Same experimental setting as in example
13.2
• Parts are a random factor, but Operators
are fixed (3 persons)
• Assume the restricted form of the mixed
model
• Minitab can analyze the mixed model
23
Example 13.2
Minitab Solution – Balanced ANOVA
DOX 6E Montgomery 24
Example 13.2
Minitab Solution – Balanced ANOVA
• There is a large effect of parts
• Small operator effect
• No Part – Operator interaction
• Negative estimate of the Part – Operator
interaction variance component
• Fit a reduced model with the Part – Operator
interaction deleted
• This leads to the same solution that we found
previously for the two-factor random model
25
The Unrestricted Mixed Model
• Two factors, factorial experiment, factor A fixed,
factor B random
i 1, 2,..., a
yijk i j ( )ij ijk j 1, 2,..., b
k 1, 2,..., n
V ( j ) 2 , V [( )ij ] 2 , V ( ijk ) 2
a
i 1
i 0
• The random model parameters are now all
assumed to be NID
26
Testing Hypotheses – Unrestricted Mixed Model
• The standard ANOVA partition is appropriate
• Relevant hypotheses:
H 0 : i 0 H 0 : 2 0 H 0 : 2 0
H1 : i 0 H1 : 2 0 H1 : 2 0
• Expected mean squares determine the test statistics:
a
bn i2
MS A
E ( MS A ) 2 n 2 i 1
F0
a 1 MS AB
MS B
E ( MS B ) 2 n 2 an 2 F0
MS AB
MS AB
E ( MS AB ) 2 n 2 F0
MS E
27
E ( MS E ) 2
Estimating the Variance Components
– Unrestricted Mixed Model
• Use the ANOVA method; equate expected mean
squares to their observed values:
2 MS B MS AB
MS B ˆ 2 nˆ2 anˆ 2 ˆ an
MS AB ˆ 2 nˆ2 2 MS AB MS E
ˆ
n
MS E ˆ 2
ˆ 2 MS E
28
Example 13.3
Minitab Solution – Unrestricted Model
DOX 6E Montgomery 29
Finding Expected Mean Squares
• Obviously important in determining the form of the
test statistic
• In fixed models, it’s easy:
E ( MS ) 2 f (fixed factor)
• Can always use the “brute force” approach – just
apply the expectation operator
• Straightforward but tedious
• Rules on page 589-591 work for any balanced model
• Rules are consistent with the restricted mixed model
– can be modified to incorporate the unrestricted
model assumptions
30
Finding Expected Mean Squares