Factorial Designs

Prof. Pritam Ranjan

Email: [email protected]
https://sites.google.com/site/drpritamranjan
 Factorial Designs

Big Picture
• CRD: 1 treatment factor with p levels + no blocking factors
• RBD: 1 treatment factor with p levels + 1 blocking factor, b blocks of size p each
• BIBD: 1 treatment factor with p levels + 1 blocking factor, b blocks of size k (< p) each
• LSD: 1 treatment factor with p levels + 2 blocking factors, p blocks of size p each
• GLSD: 1 treatment factor with p levels + 3 blocking factors, p blocks of size p each

• Objective: to test H! : 𝜏" = 𝜏# = ⋯ = 𝜏$ = 0 (all treatment levels are same)

Prof. Pritam Ranjan / OM&QT Area 2

 Factorial Designs

Big Picture
• CRD: 1 treatment factor with p levels + no blocking factors
• RBD: 1 treatment factor with p levels + 1 blocking factor, b blocks of size p each
• BIBD: 1 treatment factor with p levels + 1 blocking factor, b blocks of size k (< p) each
• LSD: 1 treatment factor with p levels + 2 blocking factors, p blocks of size p each
• GLSD: 1 treatment factor with p levels + 3 blocking factors, p blocks of size p each

• Objective: to test H! : 𝜏" = 𝜏# = ⋯ = 𝜏$ = 0 (all treatment levels are same)

• What if we have more than 1 treatment factors?

Prof. Pritam Ranjan / OM&QT Area 3

 Factorial Designs

Factorial design
Simplest 2 – treatment factor design

2 treatment factors with 2 levels each + No blocking factors

i.e., All experimental units are homogeneous
i.e., Design: 2! factorial CRD

Prof. Pritam Ranjan / OM&QT Area 4

 Factorial Designs

Factorial design
Simplest 2 – treatment factor design

2 treatment factors with 2 levels each + No blocking factors

i.e., All experimental units are homogeneous
i.e., Design: 2! factorial CRD

A factorial design means that in each complete trial or replicate of the

experiment, all possible combinations of the levels of the factors are
investigated. That is, factor A has 2 levels and B has 2 levels, then each
replicate contains all 2×2 = 4 treatment combinations
{ (A-,B-), (A+,B-), (A-,B+), (A+,B+) } ó { (1), A, B, AB }

Prof. Pritam Ranjan / OM&QT Area 5

 Factorial Designs

Factorial design
Simplest 2 – treatment factor design

2 treatment factors with 2 levels each + No blocking factors

i.e., All experimental units are homogeneous
i.e., Design: 2! factorial CRD

A factorial design means that in each complete trial or replicate of the

experiment, all possible combinations of the levels of the factors are
investigated. That is, factor A has 2 levels and B has 2 levels, then each
replicate contains all 2×2 = 4 treatment combinations.
{ (A-,B-), (A+,B-), (A-,B+), (A+,B+) } ó { (1), A, B, AB }

Objective: which treatment factors and factor combinations are significant?

Prof. Pritam Ranjan / OM&QT Area 6

 Factorial Designs

Example

Prof. Pritam Ranjan / OM&QT Area 7

 Factorial Designs

2! Factorial design
Example:

Prof. Pritam Ranjan / OM&QT Area 8

 Factorial Designs

2! Factorial design
Example:

Main effect of A:
40 + 52 20 + 30
𝑦'%& − 𝑦'%' = − = 21
2 2

Prof. Pritam Ranjan / OM&QT Area 9

 Factorial Designs

2! Factorial design
Example:

Main effect of A:
40 + 52 20 + 30
𝑦'%& − 𝑦'%' = − = 21
2 2

Main effect of B:
30 + 52 20 + 40
𝑦'(& − 𝑦'(' = − = 11
2 2

Prof. Pritam Ranjan / OM&QT Area 10

 Factorial Designs

2! Factorial design
Example:

Main effect of A:
40 + 52 20 + 30
𝑦'%& − 𝑦'%' = − = 21
2 2

Main effect of B:
30 + 52 20 + 40
𝑦'(& − 𝑦'(' = − = 11
2 2

Interaction effect of AB:

20 + 52 30 + 40
𝑦'%(& − 𝑦'%(' = − =1
2 2

Prof. Pritam Ranjan / OM&QT Area 11

 Factorial Designs

2! Factorial design
Example:

Main effect of A: 21
Main effect of B: 11
Interaction effect of AB: 1

Main effect of A:
Main effect of B:
Interaction effect of AB:

Prof. Pritam Ranjan / OM&QT Area 12

 Factorial Designs

2! Factorial design
Example:

Main effect of A: 21
Main effect of B: 11
Interaction effect of AB: 1

Main effect of A:
Main effect of B:
Interaction effect of AB:

Prof. Pritam Ranjan / OM&QT Area 13

 Factorial Designs

Regression model

Prof. Pritam Ranjan / OM&QT Area 14

 Factorial Designs

2! Factorial design
Example:
Assuming treatment factors are quantitative, define
𝑥" = −1 𝐴 𝑙𝑜𝑤 ; 𝑥" = +1 (𝐴 ℎ𝑖𝑔ℎ).
𝑥# = −1 𝐵 𝑙𝑜𝑤 ; 𝑥# = +1 (𝐵 ℎ𝑖𝑔ℎ).

The regression model

Prof. Pritam Ranjan / OM&QT Area 15

 Factorial Designs

2! Factorial design
Example:
Assuming treatment factors are quantitative, define
𝑥" = −1 𝐴 𝑙𝑜𝑤 ; 𝑥" = +1 (𝐴 ℎ𝑖𝑔ℎ).
𝑥# = −1 𝐵 𝑙𝑜𝑤 ; 𝑥# = +1 (𝐵 ℎ𝑖𝑔ℎ).

The regression model

Response surface plot

Prof. Pritam Ranjan / OM&QT Area 16

 Factorial Designs

Example

Prof. Pritam Ranjan / OM&QT Area 17

 Factorial Designs

3! Factorial design
Example: an engineer is designing a battery that will be subjected to extreme
variations in temperature and has three choices for the plate material.

Client’s question: Is there a choice of material that would give uniformly long life
regardless of temperature?

Prof. Pritam Ranjan / OM&QT Area 18

 Factorial Designs

3! Factorial design
ANOVA model : CRD with two treatment factors

Prof. Pritam Ranjan / OM&QT Area 19

 Factorial Designs

3! Factorial design
Residual analysis

Prof. Pritam Ranjan / OM&QT Area 20

 Factorial Designs

3! Factorial design
ANOVA model

Prof. Pritam Ranjan / OM&QT Area 21

 Factorial Designs

3! Factorial design
Multiple comparison
One approach to this situation is to fix factor B at a specific level and apply Tukey’s
test to the means of factor A at that level.

The pairwise comparisons yield

Prof. Pritam Ranjan / OM&QT Area 22

 Factorial Designs

Exercise

Prof. Pritam Ranjan / OM&QT Area 23

 Factorial Designs

Exercise

Prof. Pritam Ranjan / OM&QT Area 24

 Factorial Designs

Exercise

Prof. Pritam Ranjan / OM&QT Area 25

 Factorial Designs

Exercise

Prof. Pritam Ranjan / OM&QT Area 26

 Factorial Designs

#
More on 2 factorial designs

Prof. Pritam Ranjan / OM&QT Area 27

 Factorial Designs

Setup
• No blocking factors – all experimental units are homogenous
• Errors are normally distributed

• Objective: identify active factors and factor combinations

Prof. Pritam Ranjan / OM&QT Area 28

 Factorial Designs

Basics of 𝟐𝒌 factorial CRD

• Factor coding (DESIGN matrix)

• Geometric coding (+ / - ): AB is the product of A and B

• 0-1 coding: AB (interaction effect) is the sum of A and B modulo 2

Prof. Pritam Ranjan / OM&QT Area 29

 Factorial Designs

Basics of 𝟐𝒌 factorial CRD

• Factor coding (MODEL matrix)

$ '("' (" $
Main effect of A: 𝑦!!" − 𝑦!!# = %!"# & 𝐴, 𝑇𝑟 = %&
− %&
$
Main effect of B: 𝑦!)" − 𝑦!)# = %!"# & ⟨𝐵, 𝑇𝑟⟩
$
Main effect of AB: 𝑦!!)" − 𝑦!!)# = ⟨𝐴𝐵, 𝑇𝑟⟩
%!"# &
Prof. Pritam Ranjan / OM&QT Area 30
 Factorial Designs

Basics of 𝟐𝒌 factorial CRD

• Factor coding (Yates’ order) (MODEL matrix)

$ '"'("'*"'(* $ "("*"(*
Main effect of A: 𝑦!!" − 𝑦!!# = %!"# & 𝐴, 𝑇𝑟 = +&
− +&
$
Main effect of B: 𝑦!)" − 𝑦!)# = %!"# & ⟨𝐵, 𝑇𝑟⟩
$
Main effect of AB: 𝑦!!)" − 𝑦!!)# = ⟨𝐴𝐵, 𝑇𝑟⟩
%!"# &
Prof. Pritam Ranjan / OM&QT Area 31
 Factorial Designs

Example – 6.1

Prof. Pritam Ranjan / OM&QT Area 32

 Factorial Designs

Example – 6.1

Find the significant factors

Prof. Pritam Ranjan / OM&QT Area 33

 Factorial Designs

Example – 6.1
Ø lm_fit=lm(etch~A+C+A:C, data=dat)
Ø summary(lm_fit)

# Prediction
yhat = predict(lm_fit)
res = rstandard(lm_fit)

# Diagnostic plots
plot(etch,yhat)
plot(yhat,res); abline(h=0)
qqnorm(res);qqline(res)

Prof. Pritam Ranjan / OM&QT Area 34

 Factorial Designs

Example – 6.1

# Diagnostic plots
plot(etch,yhat)
plot(yhat,res); abline(h=0)
qqnorm(res);qqline(res)

Prof. Pritam Ranjan / OM&QT Area 35

 Factorial Designs

#
Unreplicated 2 factorial
designs

Prof. Pritam Ranjan / OM&QT Area 36

 Factorial Designs

Example – 6.2
• A factorial experiment is carried out in a chemical plant to
study the factors thought to influence the filtration rate of a
chemical product. The factors are temperature (A), pressure
(B), concentration of formaldehyde (C), and stirring rate (D).

Prof. Pritam Ranjan / OM&QT Area 37

 Factorial Designs

Example – 6.2
• Data analysis results

Prof. Pritam Ranjan / OM&QT Area 38

 Factorial Designs

Example – 6.2
• How to test the significance of
factors and factor interactions?

1. Half normal plot

2. Length’s test
3. Permutation test

Prof. Pritam Ranjan / OM&QT Area 39

 Factorial Designs

Half-normal plot
• How to test the significance of factors and factor interactions?

1. Half normal plot - a plot of the absolute value of the effect estimates
against their cumulative normal probabilities

2. Length’s test
3. Permutation test

Ø library(DoE.base)
Ø halfnormal(lm_fit, ME.partial = TRUE)

Prof. Pritam Ranjan / OM&QT Area 40

 Factorial Designs

Example-6.2 - diagnostics

Prof. Pritam Ranjan / OM&QT Area 41

 Factorial Designs

Example-6.2

• Further model refinement – Should we drop CD and ACD?

Prof. Pritam Ranjan / OM&QT Area 42

 Factorial Designs

Principles of Model building

• Effect Sparsity / Occam’s razor - law of parsimony
• The simplest explanation is usually the right one
• The model with fewer parameters is preferred
• The model is usually dominated by main effects and low-order interactions

• Effect Hierarchy
• Lower order effects are more important than higher order effects

• Effect Heredity
• AB is included only if both A and B are already in the model (Strong)
• AB is included only if at least one of A and B are in the model (Weak)

Prof. Pritam Ranjan / OM&QT Area 43

 Factorial Designs

Example-6.2
• Find a good final response surface model (R – illustration).

Prof. Pritam Ranjan / OM&QT Area 44

 Factorial Designs

Example-6.2

• Goodness of fit measures

? ?
• Look at both 𝑅<=> and 𝑅@ABCC

• PRESS - prediction error sum of squares (sum of the squared prediction

errors obtained by predicting the i-th data point with a model that
includes all observations except the i-th one.)

Prof. Pritam Ranjan / OM&QT Area 45

 Factorial Designs

Example – 6.6
• An experiment is conducted to test new ideas to increase
direct mail sales by the credit card division of a financial
services company The factors tested in the experiment are

Prof. Pritam Ranjan / OM&QT Area 46

 Factorial Designs

Practice Problems

Prof. Pritam Ranjan / OM&QT Area 47

 Factorial Designs

Exercises

Prof. Pritam Ranjan / OM&QT Area 48

 Factorial Designs

Exercises

Prof. Pritam Ranjan / OM&QT Area 49

 Factorial Designs

Exercises

Prof. Pritam Ranjan / OM&QT Area 50

 Factorial Designs

Exercises

• (6.22) The effect estimates from a 24 factorial design are as follows: ABCD =
−1.5138, ABC = −1.2661, ABD =−0.9852, ACD = −0.7566, BCD = −0.4842, CD =
−0.0795, BD = −0.0793, AD = 0.5988, BC = 0.9216, AC = 1.1616, AB = 1.3266, D =
4.6744, C = 5.1458, B = 8.2469, and A = 12.7151. Are you comfortable with the
conclusions that all main effects are active?

Prof. Pritam Ranjan / OM&QT Area 51

 Factorial Designs

Exercises

Prof. Pritam Ranjan / OM&QT Area 52

 Factorial Designs

Thank You

Questions ?

Reference – Chapters 5, 6

Next Topic
• Blocking and Confounding in Factorial Designs

Prof. Pritam Ranjan / OM&QT Area 53