POLYTECHNIC UNIVERSITY OF THE PHILIPPINES

COLLEGE OF ENGINEERING
DEPARTMENT OF INDUSTRIAL ENGINEERING

MODULE 4
DESIGN OF EXPERIMENTS WITH SEVERAL FACTORS

An experiment is basically a test or series of tests; they are

performed in all engineering and scientific disciplines and are an
important part of the way we learn about how systems and
processes work. The design of the experiment plays a pivotal role
in the eventual solution of the problem because the validity of the
conclusions that are drawn from it depends largely on how the
experiment was conducted.

We have already studied single-factor experiments and the use of

ANOVA as a tool to obtain sound conclusions. In this module, we
will focus on experiments that include two or more factors that
may have effects on a certain output.

FACTORIAL EXPERIMENTS

When several factors are to be studied in an experiment, a factorial

experiment design should be used. A factorial experiment is
defined as an experiment wherein all possible combinations of
the levels of the factors are investigated in each complete trial
or replicate.

The effect of a factor is defined as the change in response

produced by a change in the level of the factor. It is called a main
effect because it refers to the primary factors in the study.

Consider the figure below, which shows the different effects of

different combinations of levels of two factors, A and B:

Page 1 of 15
To read the figure below, we designate the negative sign as low
and the positive sign as high. Thus, a low level of Factor A and a
low level of Factor B will give us a result of 10, etc.

The main effect of Factor A is the difference between the average

response at the high level of A and the average response at the low
level of A:

Average response at
high level Average response at
low level

This means that changing Factor A from the low level to the high
level causes an average response increase of 20 units.

N.B. Can you prove given the data above that the main effect of
Factor B is 10?

In some experiments, the difference in response between the levels

of one factor is not the same at all levels of the other factors. When

Page 2 of 15
this occurs, this means that there is an interaction between the
factors. Consider the data given in the table below:

Factor B
Factor A Low High
Low 10 20
High 30 0

At the low level of Factor B, the Factor A effect is

A = 30 10 = 20
and at the high level of B, the A effect is

A = 0 20 = 20
Since the effect of Factor A depends on the level chosen for Factor
B, there is interaction between factors A and B. Thus, knowledge
of the AB interaction is more useful than knowledge of the main
effect. When an interaction is present, the main effects of the
factors involved in the interaction may not have much meaning.

To estimate the interaction effect in factorial experiments, we find

the difference in the diagonal averages:

Factor B
Factor A Low High
Low 10 20
High 30 0

20 + 30 10 + 0
AB = = 20
2 2

Page 3 of 15
The concept of interaction can be illustrated graphically in several
ways. One such way is a two-factor interaction plot, similar to the
figures below. They are often useful in presenting the results of
experiments.

Facotrial Experiment, Without Interaction

50
40
Observation

30 B High
20 B Low
10
0
Low High
Factor A

Factorial Experiment, With Interaction

35
30
Observation

25
20 B High
15 B Low
10
5
0
Low High
Factor A

It is important to note that factorial experiments are the only

way to discover interactions between variables.

Page 4 of 15
TWO-FACTOR FACTORIAL EXPERIMENTS

The simplest type of factorial experiment involves only two

factors, say, A and B. There are a levels of Factor A and b levels of
Factor B. The experiment has n replicates and each replicate
contains all ab treatment combinations.

Factor B
1 2 b

1 y111, y112, y121, y122, y1b1, y1b2,

., y11n ., y12n ., y1bn

2 y211, y212, y221, y222, y2b1, y2b2,

Factor A

., y21n ., y22n ., y2bn

.
.
.

a ya11, ya12, ya21, ya22, yab1, yab2,

., ya1n ., ya2n ., yabn

In performing the experiment, the abn observations would be run

in random order. Thus, like the single-factor experiment, the two-
factor factorial is a completely randomized design.

The observations may be described by the linear statistical model

a parameter associated with the

interaction of the ith and jth factor
levels of factors A and B i = 1, 2 , ,a
y ijk = + i + j + ( )ij + ijk j = 1, 2 , ,b
k = 1, 2 , ,n
a parameter associated
a parameter associated with the jth factor level of
a common parameter to all with the ith factor level of factor B called the jth
factors called the overall factor A called the ith factor effect
mean factor effect

Page 5 of 15
Statistical Analysis of the Fixed-Effects Model

Suppose that Factors A and B are fixed factors. ANOVA can be

used to test hypotheses about the main factor effects of A and B
and the AB interaction:

The hypotheses that will be tested are as follows:

1. H 0 : 1 = 2 = =a = 0
H 1 : at least one i 0
2. H 0 : 1 = 2 = = b = 0
H 1 : at least one j 0
3. H 0 : ( )1 = ( )2 = = ( )ab = 0
H 1 : at least one ( )ij 0

Page 6 of 15
The ANOVA table is constructed as follows:

EXAMPLE # 1: Aircraft primer paints are applied to aluminum

surfaces by two methods: dipping and spraying. The purpose of the
primer is to improve paint adhesion, and some parts can be primed
using either application method. The process engineering group
responsible for this operation is interested in learning whether
three different primers differ in their adhesion properties.

Application Method
Primer Dipping Spraying
1 4.0 4.5 4.3 5.4 4.9 5.6
2 5.6 4.9 5.4 5.8 6.1 6.3
3 3.8 3.7 4.0 5.5 5.0 5.0

A factorial experiment was performed to investigate the effect of

paint primer type and application method on paint adhesion. For
each combination of primer type and application method, three
specimens were painted, then a finish paint was applied, and the

Page 7 of 15
adhesion force was measured. The data are shown above. Do the
individual factors, i.e. primer type and application method, and/or
their interaction have an effect on the adhesion strength?

STEP 1: Declare your hypotheses

1. H 0 : 1 = 2 = =a = 0
H 1 : at least one i 0
2. H 0 : 1 = 2 = = b = 0
H 1 : at least one j 0
3. H 0 : ( )1 = ( )2 = = ( )ab = 0
H 1 : at least one ( )ij 0

STEP 2: State your level of confidence = 0.05

STEP 3: State the appropriate model Two-Factor Factorial

Experiment, Completely Randomized Design Fixed-Effects Model

STEP 4: Set up your ANOVA table:

Degrees
Sum of of Mean
Source of Variation Squares Freedom Square F0
Primer Types 4.58 2 2.29 28.63
Application Method 4.91 1 4.91 61.38
Interaction 0.24 2 0.12 1.50
Error 0.99 12 0.08
Total 10.72 17

Page 8 of 15
STEP 5: Declare your critical regions

For Primer Type: f 0 > f 0.05, 2 ,12 f 0 > 3.89

For Application Method: f 0 > f 0.01,1,12 f 0 > 4.75

For the Interaction: f 0 > f 0.05, 2 ,12 f 0 > 3.89

STEP 6: State your results

For Primer Type: Reject the null hypothesis

For Application Method: Reject the null hypothesis

For the Interaction: Fail to reject the null hypothesis

STEP 7: State your conclusion

Both primer type and application method have significant

individual effects on adhesion strength. Their interaction though
does not have a significant effect on adhesion strength.

GENERAL FACTORIAL EXPERIMENTS

Many experiments though involve more than two factors. Let us

take into consideration first the situation wherein there are three
factors to be studied.

ANOVA Table for a 3-Factor Experiment:

Source
Sum of Degrees of Mean
of F0
Squares Freedom Square
Variation

Page 9 of 15
 A SSA a-1 MSA MSA/MSE
B SSB b-1 MSB MSB/MSE
C SSC c-1 MSC MSC/MSE
AB SSAB (a-1)(b-1) MSAB MSAB/MSE
AC SSAC (a-1)(c-1) MSAC MSAC/MSE
BC SSBC (b-1)(c-1) MSBC MSBC/MSE
ABC SSABC (a-1)(b-1)(c-1) MSABC MSABC/MSE
Error SSE abc(n-1) MSE
Total SST abcn-1

Notice that for a 3-factor experiment, all possible effect

combinations are tested, i.e. we test the significance of one-factor
alone (A, B, C); two-factor interactions (AB, AC, BC); and three-
factor interactions (ABC). The ANOVA table gets bigger as the
number of factors to be studied increases.

EXAMPLE # 2: A mechanical engineer is studying the surface

roughness of a part produced in a metal-cutting operation. Three
factors, feed rate (A), depth of cut (B), and tool angle (C), are of
interest. All three factors have been assigned two levels, and two
replicates of a factorial design are run. Data are shown
immediately after this text.

Depth of Cut
0.025 inch 0.040 inch
Tool Angle Tool Angle
Feed Rate 15o 25o 15o 25o
9 11 9 10
20 in. per min.
7 10 11 8
10 10 12 16
30 in. per min.
12 13 15 14

Page 10 of 15
 Degrees
Source of Sum of Mean
of F0
Variation Squares Square
Freedom

Feed 45.56 1 45.563 18.693

Depth 10.56 1 10.563 4.334
Angle 3.063 1 3.063 1.257
Feed*
7.563 1 7.563 3.103
Depth
Feed*
0.062 1 0.062 0.025
Angle
Depth*
1.563 1 1.563 0.641
Angle
Feed*
Depth* 5.062 1 5.062 2.077
Angle
Error 19.499 8 2.437
Total 92.938 15

At = 0.05, and with every degree of freedom at 1, the critical

value of f 0.05,1,8 = 5.32 , we reject the null hypothesis for Feed rate
and therefore conclude that Feed rate alone has a significant effect
on the surface roughness.

2k FACTORIAL DESIGNS

Factorial designs are frequently used in experiments involving

several factors where it is necessary to study the joint effect of the
factors on a response. Some special cases of the general factorial
design are important because they are widely used in research
work and because they form the basis of other designs.

Page 11 of 15
Perhaps one of the most (if not the most) important of these special
cases is that of k factors, with each factor only having two levels.

22 Design

The simplest type of the 2k design is the 22, i.e. two factors A and
B, each at two levels (high and low levels):

High effects of both

factors A and B
High effect of
factor B; low effect
of factor A

High effect of
factor A; low effect
of factor B
Low effects of both
factors A and B

Special notations are used for the generic 2k design:

1. Low and High levels are denoted by the - and +

respectively.
2. A treatment combination is represented by a series of
lowercase letters. If a letter is present, the corresponding
factor is run at the high level in that treatment combination; if
it is absent, the factor is run at its low level.

Page 12 of 15
To compute for the different sum of squares for a 22 experiment,
we have the following equations:

SS A =
[ a + ab b (1)]
2

4n

SS B =
[b + ab a (1)]
2

4n

SS AB =
[ ab + (1) a b]
2

4n

with n corresponding to the number of replicates. The numerator,

e.g. for SSA a + ab b (1), is what we term as the contrast. In
general, the equation for a sum of squares for the different
interactions is equal to:

SS =
( contrast )
2

n2 k
In addition, the average effect of each interaction as it changes
from a low level to a high level is computed as:

contrast
Effect =
n2 k 1

Page 13 of 15
EXAMPLE # 3: An article in the AT&T Technical Journal
describes the application of two-level factorial designs to
integrated circuit manufacturing. A basic processing step in this
industry is to grow an epitaxial layer on polished silicon wafers.
The wafers are mounted on a susceptor and positioned inside a bell
jar. Chemical vapors are introduced through nozzles near the top of
the jar. The susceptor is rotated, and heat is applied. These
conditions are maintained until the epitaxial layer is thick enough.
Data collected are shown in the table below:

Thickness (in micrometers)

Replicate
Treatment
Combination 1 2 3 4
(1) 14.037 14.165 13.972 13.907
a 14.821 14.757 14.843 14.878
b 13.880 13.860 14.032 13.914
ab 14.888 14.921 14.415 14.932

The ANOVA table is given as:

Source of Sum of Degrees of Mean

F0
Variation Squares Freedom Square
A (deposition time) 2.796 1 2.796 134.458
B (arsenic flow) 0.018 1 0.018 0.871
AB 0.004 1 0.004 0.192
Error 0.250 12 0.021
Total 3.067 15

At = 0.05, and with every degree of freedom at 1, the critical

value of f 0.05,1,12 = 4.75 , we reject the null hypothesis for

Page 14 of 15
Deposition time and therefore conclude that Deposition time alone
has a significant effect on the epitaxial layer thickness.

The estimates of the effects are as follows:

A (deposition
0.836
time)
B (arsenic
0.067
flow)
AB 0.032

The values of the estimates of the effects are consistent with the
conclusion we obtained from our ANOVA. As can be seen from
above, the effect of deposition time is much more than the effect o
arsenic flow and their interaction.

EXERCISE: Construct the ANOVA table for the data presented in

Example # 2 using the generic 2k model equations.

Page 15 of 15