BSC 311: Design and Analysis of Experiments First Semester 2021/22 Academic Year

Lecture 9: Multi-way Factorial Analysis of Variance - Three-factor analysis of variance

Note: This lecture will not be taught in class. It is for your-need-to-know-basis only just in case
you come across such experiments in your research. We will do this using a computer.

Introduction

The multi-way factorial analysis of variance is performed by expanding the considerations of the
two-way analysis of variance to assess the effects on a variable of the simultaneous application
of three or more factors.

It is not unreasonable to expect somebody to perform a one-way or two-way analysis by hand

although computer programs are frequently employed, especially when the experiment consists
of a large number of observations.

However, it has become uncommon for analyses of variance with more than two factors to be
analyzed other than on a computer, owing to considerations of time, ease and accuracy.

Therefore, this lecture will not instruct the student in the computational mechanics of multi-way
analyses of variance. Rather, it will presume that established computer programs will be used to
perform the necessary mathematical manipulations.

Example 1. We want to determine the effect of species, temperature and sex on the rate of
oxygen consumption of crabs. The tables below show the results of the study:

Species 1
Low temperature Medium temperature High temperature
♂ ♀ ♂ ♀ ♂ ♀
1.9 1.8 2.3 2.4 2.9 3.0
1.8 1.7 2.1 2.7 2.8 3.1
1.6 1.4 2.0 2.4 3.4 3.0
1.4 1.5 2.6 2.6 3.2 2.7

1
Lecture 9 Wilson Wesley Lazaro Jere
BSC 311: Design and Analysis of Experiments First Semester 2021/22 Academic Year

Species 2
Low temperature Medium temperature High temperature
♂ ♀ ♂ ♀ ♂ ♀
2.1 2.3 2.4 2.0 3.6 3.1
2.0 2.0 2.6 2.3 3.1 3.0
1.8 1.9 2.7 2.1 3.4 2.8
2.2 1.7 2.3 2.4 3.2 3.2

Species 3
Low temperature Medium temperature High temperature
♂ ♀ ♂ ♀ ♂ ♀
1.1 1.4 2.0 2.4 2.9 3.2
1.2 1.0 2.1 2.6 2.8 2.9
1.0 1.3 1.9 2.3 3.0 2.8
1.4 1.2 2.2 2.2 3.1 2.9

Hypotheses

i. Hypotheses for main effects

𝐻𝑜 : Mean respiratory rate is the same in all three crab species (i.e., 𝜇1 = 𝜇2 = 𝜇3 ).

𝐻𝑎 : Mean respiratory rate is not the same in all three crab species.

𝐻𝑜 : Mean respiratory rate of crabs is the same at all three experimental temperatures (i.e.,
𝜇𝑙𝑜𝑤 = 𝜇𝑚𝑒𝑑 = 𝜇ℎ𝑖𝑔ℎ ).

𝐻𝑎 : Mean respiratory rate of crabs is not the same at all three experimental temperatures.

𝐻𝑜 : Mean respiratory rate is the same for males and females crabs (i.e., 𝜇♂ = 𝜇♀ ).

𝐻𝑎 : Mean respiratory rate is not the same for males and females crabs (i.e., 𝜇♂ ≠ 𝜇♀).

2
Lecture 9 Wilson Wesley Lazaro Jere
BSC 311: Design and Analysis of Experiments First Semester 2021/22 Academic Year

ii. Hypotheses for two-way interactions

𝐻𝑜 : Differences in mean respiratory rate among the three species of crabs are independent of
(i.e., are the same at) the three experimental temperatures; or, differences in mean
respiratory rate among the three temperatures are independent of (i.e., are the same in)
the three species. (Testing for Species x Temperature interaction).

𝐻𝑎 : Differences in mean respiratory rate among the species of crabs are not independent of
the experimental temperatures.

𝐻𝑜 : Differences in mean respiratory rate among the three species of crabs are independent of
sex (i.e., are the same for both sexes); or, differences in mean respiratory rate between
males and females are independent of (i.e., are the same in) the three species. (Testing for
Species x Sex interaction).

𝐻𝑎 : Differences in mean respiratory rate among the species of crabs are not independent of
the sex.

𝐻𝑜 : Differences in mean respiratory rate among the three experimental temperatures are
independent of (i.e., are the same in) the two sexes of crabs; or, differences in mean
respiratory rate between males and females are independent of (i.e., are the same at) the
three experimental temperatures. (Testing for Temperature x Sex interaction).

𝐻𝑎 : Differences in mean respiratory rate among the three temperatures are not independent of
the sex of crabs.

iii. Hypotheses for three-way interaction

𝐻𝑜 : Differences in mean respiratory rate among the three species of crabs (or temperatures, or
sexes) are independent of the other two factors (Testing for Species x Temperature x Sex
interaction).

𝐻𝑎 : Differences in mean respiratory rate among the three species of crabs (or temperatures, or
sexes) are not independent of the other two factors

3
Lecture 9 Wilson Wesley Lazaro Jere
BSC 311: Design and Analysis of Experiments First Semester 2021/22 Academic Year

By consulting a computer output, we can prepare the following summary

We can examine the ANOVA tables by looking at the tabulated F values or the p-values given
by the computer.

A. Examining the calculated F values

Source of Sums of Degrees of Mean Calculated Critical F

Variation Squares freedom Squares F*
1.8175 2 0.9088 24.4751 𝐹0.05(1),2,54 ≅ 3.18
Species (Factor A)
Temperature 24.6558 2 12.3279 332.0237 𝐹0.05(1),2,54 ≅ 3.18
(Factor B)
Sex (Factor C) 0.0098 1 0.0089 0.2394 𝐹0.05(1),1,54 ≅ 4.03
AxB 1.1017 4 0.2754 7.4177 𝐹0.05(1),4,54 ≅ 2.56
AxC 0.3703 2 0.1851 4.9863 𝐹0.05(1),2,54 ≅ 3.18
BxC 0.1753 2 0.0876 2.3603 𝐹0.05(1),2,54 ≅ 3.18
AxBxC 0.2206 4 0.0551 1.4850 𝐹0.05(1),4,54 ≅ 2.56
Error 2.0050 54 0.0371
*There is no critical values for error degrees of freedom = 54, so the values for the next lower
degrees of freedom, i.e., 50, were utilized.

Decisions and interpretations

When making decisions, we start by examining the higher-order interaction, in this case, the
three-way interaction.

If the higher order interaction is significant, we do not examine the lower order interactions or
main effects.

If a factor is involved in any interaction, we don't examine its main effect.

4
Lecture 9 Wilson Wesley Lazaro Jere
BSC 311: Design and Analysis of Experiments First Semester 2021/22 Academic Year

a) Three-way interaction: interaction between species, temperature and sex (A x B x C)

Since the calculated F value is less than the tabulated F value, we fail to reject the null
hypothesis. We conclude that there is sufficient evidence that there is no significant interaction
among the three factors (species, temperature and sex) on the mean respiratory response of crabs
at 5 % level of significance.

Since the three-way interaction is not significant, we can go ahead to examine the two-way
interactions.

b) Two-way interactions

i. Interaction between temperature and sex (B x C)

Since the calculated F value is less than the tabulated F value, we fail to reject the null
hypothesis. We conclude that there is sufficient evidence that there is no significant interaction
between temperature and sex on the mean respiratory response of crabs at 5 % level of
significance.

ii. Interaction between species and sex (A x C)

Since the calculated F value is greater than the tabulated F value, we reject the null hypothesis.
We conclude that there is sufficient evidence that there is significant interaction between species
and sex on the mean respiratory response of crabs at 5 % level of significance.

iii. Interaction between species and temperature (A x B)

Since the calculated F value is greater than the tabulated F value, we reject the null hypothesis.
We conclude that there is sufficient evidence that there is significant interaction between species
and temperature on the mean respiratory response of crabs at 5 % level of significance.

c) Main effects

Since the three main effects are all involved in significant interactions, we cannot examine their
main effects.

5
Lecture 9 Wilson Wesley Lazaro Jere
BSC 311: Design and Analysis of Experiments First Semester 2021/22 Academic Year

B. Examining the calculated p values

We can also read out the p-values from the computer output.

Source of Sums of Degrees of Mean Calculated p-values

Variation Squares freedom Squares F*
1.8175 2 0.9088 24.4751 0.0000
Species (Factor A)
Temperature 24.6558 2 12.3279 332.0237 0.0000
(Factor B)
Sex (Factor C) 0.0098 1 0.0089 0.2394 0.6266
AxB 1.1017 4 0.2754 7.4177 0.0000
AxC 0.3703 2 0.1851 4.9863 0.0103
BxC 0.1753 2 0.0876 2.3603 0.1041
AxBxC 0.2206 4 0.0551 1.4850 0.2196
Error 2.0050 54 0.0371

a) Three-way interaction: interaction between species, temperature and sex (A x B x C)

Since the p-value is more than 0.05, we fail to reject the null hypothesis. We conclude that there
is sufficient evidence that there is no significant interaction among the three factors (species,
temperature and sex) on the mean respiratory response of crabs at 5 % level of significance.

Since the three-way interaction is not significant, we can go ahead to examine the two-way
interactions.

b) Two-way interactions

i. Interaction between temperature and sex (B x C)

Since the p-value is more than 0.05, we fail to reject the null hypothesis. We conclude that there
is sufficient evidence that there is no significant interaction between temperature and sex on the
mean respiratory response of crabs at 5 % level of significance.

6
Lecture 9 Wilson Wesley Lazaro Jere
BSC 311: Design and Analysis of Experiments First Semester 2021/22 Academic Year

ii. Interaction between species and sex (A x C)

Since the p-value is less than 0.05, we reject the null hypothesis. We conclude that there is
sufficient evidence that there is significant interaction between species and sex on the mean
respiratory response of crabs at 5 % level of significance.

iii. Interaction between species and temperature (A x B)

Since the p-value is less than 0.05, we reject the null hypothesis. We conclude that there is
sufficient evidence that there is significant interaction between species and temperature on the
mean respiratory response of crabs at 5 % level of significance.

c) Main effects

Since the main effects are all involved in significant interactions, we cannot examine the main
effects.

Note

1. The test for a two-way interaction asks whether differences in the variable among levels of
one factor are the same at all levels of the second factor.
2. A test for a three-factor interaction may be thought of asking if the interaction between any
two of the factors is the same at all levels of the third factor.

7
Lecture 9 Wilson Wesley Lazaro Jere
BSC 311: Design and Analysis of Experiments First Semester 2021/22 Academic Year

Statistical model for a three-way factorial experiment

The model for a three-way ANOVA is as follows:

𝑌𝑖𝑗𝑘𝑙 = 𝜇 + 𝛼𝑖 + 𝛽𝑗 + 𝛾𝑘 + 𝛼𝛽𝑖𝑗 + 𝛼𝛾𝑖𝑘 + 𝛽𝛾𝑗𝑘 + 𝛼𝛽𝛾𝑖𝑗𝑘 + 𝜀𝑖𝑗𝑘𝑙

Where:

𝑌𝑖𝑗𝑘𝑙 = lth response associated with ith level of factor A, jth level of factor B and kth
level of factor C.

𝜇 = overall mean

𝛼𝑖 = ith effect of factor A

𝛽𝑗 = jth effect of factor B

𝛾𝑘 = kth effect of factor C

𝛼𝛽𝑖𝑗 = interactive effect of ith level of factor A and jth level of factor B

𝛼𝛾𝑖𝑘 = interactive effect of ith level of factor A and kth level of factor C

𝛽𝛾𝑗𝑘 = interactive effect of jth level of factor B and kth level of factor C

𝛼𝛽𝛾𝑖𝑗𝑘𝑙 = interactive effect of ith level of factor A, jth level of factor B and kth

level of factor C

𝜀𝑖𝑗𝑘 = kth random error associated with ith level of factor A and jth level of factor B.

8
Lecture 9 Wilson Wesley Lazaro Jere