The Two-Factor Analysis of Variance You may have noticed that some people seem to work by the motto,
me people seem to work by the motto, “If at first you
14.1 INTRODUCTION
Analysis of Variance allows us to observe simultaneously the effects of many levels
of one independent variable on the dependent variable.
One-Way ANOVA is used to compare a number of groups that represent different
levels of a single independent variable. Also, it frees us from the limitations of the
two-group experiment (e.g., experimental versus control group design) and the use
of many t-ratios for one set of data.
Two-Factor or Two-Way Analysis of Variance is a statistical technique that allows us
to examine simultaneously the effects of two independent variables on a specific
behavior and the complex conditions that influence behavior.
14.2 LOGIC OF THE TWO-WAY ANOVA
The two-way ANOVA allows us to determine how much of the total variance is
explained by each of the independent variables by themselves and how much
variance is explained by combinations of the variables.
There are two different independent variables: qualitative or quantitative.
Qualitative variables differ in kind. A researcher would want to know which of
several teaching methods is most effective in helping illiterate adults learn to read
or which type of psychotherapy is most effective for the treatment of people who
batter their spouse or significant other.
Quantitative variables differ in “how much”. For example, testing the effects of a
new drug, it can vary in the dosage and to examine the rate of recovery.
Empirical Research on Behavior.
Why is it that some people, when faced with a difficult problem, work hard at
solving the problem, whereas other people seem to give up quickly?
don’t succeed, try, try again, ‘whereas others seem to work by the motto, “Never
throw good money after bad.”
In other words, we want to know why some people are more persistent than
others. What are some of the factors that influence how persistent a person will
be?
Chances are that you can think of many such factors that influence how persistent a
person will be, such as,
1. The characteristics of the task (hard versus easy);
2. The skill level of the task (skill versus luck is required);
3. The outcome of completing the task (much reward versus little reward);
4. It could also be individual differences. Some people may find the task
interesting, whereas others find it boring. Some people may believe that
they have a particular skill that will allow them to succeed, whereas others
may believe that they are not qualified to complete the task.
Sanderlands, Brocker, & Glynn (1988) conducted an experiment that examined
persistence. In the experiment, they had subjects attempt to solve a long list of
anagrams (e.g. WOREP is POWER). Some of the anagrams had solutions; the others
were impossible to solve. The dependent variable was the amount of time spent on
the anagrams. The subjects differed in terms of how much time they spent on each
unsolvable anagram. Some subjects devoted much time to working on each
anagram, whereas others spent much less time trying to solve each problem.
Sanderlands, Brocker, & Glynn used two independent variables in their study. The
first variable was the self-esteem of the subjects. The researchers selected subjects
with high or low self-esteem to participate in the study, with two levels of self-
esteem: high or low. The second variable was the type of instruction the subjects
received, with two levels of instruction: persist or don’t persist. Half of the subjects
were told that all the anagrams could be solved and that persistence on each
problem was a good strategy. The researchers told the remaining subjects that
some of the anagrams could not be solved and that giving up on a difficult problem
was a good strategy.
The variability among the scores of all the subjects in all the groups represents the high self-esteem. Also, half of all the subjects are told to persist, whereas the other
total variation. Using the two-way ANOVA, we can determine how much of the half are told not to persist.
variance is systematically related to the independent variables of the experiment
Table 14.2
and how much is due to random and unidentified effects.
A Representation of the Subscript Notation for a Two-Way ANOVA
Table 14.1

Illustration of a Two-Way ANOVA and the Assignment of Subjects to Treatment X ijk

Conditions
i=Individual observation
FACTOR A (Self-Esteem)
j=Factor A : Level of self −esteem
FACTOR B Low High
(Type of k =Factor B :Type of instruction
Instruction)
Persist Low self-esteem High self-esteem Subjects X ij Xj Xk
subjects subjects instructed to
Sum of all values for a cell, Sum of all values for Sum of all values for
instructed to instructed to persist;
persist; persist; n=30 the j th treatment subjects at a level of subjects at a level of
, factor A factor B
n=15 n=15 condition of factor A, and
, , X¿ ¿ k th treatment
X 11 X 12 the
condition of factor B
Don’t Persist Low self-esteem High self-esteem Subjects
subjects subjects instructed not to
instructed not to instructed not to persist; Table 14.2 illustrates the subscript notation of the two-way ANOVA. First, the design
persist; persist; n=30 , of the experiment allows us to look at all possible combinations of the variables we
n=15 n=15
, , X¿ ¿ chose to study. That is, we study subjects with high or low self-esteem under
X 21 X 22 conditions of instructions to either persist or not to persist.
Subjects with low Subjects with N=60 , A second important feature of the two-way ANOVA is that we can use different
self-esteem; high self-esteem;
n=30 n=30
X __ types of independent variables. Self-esteem is a subject variables and type of
, , instruction is a manipulated variable. Here again is the value of the two-way
X¿ 1 X¿2 ANOVA: we are able to combine in one experiment any pattern of independent
variables. In this experiment, we combined a subject and a manipulated variable.
Recall that we can also have combination of quantitative and qualitative
Table 14.1 illustrates the two factors and the assignment of the subjects to the independent variables. Be sure that you recognize, however, that the dependent
treatment groups. For the entire experiment there are 60 subjects. Half are variable must always represent an interval or ratio scale.
identified as having low self-esteem, whereas the other half are classified as having
Figure 14.1 X ijk = individual observation within a specific treatment condition. The i in
Illustration of the two-factor analysis of variance. The total variation is partitioned
the subscript represents the individual observation, the j represents the level of
into two general categories, between-group variation and within-group variation.
The between group variation is further subdivided into effects due to factor A, the first factor, and the k represents the level of the second factor.
factor B, and the interaction of factors A and B.
μ= population mean for the base population
Effects of factor
A: Self-esteem i effects of a specific level of the first treatment factor

βk= effects of a specific level of the second treatment factor

Between-group Effects of factor B:
variaton Instructions
αβ jk = effects of the interaction of the two factors at specific levels
Total Variation
Joint effects of
Within-group
variation
factors A and B ε ijk = undifferentiated error assumed to be random with a mean effect of 0
Interaction
Equation 14.1 will allow us to estimate the degree to which persistence is
Figure 14.1 illustrates the logic of the steps we will take with the two-way ANOVA.
influenced by self-esteem (represented as α in the equation), type of instruction
Observe that the total variation among subjects is partitioned into two general
components, the between-group variation and the within-group variation. The (represented as β in the equation), and the interaction of these two variables
within-group variation represents the variation among subjects that cannot be
(represented as αβ ).
predicted or accounted for the independent variables. Therefore, the within-group
variation represents sampling error. The between-group variation represents 14.3 ADVANTAGES OF THE TWO-WAY ANOVA
variance that can be attributed to systematic differences in the independent
variables. Specifically, the between-group variation consists of the effects due to 1. Two-way ANOVA allow us to examine the effects of more than one independent
each of the independent factors and the interaction of these factors. variable on the dependent variable.

Equation 14.1 The general linear model to represent the logic of the two-way 2. Two-way ANOVA are more cost-effective than multiple one-way ANOVAs.
ANOVA. 3. Two-way ANOVA tend to increase power and improve the generalizability of the
data.
X ijk =μ+α j + β k +αβ jk + ε ijk
4. Two-way ANOVA allow us to look for the presence of an interaction between
Where: independent variables.

Figure 14.2
Experiment 1 One-way ANOVA comparison of low and high self-esteem where the 14.4.1 THE CONCEPT OF A TREATMENT COMBINATION
sample size is 60.
One of the advantages of the two-way ANOVA is that it allows us to examine
Low Self-Esteem High Self-Esteem Total simultaneously the effects of two independent variables and their combined effects.
n = 30 n = 30 N = 60 When we design a two-factor experiment, we select the number of levels that we
want to use for each factor. Because we are combining the two factors into one
experimental design, we are creating a factorial design. In a factorial design, each
Experiment 2 One-way ANOVA comparison of instructions to persist or not where
treatment condition represents the combination of each factor, and the number of
the sample size is 60.
treatment conditions equals all the possible combinations of the factors.
Persist Don’t Persist Total Table 14.3
n = 30 n = 30 N = 60
Model of a 2×2 Factorial Design
Experiment 3 Two-way ANOVA comparing levels of self-esteem and instructtions Factor B Factor A (Self-Esteem) Row
where the sample size is 60. (Instructions)
Low (
a1 ) High (
a2 )
Total
Factor B Factor A (Self-Esteem) Row a1 b1 a2 b1 n = 30
(Instructions) Low High Total
Persist n = 15 n = 15 n = 30 Persist (
b1 )
X 111 X 121
Don’t Persist n = 15 n = 15 n = 30 X 211 X 221
Column Total n = 30 n = 30 N = 60
X 311 X 321
X 411 X 421
Figure 14.2 Representation of the advantages of the two-way ANOVA. Experiments ¿ ¿
represent simple one-way ANOVAs for each factor, self-esteem and instructions. ¿ ¿
Together, both experiments require 60 subjects. In contrast, Experiment 3 requires ¿ ¿
only 60 subjects and allows one to examine each main effect and the joint effects of
the two variables. X n 11 X n 21
X ¿ 11 X ¿ 21 X ¿⋅1
a 1 b2 a2 b 2
14.4 FACTORIAL DESIGNS: FACTORS, LEVEL, AND CELLS
Don’t Persist (
b2 X 112 X 122
A factorial design is a research design in which each treatment condition represents ) X 212 X 222
the combination of all levels of each factor. The number of treatment conditions for
a two-way ANOVA equals the number of levels for Factor A multiplied by the
X 312 X 322
number of levels for Factor B. X 412 X 422
¿ ¿
 ¿ ¿ Main effect means the effect of one of the factors without regard to the other
¿ ¿ factor and the interaction. It is based on the means of all subjects who are exposed
X n 12 X n 22 to a common level of a factor.
Column Total X ¿ 12 X ¿ 22 X ¿⋅2 In all two-way ANOVAs, there are two main effects to be examined, the effect of
Factor A and the effect of Factor B. In the current example, one main effect is
X ¿ 1⋅¿ ¿ X ¿ 2⋅¿ ¿ X ¿⋅¿ attributed to the subject’s self-esteem and the other is due to the instructions. Each
main effect can be either statistically significant.
Table 14.3 has four cells or treatment combinations. Each cell represents a different
Four Potential Patterns of Main Effects:
treatment combination. For example, cell
a1 b1
represents a different
1. A significant main effect for factor A,
treatment combination of low self-esteem and instructions to persist, whereas cell
2. A significant main effect for factor B,
a1 b2 represents the treatment combination of high self-esteem and 3. A simultaneous significant main effect for factors A and B, and
instructions not to persist. Therefore, a cell represents a specific treatment 4. A not significant main effect for factors A and B.
combination of the factors used in the experiment.
Significant Main Effect for Factor A: Self-Esteem
Factorial designs can vary in complexity from the very simple 2×2 design to Figure 14.3 represents data where there is a significant difference between levels of
more complex designs such as a 3×5 or a 4×6 factorial. self-esteem. There is no significant difference between the persist and don’t persist
instruction conditions, and there is no interaction between the variables.

14.5 TREATMENT COMBINATIONS: Main Effects and Interaction
The whole purpose of the two-way ANOVA is to examine how the two variables
combine and possibly interact with one another to produce the effects of the
experiment. Specifically, what it means by main effects and interactions.
Main Effect represents conditions wherein one or both of the factors have a
significant effect on the dependent variable.
Interaction represents a condition where both factors have an effect on the
dependent variable but the effect differs across the treatment combinations.
People with lower self-esteem spent less time on the task (an average of 20
seconds) than people with high self-esteem (an average of 40 seconds). The type of
instruction had no influence on the results. Subjects told to persist worked as long
on the task (30 seconds) as people told to give up and move along (also 30
seconds). We can conclude from these data that the differences in persistence
among subjects are due primarily to differences in self-esteem.
Figure 14.3
An example of a significant main effect for Factor A. Note that in the data and in the
graph subjects in
a1 are consistently less than subjects in
a2 . There are no

differences between subjects in conditions

b1 and
b2 .

14.5.1 MAIN EFFECTS

 m e a n ti m e s p e n t o n e a c h a n a g r a m ( s e c
M e a n ti m e s p e n t o n e a c h a n a g r a m ( s e
50 50
40 40
30 30
20 20
10 10
0 0
Low High Low High
Self-Esteem Self-Esteem

Hypothetical mean scores in each cell for this figure: Hypothetical mean scores in each cell for this figure:

Instruction Self-Esteem Mean Instruction Self-Esteem Mean

Low (
a1 ) High (
a2 )
a1 a2
Low ( ) High ( )
Persist ( )
b1 20 40 30 b1 40 40 40
Persist ( )
Don’t Persist ( 20 40 30 b
Don’t Persist ( 2
20 20 20
b2) )
Mean 20 40 30 Mean 30 30 30
Figure 14.4 Represent a condition where there is a significant main effect for the
type of instructions but no main effect for self-esteem or interaction. Specifically,
Significant Main Effect for Factor B: Instructions
these data suggest that people were less persistent when told that moving on to
Figure 14.4 An example of a significant main effect for Factor B. Note that in the the next problem was the best strategy and worked longer when told that
b1 b2 persistence was the best strategy. The subject’s self-esteem had no apparent
data and in the graph subjects in are consistently less than subjects in . influence on the persistence of the subjects.
There are no differences between subjects in conditions
a1 and
a2 . Significant Main Effects for Factor A and Factor B

Figure 14.5 An example of a significant main effect for Factors A and B. Note that in

the data and in the graph subjects in

a1 are consistently less than subjects in . In

addition there are difference between subjects in conditions 1 and 2 . The b b

difference between the groups is consistent, however. Therefore, there is no
interaction.
 m e a n ti m e s p e n t o n e a ch a n a g r a m ( s e c
50 What is the difference between the persist and don’t persist instruction conditions
40
30
for each level of self-esteem? For both levels of self-esteem, the difference is 20. It
20 means there is a consistent difference to indicate that high-self-esteem subjects are
10 more persistent than low-self-esteem subjects. In other words, having higher self-
0
Low High esteem consistently added approximately 20 seconds to a subject’s persistence at
Self-Esteem
the task.

Is there a consistent effect across types of instruction? The difference between the
persist conditions (40 – 30 = 10) is the same as the difference between the don’t
persist conditions (20 – 10 = 10). Therefore, we can conclude that regardless of a
Hypothetical mean scores in each cell for this figure: subject’s self-esteem, the instructions to persist added approximately 10 seconds to
the time spent on the task.
Instruction Self-Esteem Mean
a1 a2 The main point to recognize for this scenario is that differences between the
Low ( ) High ( ) conditions remained constant. Specifically, the effects of the instructions were
Persist (
b1 )
30 40 35 consistent across levels of self-esteem. Similarly, the effect of self-esteem was
consistent across the instruction conditions. When the effect of a variable is not
Don’t Persist (
b2 10 20 15
consistent across levels of the other variables, we conclude that the variables
) interact.
Mean 20 30 25

Figure 14.5 represent a condition where both main effects are significant and there 14.5.2 The Concept of Interaction
is no interaction. If you look at the figure, you will see that overall, high-self-esteem
Two factors are said to interact when the effect of one variable on some measure of
subjects are more persistent than low-esteem subjects, and subjects told to persist
behavior depends on either the presence or the amount of another variable. The
were more persistent than subjects told not to persist. Because the difference
finding of a significant interaction is of great importance to our interpretation of the
between the low- and high self-esteem subjects is consistent across the two sets of
data. A significant interaction means that we cannot claim that the effects of one
instructions, we say that the effects of reinforcement are consistent for each
variable is the same at all levels of a second variable. That means you cannot
personality characteristic. In other words, there is no interaction.
explain interaction by merely describing the separate effects of the independent
In this example, both main effects are significant. There fore, we can conclude that variable. Instead, you must examine the unique combinations that are created
both variables contributed to the overall persistence of the subjects. Notice that the when different levels of each variable are combined.
lines are parallel that means there is no interaction and that the two variables are
independent of each other. Statistician s call this type of condition an additive
effect because each variable adds a constant effect to the total variance of the data. Significant Main Effect for Factor A and an Interaction
If the lines are reasonably parallel, then there is an additive effect.
Figure 14.6
 m e a n ti m e s p e n t o n e a ch a n a g r a m (s e co n d s )
An example of a main effect for Factor A and an interaction. Note that in the data receive, they are equally persistent under both conditions. High-self-esteem
and in the graph there are significant differences between subjects in conditions subjects are influenced by the instructions. When told to persist, they are very
a1 and 2 . This difference is not consistent. Therefore, there is an a persistent. When told not to persist, they stop the task quickly.

interaction. There appears to be no effect on self-esteem when the subjects were 40

m e a n ti m e s p e n t o n e a ch a n a g r a m (s e co n d s )

told to persist. In contrast, when the subjects were told not to persist, low-self- 30
esteem subjects were very persistent whereas high-self-esteem subjects were less 20
persistent. 10

0
50 Low High
40 Self-Esteem
30
20
10
0
Low High
Self-Esteem
Hypothetical mean scores in each cell for this figure:

Instruction Self-Esteem Mean

a
Low ( 1 )
a
High ( 2 )
Hypothetical mean scores in each cell for this figure: Persist (
b1 )
25 40 32

Instruction Self-Esteem Mean Don’t Persist (

b2 25 10 17
a
Low ( 1 )
a
High ( 2 ) )
Mean 25 25 25
Persist (
b1 )
30 30 30

Don’t Persist ( 2
b 45 15 30
Figure 14.7 represent an interaction. The lines are not parallel to one another, and
) therefore we can conclude that there are inconsistent effects across the treatment
Mean 37.5 22.5 30 conditions. Specifically, it appears that the subjects with low self-esteem are not
Significant Main Effect for Factor B and an Interaction influenced by the instructions: They spend the same average amount of time (25
Figure 14.7 seconds) on the anagrams regardless of the instructions. There is a different pattern
for subjects with high self-esteem. These subjects spend a great deal of time on
An example of a main effect for Factor B and an interaction. Note that in the data each problem (40 seconds) when told to persist, and little time on a task when told
and in the graph there are significant differences between subjects in conditions not to persist (10 seconds).
b1 and 2 . This difference is not consistent. Therefore, there is an b Significant Main Effects for A and B and an Interaction
interaction. Low-self-esteem subjects are not influenced by the instructions they
This is an interesting type of interaction because all the effects are significant. First,
we find that subjects with high-self-esteem are less persistent than subjects with
low self-esteem. Second, subjects told to persist are more persistent than subjects
told not to persist. Finally, there is an interaction between these two variables
suggesting that the instruction not to persist has a greater effect on subjects with
high self-esteem than on subjects with low self-esteem.
In this example, subjects with high self-esteem appear to be more influenced by the
instructions than subjects with low-self-esteem. Compared to subjects with low-
self-esteem, subjects with high self-esteem persist more when the instructions
m ea n ti m e sp en t o n ea ch a n a g r a m (seco n d s)
An example of significant main effects for Factors A and B and an interaction. Note
that in the data and in the graph both the personality and the instructions influence
the results. In general, low-self-esteem subjects are more persistent than high-self-
esteem subjects. Also, instructions to persist create greater persistence than
instructions not to persist. The pattern of results is not consistent, however. Note
that effects of the instructions not to persist is greater for the high-self-esteem
subjects than the other conditions. This difference represents the interaction.

recommend persistence and are less persistent when the instructions recommend
not to persist. Significant Interaction and No Main Effects
45
40
Figure 14.9 illustrates no main effects but there is a significant interaction. It shows
35 there is no significant main effects for either variable because the treatment
30
25 conditions have opposite effects on each other. In the graph and the data, the
20
15 instruction to persist created much persistence in the low-self-esteem subjects but

m e a n ti m e s p e n t o n e a ch a n a g r a m (s e co n d s )
10
5 not in the high-self-esteem subjects. In contrast, the instructions not to persist
0
Low High created great persistence in the high-self-esteem subjects but not the low-self-
Self-Esteem esteem subjects.

50
40
30
20
Hypothetical mean scores in each cell for this figure: 10
0
Low High
Instruction Self-Esteem Mean Self-Esteem
a
Low ( 1 )
a
High ( 2 )

Persist (
b1 )
40 35 37.5

Don’t Persist (
b2 30 10 20
) Hypothetical mean scores in each cell for this figure:
Mean 35 22.5 28.75
Instruction Self-Esteem Mean

Low (
a1 ) High (
a2 )
Figure 14.8
 2
Persist (
b1 )
40 10 25
( X ¿⋅k )
2
(∑ X ijk )
b2 10 40 25 SS B=∑ −
Don’t Persist ( nk N
)
Mean 25 25 25 (14.6) The Degrees of Freedom for Factor B

df B=k −1
Figure 14.9 An example of a significant interaction but no significant main effects.
These data indicate that high- and low-self-esteem subjects react in opposite ways (14.7) The Mean Square of Variance Estimate for Factor B
when given instructions to either persist or not persist.
SS B
MS B =
14.6 PARTITIONING THE SUM OF SQUARES df B
The partitioning of the total sum of squares into individual components.
14.6.3 INTERACTION VARIANCE
SS tot =SS A + SS B + SS AB + SS W (14.8) The Sum of Squares for the Interaction

14.6.1 Between-Group Variance: Factor A 2 2

( X ¿ jk ) (∑ X ijk )
(14.2) The Sum of Squares of Factor A SS AB =∑ − −( SS A+ SS B )
n jk N

SS A=∑ ¿¿¿¿¿ (14.9) The Degrees of Freedom for the Interaction

df AB =( j−1 )( k −1 )
(14.3) The Degrees of Freedom for Factor A
(14.10) The Mean Square for the Interaction
df A = j−1 SS AB
MS AB=
(14.4) The Mean Square of Variance Estimate for Factor A df AB
SS A 14.6.4 WITHIN-GROUP VARIANCE
MS A=
df A (14.11) The Estimate of the Error Variation All Groups

14.6.2 Between-Group Variance: Factor B 2

2
ijk (∑ X ¿ jk )
(14.5) The Sum of Squares of Factor B SS W =∑ X −
n jk
(14.12) The Degrees of Freedom for Within-Group Variance Effect F-Ratio Estimated Effects
H0 H0
df W =∑ ( n jk −1 ) : True : False
Factor MS A σ2 ( σ 2 +σ 2 )
(14.13) The Mean Square for the Within-Group Variance A FA = F A ≈ ε2 =1. 00 F A = α 2 ε > 1. 00
MS W σε σε
2 2 2
SS W Factor MS B σε ( σ β +σ ε )
MS W = B F B= F B ≈ 2 =1. 00 F B= >1. 00
df W MSW σε σε
2

14.6.5 TOTAL VARIANCE

Factor MS AB σ 2ε ( σ 2αβ +σ 2ε )
AB F AB = F AB ≈ 2 =1. 00 F AB = >1 .00
MS W σε σ 2ε
(14.14) The Total Variance
2
2
ijk (∑ X ijk ) If the null hypothesis is true, then we assume that there is no variance between the
SS tot =∑ X −
N
groups. For example, if there is no effect for factor A, we assume that
σ 2α =0 .
(14.15) The Degrees of Freedom for the Total Variance σ2
F A ≈ ε2 =1. 00
df tot =N−1 As a consequence, the F-ratio can be represented as σε when
14.6.6 THE F-RATIOS FOR THE TWO-WAY ANOVA σ 2α =0 . If the null hypothesis is false, then
σ 2α >0 and the F-ratio can be
2 2
The two-way ANOVA to partition the total variance into four sources of variance. ( σ α +σ ε )
The first three sources represent the two main effects and the interaction. The FA= 2
> 1. 00
fourth source of variance represents the random error that is common to the entire represented as σε . Therefore, if the F-ratio is not
experiment. F-ratio is used to determine if the variance associated with a treatment significantly greater than 1, we must assume that the treatment had no significant
condition is significantly greater than that which would occur by chance alone. effect on the data.

The logic of the F-ratios for the two-way ANOVA is it assume that the mean square
for each effect (e.g. factor A, factor B, and factor AB) represents the variance due to
the treatment conditions plus the variance due to random error. The within-group
mean square represents the variance due to random error.
Table 14.4
Table representing the F-Ratios for a Two-Way ANOVA and the Estimated F-Ratios
By contrast, if the F-ratio is greater than 1, we assume that part of the variance in
the data may be due to random effects but that a greater part is due to the
treatment effect.
14.7 A WORKED EXAMPLE
A two-way ANOVA applied by Sanderlands, Brocker, & Glynn (1988) in the conduct
of an experiment that examined persistence.

when
H0 is True and when
H0 is False Null Hypotheses:
 H0: σ 2A =σ 2W ; There is no main effect for self-esteem.
(Instructions)
Low (
a1 ) High (
a2 )
a1 b 1 a2 b 1
H0: σ 2B =σ 2W ; There is no main effect for instructions. b1 X 111 =2. 6 X 121 =3. 5
Persist ( )
X 211 =2 . 9 X 221 =3. 0
H0: σ 2AB =σ 2W ; There is no interaction between the two factors.
X 311 =2 .5 X 321 =2. 8
Alternative Hypotheses: X 411=2. 8 X 421=3 .1
H1: σ 2A =σ 2W X 511=2 .3 X 521 =2. 9
; There is a main effect for self-esteem.
X ¿ 11=13 . 1 X ¿ 21=15. 3 X ¿⋅1 =28 . 4
H1: σ 2B =σ 2W ; There is a main effect for instructions.
n11 =5 n21=5 n¿ 1=10
X 11 =2. 62 X 21=3 . 06 X ¿ 1=2 . 84
H1: σ 2AB =σ 2W ; There is an interaction between the two factors. a 1 b2 a 2 b2
Statistical Test: The two-way ANOVA is used because we are simultaneously
Don’t Persist ( X 112 =2 .5 X 122 =2. 0
comparing the effects if two independent variables. We assume that the data are
b2 ) X 212 =2. 3 X 222 =1. 9
normally distributed and that the variances of the groups are equal.
2
X 312 =2. 4 X 322 =1.8
Significance Level: α=0 . 05 . If the F-ratio is sufficiently large, we will reject the X 412=1. 9 X 422=2. 2
null hypothesis in favor of the alternative hypothesis. X 512 =2. 2 X 522 =1. 7
Sampling Distribution: The F-distribution with df-1.16 for each hypothesis. X ¿ 12=11.3 X ¿ 22=9 .6 X ¿ 2=20. 9
n12=5 n22=5 n¿ 2=10
Critical Region for Rejection of
H 0 : F0 .5 ≥4.49 . A critical region is that
X 12=2 . 26 X 22=31 . 92 X ¿ 2=2. 09
portion of the area under the F-distribution that includes values of a statistic that
H0 X ¿ 1⋅¿=24 .4 ¿ X ¿ 2⋅¿=24 .9 ¿ X ¿⋅¿ =49 .3
lead to rejection of . If the F-ratio is equal to or larger than 4.49, we will n1⋅¿=10 ¿ n2⋅¿=10 ¿ N . .=20
reject
H0 . X 1⋅¿=2 . 44 ¿ X 2⋅¿=2 . 49 ¿ X ¿⋅¿=2 . 465 ¿
2
Table 14.5 ∑ X ijk=49 .3 ∑X ijk
=125. 99
2
Hypothetical data for Example ( 49. 3 )
SS tot =125 . 99−
20
Factor B Factor A (Self-Esteem)
SS tot =125 . 99−121 . .52
 2
SS tot =4 . 47 ( X ¿⋅k )
2
(∑ X ijk )
SS B=∑ −
nk N
Step 1: Calculate the sum of squares for factor A.
( 28 . 4 )2 ( 20 . 9 )2 ( 49 . 3 )2
SS B = + −
SS A=∑ ¿¿¿¿¿ 10
SS B =2. 813
10 20

( 24 . 4 )2 (24 .9 )2 ( 49 .3 )2 Step 5: Calculate the degrees of freedom for factor B.

SS A= + −
10 10 20
df B=k −1
SS A=0. 013
df B=2−1
Step 2: Calculate the degrees of freedom for factor A.
df B =1
df A = j−1
Step 6: Calculate the mean square for factor B.
df A =2−1
SS B
df A =1 MS B =
df B
Step 3: Calculate the mean square for factor A. 2. 813
MS B =
SS A 1
MS A=
df A MS B =2 . 813
0 . 013 Step 7: Calculate the sum of squares for the interaction of AB.
MS A=
1 2 2
( X ¿ jk ) (∑ X ijk )
MS A=0 . 013 SS AB =∑ − −( SS A+ SS B )
n jk N
Step 4: Calculate the sum of squares for factor B.
( 13 . 1 )2 ( 11 . 3 )2 ( 15 . 3 )2 ( 9. 6 )2 ( 49 .3 )2
SS AB = + + + − −( 0 .013+2. . 813 )
5 5 5 5 20
 SS AB =0 .760 df W =16
Step 8: Calculate the degrees of freedom for the interaction of AB. Step 12: Calculate the mean square for the within-group error.

df AB =( j−1 )( k −1 ) SS W
MS W =
df W
df AB =( 1 ) (1 )
0 . 880
df AB =1 MS W =
16
Step 9: Calculate the mean square for the interaction of AB.
MS W =0 . 055
SS AB
MS AB= Step 13: Calculate the Total sum of squares.
df AB
2
0 . 760
2
ijk (∑ X ijk )
MS AB= SS tot =∑ X −
1 N

MS AB=0. 760 ( 49. 3 )2

SS tot =125 . 990−
20
Step 10: Calculate the sum of the squares for the within-group error.
2
SS tot =4 . 466
2
ijk (∑ X ¿ jk )
SS W =∑ X − We can check our calculations using
n jk
SS tot =SS A+ SS B + SS AB + SS W
( 13 . 1 )2 ( 11. 3 )2 ( 15 . 3 )2 ( 9. 6 )2
SS W =125 . 990− + + + SS tot =0. 013+2 . 813+0 .760+0. 880
5 5 5 5
SS W =0. 880 SS tot =4 . 466
Step 11: Calculate the degrees of freedom for within-group error. Step 14: Calculate the degrees of freedom for the total variance

df W =∑ ( n jk −1 ) df tot =N−1

df W =4 +4 +4 +4 df tot =20−1
 df tot =19
Therefore, the effect of the instructions (factor B) is
Step 15: Calculate the F-ratio for each treatment effect.

MS A 0. 013 1 ( 51. 145−1 )

Factor A
FA= =0. 236 ω2B =
FA = 1 ( 51 .145−1 ) +20
MS W 0. 055
Factor B MS B 2. 813 50 .145
F B= F B= =51 . 145 ω2B=
MSW 0 .055 70 .145
Factor AB MS AB 0 .760 ω2B =0 . 715
F AB = F AB = =13. 818
MS W 0 .055
The effect for the interaction (factor AB) is

Summary Table for the Data 1 ( 13 .827−1 )

ω2AB=
Source SS df MS F
1 ( 13 . 827−1 )+20
A 0.013 1 0.013 0.236
12 . 827
B 2.813 1 2.813 51.145 ω2AB=
AB 0.760 1 0.760 13.818 32 . 827
Within 0.880 16 0.055
Total 4.466 19 ω2AB=0 . 391
The table tells us the result of the tests of the effects for significance, that there is a
The instructions that subjects received accounted for approximately 71.5% of the
significant main effect for the instructions, F ( 1,16 )=51.145 , p<0 . 05 ;a total difference among subjects. The interaction between self-esteem and the
instructions accounted for 39.1% of the variance that cannot be explained by either
significant interaction, F ( 1,16 )=13 . 818 p<0 . 05 ; but that the effect for
, level of self-esteem or type of instruction. More specifically, telling subjects not to
self-esteem is not significant, F ( 1,16 )=0.2326 , p>0 . 05 . persist at the task caused subjects with high self-esteem to spend less time on each
problem but caused subjects with low self-esteem to spend more time on each
The relation between the independent and dependent variable. Note that even very problem.
small effects may be statistically significant if the sample size is sufficiently large.
2
Hence, we need to calculate ω for each of the significant F-ratios. The statistic
df effect ( F effect −1 )
ω2 =
ω
2 df F
effect ( effect )−1 + N
is defined as , where the subscript
“effect” refers to a particular effect such as Factor A, factor B, or Factor AB.
EXERCISE

Does one’s mood influence how well things are remembered? To conduct the
research, a psychologist decides to use two factors. Factor A represents the
emotional condition under which the subjects are required to learn a list of words.
a a
Subjects are made to feel either sad ( 1 ) or happy ( 2 ). Factor B represents
the emotional condition under which the subjects’ memory is tested. Subjects are
b b
made to feel either sad ( 1 ) or happy ( 2 ) before they are told to recall the
words. The data appear in the accompanying table.

Emotion Experienced Emotion Experienced When Learning List

During the Recall of List a
Sad ( 1 )
a
Happy ( 2 )

Sad (
b1 )
70 48
77 41
79 47
77 49
78 44

Happy (
b2 )
39 85
40 83
40 88
36 88
38 86
 Null Hypotheses:

a. Draw a graph of these data and describe the results. Does there appear to be a H0: σ 2A =σ 2W ; There is no main effect on the emotional condition
main effect for factor A? Does there appear to be main effect for factor B?
Does there appear to be an interaction of the two factors? under which the subjects are required to learn a list of words.
b. Conduct a two-way ANOVA on these data and prepare a summary table.
c. Is there a significant interaction between these two factors?
H0: σ 2B =σ 2W ; There is no main effect on the emotional condition
2
Calculate ω
under which the subjects are required to learn a list of words after they are exposed
d. for each main effect and the interaction.
to a specific emotional condition.
e. Write a brief description of these results based on the analysis conducted.

ANSWER H0: σ 2AB =σ 2W ; There is no interaction between the two factors.

a. Table 1 Main Effect for Factor A and Factor B and their Interaction Alternative Hypotheses:
140
H1: σ 2A ≠σ 2W
No. of Words Memorzed

120
; There is a main effect for self-esteem.
100
80 H1: σ 2B≠σ 2W ; There is a main effect for instructions.
60
40 H1: σ 2AB ≠σ 2W ; There is an interaction between the two factors.
20
0 Statistical Test: The two-way ANOVA is used because we are simultaneously
a1 a2
comparing the effects if two independent variables. We assume that the data are
Emotional Condition normally distributed and that the variances of the groups are equal.

b1 b2 Significance Level: α=0 . 05 . If the F-ratio is sufficiently large, we will reject the
null hypothesis in favor of the alternative hypothesis.
Table 1 illustrates the emotional condition under which subjects’ memory
is tested. There is no significant main effect for factor A since the Sampling Distribution: The F-distribution with df-1;16 for each hypothesis.
emotional conditions in factor A has opposite direction but there is a
significant main effect for factor B, that is, subjects who are made to feel Critical Region for Rejection of
H 0 : F0 .5 ≥4.49 . A critical region is that
sad is before they are told to recall the words consistently less than those portion of the area under the F-distribution that includes values of a statistic that
subjects who are made to feel happy and there difference is significant or
not consistent. Therefore, there is an interaction. lead to rejection of
H0 . If the F-ratio is equal to or larger than 4.49, we will
b. Two-Way ANOVAx
reject
H0 .
Table 14.5 ( 1233 )2
SS tot =84073−
Hypothetical data for Example 20
SS tot =84073−76014 . 45
Factor B Factor A (With Intervention)
(No a1 a2 SS tot =8058. 55
Intervention) Sad ( ) Happy ( )
a1 b1 a2 b1
Step 1: Calculate the sum of squares for factor A.
Sad (
b1 )
70 48
77 41
79
77
47
49 SS A=∑ ¿¿¿¿¿
78 44
∑ a 1 b1=381 ∑ a 2 b1=229 ∑ b 1=610 ( 574 )2 ( 659 )2 ( 1233 )2
n11 =5 n¿ 1=10 SS A= + −
10 10 20
X 11 =76 .2 n21=5 X ¿ 1=61
SS A =361. 25
X 21=45 .8
a1 b2 a 2 b2 Step 2: Calculate the degrees of freedom for factor A.

Happy (
b2 )
39 85
df A = j−1
40 83
40 88
36 88 df A =2−1
38 86
df A =1
∑ a 1 b2=193 ∑ a 2 b2=430 ∑ b 2=623
n12=5 n22=5 n¿ 2=10 Step 3: Calculate the mean square for factor A.
X 12=38 . 6 X 22=86 X ¿ 2=62. 3 SS A
∑ a 1=574 ∑ a 2=659 MS A=
∑ a=1233 df A
n1⋅¿=10 ¿ n2⋅¿=10 ¿ N . .=20
X a 1 =57 . 4 X a 2 =65 .9 X t =61. 65 361 .25
MS A=
2 1
∑ X ijk=1233 ∑X ijk
=84073
MS A=361 .25
Step 4: Calculate the sum of squares for factor B. ( 381 )2 ( 193 )2 ( 229 )2 ( 430 )2 ( 1233 )2
SS AB = + + + − −( 361. 25+8 . 45 )
2 5 5 5 5 20
( X ¿⋅k )2 (∑ X ijk )
SS B=∑ −
nk N SS AB =7566 .05

( 610 )2 ( 623 )2 ( 1233 )2 Step 8: Calculate the degrees of freedom for the interaction of AB.
SS B = + −
10 10 20 df AB =( j−1 )( k −1 )
SS B=8 . 45 df AB =( 1 ) (1 )
Step 5: Calculate the degrees of freedom for factor B.
df AB =1
df B=k −1
Step 9: Calculate the mean square for the interaction of AB.
df B=2−1 SS AB
MS AB=
df B =1 df AB

Step 6: Calculate the mean square for factor B. 7566 . 05

MS AB=
1
SS B
MS B = MS AB=7566 . 05
df B
Step 10: Calculate the sum of the squares for the within-group error.
8 . 45
MS B=
1 2
2
ijk (∑ X ¿ jk )
SS W =∑ X −
MS B =8 . 45 n jk
Step 7: Calculate the sum of squares for the interaction of AB.
( 381 )2 ( 193 )2 (229 )2 ( 430 )2

SS AB =∑
( X ¿ jk )
2
−
(∑ X ijk )
2

−( SS A+ SS B )
SS W =84073−
5
+
5 ( +
5
+
5 )
n jk N SS W =122 . 8
Step 11: Calculate the degrees of freedom for within-group error.
 df W =∑ ( n jk −1 ) df tot =N−1

df W =4 +4 +4 +4 df tot =20−1

df W =16 df tot =19

Step 12: Calculate the mean square for the within-group error. Step 15: Calculate the F-ratio for each treatment effect.

SS W Factor A MS A 361. 25
MS W = FA = FA= =47 . 068
df W MS W 7 . 675
Factor B MS B 8. 45
122 .8 F B= F B= =1 .101
MS W = MSW 7 .675
16
Factor AB MS AB 7566 . 05
F AB = F AB = =985 .805
MS W =7 . 675 MS W 7 .675
Step 13: Calculate the Total sum of squares.
2 Summary Table for the Data
2
ijk (∑ X ijk )
SS tot =∑ X − Source SS df MS F
N A 361.25 1 361.25 47.068
B 8.45 1 8.45 1.101
( 1233 )2 AB 7566.05 1 7566.05 985.805
SS tot =84073−
20 Within 122.8 16 122.8
Total 8058.55 19
SS tot =8058. 55 The table tells us the result of the tests of the effects for significance, that there is a

We can check our calculations using
SS tot =SS A+ SS B + SS AB + SS W
SS tot =361 . 25+8 . 45+7566 . 05+122. 8
SS tot =8058. 55
Step 14: Calculate the degrees of freedom for the total variance
significant main effect for factor A, F ( 1,16 )=47. 068 , p<0 . 05 . On the
F ( 1,16 )=1.101 ,
other hand there is no significant main effect for factor B,
p>0 . 05 . In addition, there is a significant interaction in factors A and B with
F ( 1,16 )=985 .805 , p<0 . 05 .
 df effect ( F effect −1 )
ω2 =
df F
effect ( effect ) −1 + N
We use the formula, , to determine the amount
of effect of each factor which has a significant main effect on the dependent
variable.

The exposure of the subjects to a particular emotional condition before they were
told to memorize has a 69.73% significant effect on the amount of words they could
memorize.

1 ( 47 . 068−1 )
ω2A = =0 . 6973
1 ( 47 .068−1 ) +20
The interaction of before and the actual emotional condition of the subject has
98.01% of the variance that cannot be explained.

1 ( 985 . 805−1 )
ω2AB= =0 . 9801
1 ( 985. 805−1 )+20