Chapter Sixteen

Analysis of Variance and

Covariance
Relationship Among Techniques
Analysis of variance (ANOVA) is used as a test of
means for two or more populations. The null hypothesis,
typically, is that all means are equal.
Analysis of variance must have a dependent variable
that is metric (measured using an interval or ratio scale).
There must also be one or more independent variables
that are all categorical (nonmetric). Categorical
independent variables are also called factors.
Relationship Among Techniques
A particular combination of factor levels, or categories, is
called a treatment.
One-way analysis of variance involves only one
categorical variable, or a single factor. In one-way
analysis of variance, a treatment is the same as a factor
level.
If two or more factors are involved, the analysis is
termed n-way analysis of variance.
If the set of independent variables consists of both
categorical and metric variables, the technique is called
analysis of covariance (ANCOVA). In this case, the
categorical independent variables are still referred to as
factors, whereas the metric-independent variables are
referred to as covariates.
Relationship Amongst Test, Analysis of
Variance, Analysis of Covariance, &
Fig. 16.1
Regression
Metric Dependent Variable

One Independent
Variable One or More
Independent Variables

Categorical: Categorical
Binary Interval
Factorial and Interval

Analysis of Analysis of
t Test Variance Covariance Regression

More than
One Factor One Factor

One-Way Analysis N-Way Analysis

of Variance of Variance
 One-way Analysis of Variance
Marketing researchers are often interested in examining
the differences in the mean values of the dependent
variable for several categories of a single independent
variable or factor. For example:

Do the various segments differ in terms of their volume

of product consumption?
Do the brand evaluations of groups exposed to different
commercials vary?
What is the effect of consumers' familiarity with the store
(measured as high, medium, and low) on preference for
the store?
 Statistics Associated with One-way
Analysis

of Variance
eta2 ( 2). The strength of the effects of X (independent

variable or factor) on Y (dependent variable) is
measured by eta2 ( 2). The value of 2 varies between 0
and 1.

F statistic. The null hypothesis that the category means

are equal in the population is tested by an F statistic
based on the ratio of mean square related to X and
mean square related to error.

Mean square. This is the sum of squares divided by the

appropriate degrees of freedom.
 Statistics Associated with One-way
Analysis of Variance
SSbetween. Also denoted as SSx, this is the variation in Y
related to the variation in the means of the categories of
X. This represents variation between the categories of
X, or the portion of the sum of squares in Y related to X.

SSwithin. Also referred to as SSerror, this is the variation

in Y due to the variation within each of the categories of
X. This variation is not accounted for by X.

SSy. This is the total variation in Y.

 Anova and the
Design of Experiments

.Why and How to use ANOVA in Marketing Research?

What is the Difference between one way ANOVA, N-way

ANOVA, MANOVA and ANCOVA
 Analysis of variance (ANOVA) is used as a test of means for
two or more populations. The null hypothesis, typically, is that
all means are equal.
Analysis of variance must have a dependent variable that is
metric (measured using an interval or ratio scale).
There must also be one or more independent variables that are
all categorical (nonmetric). Categorical independent variables
are also called factors.

Design
Completely Randomised Design in a One-Way ANOVA (Single
Factor)
Randomized Block Design (Single Blocking Factor)
Factorial Design with 2 or more Factors.
 One-way Analysis of Variance
Application in MR

Do the various segments differ in terms of their volume

of product consumption?
Do the brand evaluations of groups exposed to different
commercials vary?
What is the effect of consumers' familiarity with the store
(measured as high, medium, and low) on preference for
the store?
Conducting
Fig. 16.2
One-way ANOVA
Identify the Dependent and Independent Variables

Decompose the Total Variation

Measure the Effects

Test the Significance

Interpret the Results

 Conducting One-way Analysis of
Variance
Decompose the Total Variation
The total variation in Y, denoted by SSy, can be
decomposed into two components:

SSy = SSbetween + SSwithin

where the subscripts between and within refer to the

categories of X. SSbetween is the variation in Y related to
the variation in the means of the categories of X. For
this reason, SSbetween is also denoted as SSx. SSwithin is
the variation in Y related to the variation within each
category of X. SSwithin is not accounted for by X.
Therefore it is referred to as SSerror.
 Decomposition of the Total Variation:
Table 16.1 One-way ANOVA
Independent Variable X
Total
Categories Sample
Within X1 X2 X3 Xc
Category Y1 Y1 Y1 Y1 Y1 Total
Variation Variation
Y2 Y2 Y2 Y2 Y2 =SSy
=SSwithin : :
: :
Yn Yn Yn Yn YN
Category Y1 Y2 Y3 Yc Y
Mean
Between Category Variation = SSbetween
 Conducting One-way Analysis of
Variance
In analysis of variance, we estimate two measures of
variation: within groups (SSwithin) and between groups
(SSbetween). Thus, by comparing the Y variance
estimates based on between-group and within-group
variation, we can test the null hypothesis.

Measure the Effects

The strength of the effects of X on Y are measured as
follows:

2 = SSx/SSy = (SSy - SSerror)/SSy

The value of 2 varies between 0 and 1.

 Conducting One-way Analysis of
Variance
Test Significance
In one-way analysis of variance, the interest lies in testing the null
hypothesis that the category means are equal in the population.

H0: 1 = 2 = 3 = ........... = c

Under the null hypothesis, SSx and SSerror come from the same source of
variation. In other words, the estimate of the population variance of Y,
2
S y = SSx/(c - 1)
= Mean square due to X
= MSx
or
2
S y
= SSerror/(N - c)
= Mean square due to error
= MSerror
 Conducting One-way Analysis of
Variance
Test Significance
The null hypothesis may be tested by the F statistic
based on the ratio between these two estimates:
SS x /(c - 1) MS x
F= =
SS error/(N - c) MS error

This statistic follows the F distribution, with (c - 1) and

(N - c) degrees of freedom (df).
 Conducting One-way Analysis of
Variance
Interpret the Results
If the null hypothesis of equal category means is not
rejected, then the independent variable does not have a
significant effect on the dependent variable.
On the other hand, if the null hypothesis is rejected, then
the effect of the independent variable is significant.
A comparison of the category mean values will indicate
the nature of the effect of the independent variable.
 Illustrative Applications of One-way
Analysis of Variance
We illustrate the concepts discussed in this chapter
using the data presented in Table 16.2.

The department store is attempting to determine the

effect of in-store promotion (X) on sales (Y). For the
purpose of illustrating hand calculations, the data of
Table 16.2 are transformed in Table 16.3 to show the
store sales (Yij) for each level of promotion.

The null hypothesis is that the category means are

equal:
H0: 1 = 2 = 3.
Effect of Promotion and Clientele
Table 16.2 on Sales
Store Num ber Coupon Level In-Store Prom otion Sales Clientel Rating
1 1.00 1.00 10.00 9.00
2 1.00 1.00 9.00 10.00
3 1.00 1.00 10.00 8.00
4 1.00 1.00 8.00 4.00
5 1.00 1.00 9.00 6.00
6 1.00 2.00 8.00 8.00
7 1.00 2.00 8.00 4.00
8 1.00 2.00 7.00 10.00
9 1.00 2.00 9.00 6.00
10 1.00 2.00 6.00 9.00
11 1.00 3.00 5.00 8.00
12 1.00 3.00 7.00 9.00
13 1.00 3.00 6.00 6.00
14 1.00 3.00 4.00 10.00
15 1.00 3.00 5.00 4.00
16 2.00 1.00 8.00 10.00
17 2.00 1.00 9.00 6.00
18 2.00 1.00 7.00 8.00
19 2.00 1.00 7.00 4.00
20 2.00 1.00 6.00 9.00
21 2.00 2.00 4.00 6.00
22 2.00 2.00 5.00 8.00
23 2.00 2.00 5.00 10.00
24 2.00 2.00 6.00 4.00
25 2.00 2.00 4.00 9.00
26 2.00 3.00 2.00 4.00
27 2.00 3.00 3.00 6.00
28 2.00 3.00 2.00 10.00
29 2.00 3.00 1.00 9.00
30 2.00 3.00 2.00 8.00
 Illustrative Applications of One-way
Analysis of Variance
It can be verified that
SSy = SSx + SSerror
as follows:
185.867 = 106.067 +79.80
The strength of the effects of X on Y are measured as follows:
2 = SSx/SSy
= 106.067/185.867
= 0.571

In other words, 57.1% of the variation in sales (Y) is accounted for

by in-store promotion (X), indicating a modest effect. The null
hypothesis may now be tested.
SS x /(c - 1) MS X
F= =
SS error/(N - c) MS error
106.067/(3-1)
F=
79.800/(30-3)
= 17.944
 Illustrative Applications of One-way
Analysis of Variance
From Table 5 in the Statistical Appendix we see that for 2
and 27 degrees of freedom, the critical value of F is 3.35
for .= Because
0.05 the calculated value of F is greater
than the critical value, we reject the null hypothesis.
We now illustrate the analysis of variance procedure
using a computer program. The results of conducting the
same analysis by computer are presented in Table 16.4.
 One-Way ANOVA:
Effect of In-store Promotion on Store
Table 16.3
Sales
Source of Sum of df Mean F ratio F prob.
Variation squares square
Between groups 106.067 2 53.033 17.944 0.000
(Promotion)
Within groups 79.800 27 2.956
(Error)
TOTAL 185.867 29 6.409

Cell means

Level of Count Mean

Promotion
High (1) 10 8.300
Medium (2) 10 6.200
Low (3) 10 3.700
TOTAL 30 6.067
N-way Analysis of Variance
In marketing research, one is often concerned with the
effect of more than one factor simultaneously. For
example:

How do advertising levels (high, medium, and low) interact

with price levels (high, medium, and low) to influence a
brand's sale?

Do educational levels (less than high school, high school

graduate, some college, and college graduate) and age
(less than 35, 35-55, more than 55) affect consumption of
a brand?

What is the effect of consumers' familiarity with a

department store (high, medium, and low) and store
image (positive, neutral, and negative) on preference for
the store?
 N-way Analysis of Variance
Consider the simple case of two factors X1 and X2 having categories c1 and c2.
The total variation in this case is partitioned as follows:

SStotal = SS due to X1 + SS due to X2 + SS due to interaction of X1 and X2 +

SSwithin

or
SS y = SS x 1 + SS x 2 + SS x 1x 2 + SS error

The strength of the joint effect of two factors, called the overall effect, or
multiple 2, is measured as follows:
Multiple 2 = (SS x 1 + SS x 2 + SS x 1x 2)/ SS y
N-way Analysis of Variance
The significance of the overall effect may be tested by an F test, as
follows

(SS x 1 + SS x 2 + SS x 1x 2)/dfn
F=
SS error/dfd
SS x 1,x 2,x 1x 2/ dfn
=
SS error/dfd
MS x 1,x 2,x 1x 2
=
MS error
where

dfn = degrees of freedom for the numerator

= (c1 - 1) + (c2 - 1) + (c1 - 1) (c2 - 1)
= c1c2 - 1
dfd = degrees of freedom for the denominator
= N - c1c2
MS = mean square
 N-way Analysis of Variance
If the overall effect is significant, the next step is to examine the
significance of the interaction effect. Under the null hypothesis of
no interaction, the appropriate F test is:
SS x 1x 2/dfn
F=
SS error/dfd
MS x 1x 2
=
MS error

where

dfn = (c1 - 1) (c2 - 1)

dfd = N - c 1c 2
 N-way Analysis of Variance
The significance of the main effect of each factor may be tested
as follows for X1:
SS x 1/dfn
F=
SS error/dfd
MS x 1
=
MS error

where

dfn = c1 - 1
dfd = N - c 1c 2
 Two-way
Table 16.4
Analysis of Variance
Source of Sum of Mean Sig. of
Variation squares df square F F 2
Main Effects
Promotion 106.067 2 53.033 54.862 0.000 0.557
Coupon 53.333 1 53.333 55.172 0.000 0.280
Combined 159.400 3 53.133 54.966 0.000
Two-way 3.267 2 1.633 1.690 0.226
interaction
Model 162.667 5 32.533 33.655 0.000
Residual (error) 23.200 24 0.967
TOTAL 185.867 29 6.409
 Two-way
Table 16.4 cont.
Analysis of Variance
Cell Means
Promotion Coupon Count Mean
High Yes 5 9.200
High No 5 7.400
Medium Yes 5 7.600
Medium No 5 4.800
Low Yes 5 5.400
Low No 5 2.000
TOTAL 30

Factor Level
Means
Promotion Coupon Count Mean
High 10 8.300
Medium 10 6.200
Low 10 3.700
Yes 15 7.400
No 15 4.733
Grand Mean 30 6.067
 One-Way ANOVA
This particular design is used when there is only one categorical independent
variable, and one dependent (metric) variable.
Each category of an independent variable is called a level. The independent
variable may be different levels of prices, or different pack sizes, or different
product colours, and the effect (dependent variable) could be sales, preferences or
attitudes towards the brand.
In the example that follows, we will look at advertising copy alternatives as the
independent variable, and preference rating for the advertising copy as the
dependent variable.
Worked Example Problem:
In this example, we assume that three different versions of advertising copy have
been created by an advertising agency for a campaign. Let us call these versions of
copy ADCOPY 1, 2 and 3. Now, the ad agency wants to test which of these three
versions of the advertising copy is preferred by its target population, before they
launch the campaign.

A sample of 18 respondents is selected from the target population in the nearby

areas of the city. At random, these 18 respondents are assigned to the 3 versions of
ad copy. Each version of ad copy is thus shown to six of the respondents.
The respondents are asked to rate their liking for the ad copy shown to them on a
scale of 1 to 10. (1 = Not liked at all, 10 = Liked a lot, and other values in between
these two). The ratings given by the 18 respondents are tabulated.
Slide 5

Input Data

Fig 1. shows the input data for the 18 respondents.

Fig. 1.

Sr. Ad rating
No. copy
1 1 6.00
2 1 7.00
3 1 5.00
4 1 8.00
5 1 8.00
6 1 8.00
7 2 4.00
8 2 4.00
9 2 5.00
10 2 7.00
Slide 5 contd...

Fig. 1. Contd

Sr. Ad rating
No. copy
11 2 7.00
12 2 6.00
13 3 5.00
14 3 5.00
15 3 4.00
16 3 7.00
17 3 8.00
18 3 7.00

The codes in the ad copy, column (1,2,3) indicate

the different versions of the ad. The last column,
rating, is the rating given by a respondent to the
adcopy seen by him/her. Thus, six respondents have
rated each ad. Please note, that these eighteen
respondents were randomly assigned to each of the
three ad versions. This random assignment is called a
completely randomised assignment or design.
Slide 6

The input data in fig 1 is input into a statistical

package for performing a One-Way ANOVA,
because we have only 1 categorical factor (Ad copy)
at 3 levels 1, 2, 3 and 1 dependent variable
Rating.

Output

The output of the computerised One-Way ANOVA

is shown in fig. 2.
Fig. 2

Source of Sum of DF Mean F Sig.

Variation Squares Square of F

Main 7.000 2 3.500 1.780 .203

Effects
ADCOPY 7.000 2 3.500 1.780 .203
Explained 7.000 2 3.500 1.780 .203
Residual 29.500 15 1.967
Total 36.500 17 2.147
Slide 6 contd.

The first column is titled Source of Variation. Under this, labeled Main
Effects, is the single independent variable called ADCOPY.

We then go to the last column, where the significance of the F test is given. It
is .203 in this case, for the factor ADCOPY. This indicates that at the
confidence level of 95 percent, (corresponding to significance level of 0.05),
the F-test proves the model is not significant. In other words, the Ratings
given to the three ad copy versions are not significantly different from each
other.
Slide 7

The ANOVA has thus told us what we may not have been able to gauge if we had
simply looked at the mean ratings for each ad copy by computing these.

For example, the ratings for the ad copy version 1 are 6,7,5,8,8,8 and the mean
rating is (6+7+5+8+8+8) / 6, or 42/6 = 7. Similarly, the mean rating of ad copy
version 2 is (4+4+5+7+7+6) / 6, or 33/6 = 5.5. The mean rating for ad copy
version 3 is (5+5+4+7+8+7) / 6, or 36/6 = 6.

At a glance, the three mean ratings appear to be different 7, 5.5 and 6. But the
ANOVA tells us that this difference is not statistically significant at the 95 percent
confidence level.
It does this by performing an F-test. The null hypothesis for this F-test is that there
is no significant difference in the mean ratings for the three ad copy versions. (H0:
M1 = M2 = M3 where M1, M2 and M3 are the mean ratings for the three versions of
ad copy). Thus, in this case, we have accepted the null hypothesis (or failed to
reject the null hypothesis), at the 95 percent confidence level.

If the significance of F in the last column of fig. 2 had been less than 0.05, we
would have rejected the null hypothesis. In that case, we would have concluded
that significant differences exist between mean ratings given to the three ad copy
Slide 8
1. Randomised Block Design :
Let us continue with the same input data as in fig. 1,
with one more column added to it. This dataset is
shown in fig. 3.
Fig. 3
sr. adcopy rating magazine
no.
1 1 6.00 1
2 1 7.00 2
3 1 5.00 3
4 1 8.00 4
5 1 8.00 5
6 1 8.00 6
7 2 4.00 1
8 2 4.00 2
9 2 5.00 3
10 2 7.00 4
11 2 7.00 5
12 2 6.00 6
13 3 5.00 1
14 3 5.00 2
15 3 4.00 3
16 3 7.00 4
17 3 8.00 5
18 3 7.00 6
Slide 8 contd..
We have made a slightly different assumption in this
case. We assume that the three versions of the adcopy
were each used in 6 different magazines. These six
magazines are coded 1, 2, 3, 4, 5, 6 and appear in the
column titled magazine. Out of the people who saw
these ads, 18 randomly chosen respondents are
picked, one from each magazine who saw a particular
version of ad. Thus, we finally have one respondent
who has seen a given version of the ad in a given
magazine. In other words, we have one respondent
for every combination of magazine and adcopy.
Slide 9
Hypothesis

1. The assignment of our sample of 18 in the above manner assumes that the
magazine in which the version of adcopy appears may have an impact on the
ratings. We can test this hypothesis - in fact, two hypotheses - by doing an ANOVA
with a randomised block design.

2. For this purpose, we use the variable Rating as the dependent variable, and
Adcopy as the factor, and Magazine as the block.

3. A block is defined as some variable which could affect the relationship between
the independent factor and the dependent variable under study in an ANOVA. In
our example, the magazine in which the advertisement appears could influence the
Rating given to Adcopy by the respondents. We are trying to remove the effect of
the magazine used, by "blocking" its effect, or treating the block separately.

4. If we do not block on a variable, its effect gets included with the error (residual)
term. This may lead to wrong conclusions about the relationship between the
independent and dependent variables. In that sense, a randomised block design is
more "powerful" than a simple one-way ANOVA, if the block effect is
significantly influencing the relationship.
 Slide 10
Output
The computer output for this problem using a randomised block design is shown in
fig. 4.
Fig. 4
Tests of significance for RATING using UNIQUE sums of squares.
Source of SS DF MS F Sig
Variation of F
Residual 3.67 10 .37
Adcopy 7.00 2 3.50 9.55 .005
Magazine 25.83 5 5.17 14.09 .000
(Model) 32.83 7 4.69 12.79 .000
(Total) 36.50 17 2.15

This table is similar to the output table of the one-way ANOVA we got earlier (fig. 2),
except that there is an additional source of variation called Magazine in the first
column of fig. 4. This is the block we have used, to test the null hypotheses

.The first null hypothesis is that mean rating of the ADCOPY is the same for all 3
versions. This is the same as the null hypothesis we had used earlier for the one-way
ANOVA.

.The second null hypothesis is that the block used (Magazine in this case) has no effect
on mean ratings given to ADCOPY versions by respondents.
Slide 11

1. To test if the null hypotheses are rejected or not, we turn to the last column of fig.
4, which gives the result of an F-test for any assumed confidence level. We will
assume we wanted to test these hypotheses at the 95 percent confidence level.
2. We know that the significance level of F in the last column should be less than 0.05
for the null hypothesis to be rejected. We see that for both the rows labelled ADCOPY
and MAGAZINE, the significance of F is less than .05. It is .005 for ADCOPY and
.000 for MAGAZINE. This means that both the null hypotheses are rejected.
3. We conclude that the mean ratings given to the 3 versions of ADCOPY are
significantly different, and also that the MAGAZINE in which the ADCOPY appears
has an impact on its rating.
4. Please note that the Blocking Factor being considered separately has now led us to
a different conclusion from that in a completely randomized test of the same basic
data. This makes the randomized block test a better test when we suspect that a
blocking factor affects the relationship between the independent variable and the
dependent variable.
 Slide 12

Latin Square Design

The Latin Square Design is an extension of the Randomised Block Design. It

consists of one independent variable (FACTOR) and two Blocks, instead of
one which we saw in the Randomised Block Design. It has no special
significance in marketing research, so we will move on to the more general
case of a factorial design where any number of factors can be tested
simultaneously for their effects on the dependent variable.

Factorial Designs

This type of design is employed when we have 2 or more independent

variables or factors. The major advantage of this design is that multiple
factors can be simultaneously tested. There are two kinds of effects that we
can test. One is called the Main Effect. The second is called the Interaction
Effect. To illustrate, we will take up an example.
Slide 13

Worked Example

In this example, we assume that we are testing for a toilet soap brand, the effect of
two Factors (independent variables) Pack Design and Price - on Sales
(dependent variable). We would like to know (1) if each of the Factors
independently affects Sales (called the Main Effects), and (2) if there is a
combined effect of Pack Design and Price (called the 2 way Interaction Effect) on
Sales.

Incidentally, if there are 3 factors in a study, then we could test for all 2-way
interaction effects and the 3-way interaction effect, in addition to the Main Effects
of the individual factors.

To continue with our example, the experiment is conducted in a simulated

environment on 18 randomly selected respondents. There are 3 levels of price
Rs. 8, Rs. 11 and Rs. 14, and 3 levels of Pack Design designated by the main
colours used Blue, Red and Green.

The coding of these variables is 1, 2, 3 respectively for Rs. 8, 11 and 14 and 1, 2,

3 for Blue, Red and Green in the case of Pack Design.
 Input Data

The input dataset is shown in .

sr. no. sales packdesn price
1 500 1 1
2 440 2 1
3 360 3 1
4 300 1 2
5 280 2 2
6 250 3 2
7 200 1 3
8 150 2 3
9 250 3 3
10 600 1 1
11 450 2 1
12 510 3 1
13 400 1 2
14 350 2 2
15 300 3 2
16 250 1 3
17 275 2 3
18 220 3 3
Column 1 is Sales, column 2 is Pack Design and Column 3 is Price. Please note that
even though Price is a continuous metric variable, for the purpose of ANOVA, being
an independent variable, it has to be treated as a categorical variable. Hence the
coding (1, 2, 3) for Price.
Slide 15
Also note from fig.5 that each combination of Price and Pack Design appears twice in
the dataset. For example, Packdesign = 1 and Price = 1 appears in Row 1 and also
Row 10. This is known as a replication in design of experiments. This is similar to
having a higher sample size in a survey.
Depending on the number of Factors and the number of levels of each Factor, the
minimum sample size required for ANOVA may go up. In such cases, multiple
observations or replications become necessary. In general, replications reduce chances
of random error affecting the results of ANOVA experiments, similar to the effects of
increasing sample size in surveys.
Output:
The output data
Source of for ourSum
Variation
factorial
of
Squares
experiment
DF are Square
Mean presented inFfig. 6. SigF of
Fig 6 Main 209305.556 4 52326.389 13.645 .001
Effects
12536.111 2 6268.056 1.635 .248
Packdesn
Price 196769.444 2 98384.722 25.656 .000
2-Way 9838.889 4 2459.722 .641 .646
Interactions
9838.889 4 2459.722 .641 .646
Packdesn
Price
Explained 219144.444 8 27393.056 7.143 .004
Residual 34512.500 9 3834.722
Total 253656.944 17 14920.997
Slide 16

Let us first look at Sources of Variation listed in the first column. The last
source of variation listed is the Residual or error term. But we are interested
in the two Main Effects and one Interaction Effect.

In this case, we are testing three hypotheses

The mean level of Sales remains the same for all 3 levels of
Pack Design (Main Effect 1).
The mean level of Sales remains the same for all 3 levels of
Price (Main Effect 2).
The mean level of Sales remains the same for all combinations
of Pack Design and Price (Interaction Effect).

Assuming 0.05 level of significance, we check whether for each of the rows
corresponding to the above hypotheses, the significance of F is below 0.05
in the last column of fig. 6.
Slide 17

We find that the significance of F values are

Pack Design - .248 (Main Effect 1)

Price - .000 (Main Effect 2)
Pack Design by Price - .646 (Interaction Effect)

Therefore, only the Price effect, one of the two main effects, is significant
statistically, at 95 percent confidence level. This means that hypothesis no. 2
is rejected.

Hypothesis 1 and 3 cannot be rejected, as the significance of F values are

greater than .05 in both cases - .248 and .646 respectively).

Thus, we conclude that Price alone has an impact on Sales. Neither Pack
Design alone nor the combination of Pack Design with Price have any
significant impact on Sales of the toilet soap.
Slide 18 Additional Comments

Experiments are today widely used in many ways in Marketing Research. For
example, test marketing of new concepts, products or prototypes is usually done
through procedures explained above, or similar to these.
STM or simulated Test Marketing procedures are extensions of the basic ANOVA type
experiments, with the added tools of forecasting based on the results of experiments
conducts. Separate software packages are now available for many specialised
applications such as STM.

Pairwise Tests

If any main effect/interaction effect turns out significant, and has more than two levels,
there is one additional test required to check for pairwise differences in the means.
For instance, in our example of one-way ANOVA, if the mean Ratings had turned out
to be significantly different at the 95 percent confidence level, we still would not know
whether only one of the pairs (say, ADCOPY 1 and ADCOPY 2) are significantly
different from each other, or if the remaining pairs (ADCOPY 1 and 3, and ADCOPY
2 and 3) are also significantly different.
To find out, we can use tests such as Tukey's Test, Duncan's Test or Scheffe's Test.
These can be requested while doing the ANOVA on most computer packages. These
tests give us a pairwise test result of significant difference among means.
These are meaningful only if the F test value for a main effect/interaction effect with
more than two levels turns out to be significant.
 Analysis of Covariance
When examining the differences in the mean values of the
dependent variable related to the effect of the controlled
independent variables, it is often necessary to take into account the
influence of uncontrolled independent variables. For example:

In determining how different groups exposed to different

commercials evaluate a brand, it may be necessary to control for
prior knowledge.
In determining how different price levels will affect a household's
cereal consumption, it may be essential to take household size into
account. We again use the data of Table 16.2 to illustrate analysis of
covariance.
Suppose that we wanted to determine the effect of in-store
promotion and couponing on sales while controlling for the affect of
clientele. The results are shown in Table 16.6.
 Analysis
Table 16.5
of Covariance
Sum of Mean Sig.
Source of Variation Squares df Square F of F
Covariance
Clientele 0.838 1 0.838 0.862 0.363
Main effects
Promotion 106.067 2 53.033 54.546 0.000
Coupon 53.333 1 53.333 54.855 0.000
Combined 159.400 3 53.133 54.649 0.000
2-Way Interaction
Promotion* Coupon 3.267 2 1.633 1.680 0.208
Model 163.505 6 27.251 28.028 0.000
Residual (Error) 22.362 23 0.972
TOTAL 185.867 29 6.409
Covariate Raw Coefficient
Clientele -0.078
 Issues in Interpretation
Important issues involved in the interpretation of ANOVA
results include interactions, relative importance of factors,
and multiple comparisons.

Interactions
The different interactions that can arise when conducting
ANOVA on two or more factors are shown in Figure 16.3.

Relative Importance of Factors

Experimental designs are usually balanced, in that each
cell contains the same number of respondents. This
results in an orthogonal design in which the factors are
uncorrelated. Hence, it is possible to determine
unambiguously the relative importance of each factor in
explaining the variation in the dependent variable.
A Classification of Interaction
Figure 16.3 Effects
Possible Interaction Effects

No Interaction Interaction
(Case 1)

Ordinal
Disordinal
(Case 2)

Noncrossover Crossover
(Case 3) (Case 4)
 Patterns
Figure 16.4
of Interaction
Case 1: No Interaction Case 2: Ordinal Interaction
X X
22 22
Y X Y X
21 21

X X X X X X
11 12 13 11 12 13
Case 3: Disordinal Interaction: Case 4: Disordinal Interaction:
Noncrossover Crossover
X X
22 22
Y X Y
21
X
21

X X X X X X
11 12 13 11 12 13
 Issues in Interpretation
2
The most commonly used measure in ANOVA is omega squared,
. This measure indicates what proportion of the variation in the
dependent variable is related to a particular independent variable or
factor. The relative contribution of a factor X is calculated as follows:
SS x - (dfx x MS error)
2x =
SS total + MS error
2

Normally, is interpreted only for
2 statistically significant effects. In
Table 16.5,

associated with the level of in-store promotion is
calculated as follows:
2 106.067 - (2 x 0.967)
p =
185.867 + 0.967

104.133
=
186.834

= 0.557
 Issues in Interpretation
Note, in Table 16.5, that
SStotal = 106.067 + 53.333 + 3.267 + 23.2
= 185.867
Likewise, the 2 associated with couponing is:
2
53.333 - (1 x 0.967)
c =
185.867 + 0.967

52.366
=
186.834
= 0.280

As a guide to interpreting , a large experimental effect produces

an index of 0.15 or greater, a medium effect produces an index
of around 0.06, and a small effect produces an index of 0.01. In
Table 16.5, while the effect of promotion and couponing are both
large, the effect of promotion is much larger.
 Issues in Interpretation
Multiple Comparisons
If the null hypothesis of equal means is rejected, we can
only conclude that not all of the group means are equal.
We may wish to examine differences among specific
means. This can be done by specifying appropriate
contrasts, or comparisons used to determine which of
the means are statistically different.
A priori contrasts are determined before conducting the
analysis, based on the researcher's theoretical
framework. Generally, a priori contrasts are used in lieu
of the ANOVA F test. The contrasts selected are
orthogonal (they are independent in a statistical sense).
 Issues in Interpretation
Multiple Comparisons
A posteriori contrasts are made after the analysis.
These are generally multiple comparison tests. They
enable the researcher to construct generalized
confidence intervals that can be used to make pairwise
comparisons of all treatment means. These tests, listed
in order of decreasing power, include least significant
difference, Duncan's multiple range test, Student-
Newman-Keuls, Tukey's alternate procedure, honestly
significant difference, modified least significant
difference, and Scheffe's test. Of these tests, least
significant difference is the most powerful, Scheffe's the
most conservative.
 Repeated Measures ANOVA
One way of controlling the differences between subjects
is by observing each subject under each experimental
condition (see Table 16.7). Since repeated
measurements are obtained from each respondent, this
design is referred to as within-subjects design or
repeated measures analysis of variance. Repeated
measures analysis of variance may be thought of as an
extension of the paired-samples t test to the case of
more than two related samples.
 Decomposition of the Total Variation:
Repeated
Table 16.6 Measures ANOVA
Independent Variable X
Subject Categories Total
No. Sample
X1 X2 X3 Xc
Between 1 Y11 Y12 Y13 Y1c Y1
People
Variation Total
2 Y21 Y22 Y23 Y2c Y2 Variation
=SSbetween : :
people : : =SSy

n Yn1 Yn2 Yn3 Ync YN

Category Y1 Y2 Y3 Yc Y
Mean

Within People Category Variation = SSwithin people

Repeated Measures ANOVA
In the case of a single factor with repeated measures, the
total variation, with nc - 1 degrees of freedom, may be
split into between-people variation and within-people
variation.

SStotal = SSbetween people + SSwithin people

The between-people variation, which is related to the

differences between the means of people, has n - 1
degrees of freedom. The within-people variation has
n (c - 1) degrees of freedom. The within-people variation
may, in turn, be divided into two different sources of
variation. One source is related to the differences
between treatment means, and the second consists of
residual or error variation. The degrees of freedom
corresponding to the treatment variation are c - 1, and
those corresponding to residual variation are
(c - 1) (n -1).
 Repeated Measures ANOVA
Thus,

SSwithin people = SSx + SSerror

A test of the null hypothesis of equal means may now be

constructed in the usual way:
SS x /(c - 1) MS x
F= =
SS error/(n - 1) (c - 1) MS error

So far we have assumed that the dependent variable is

measured on an interval or ratio scale. If the dependent
variable is nonmetric, however, a different procedure
should be used.
 Nonmetric Analysis of Variance
Nonmetric analysis of variance examines the
difference in the central tendencies of more than two
groups when the dependent variable is measured
on an ordinal scale.
One such procedure is the k-sample median test.
As its name implies, this is an extension of the
median test for two groups, which was considered in
Chapter 15.
 Nonmetric Analysis of Variance
A more powerful test is the Kruskal-Wallis one way analysis
of variance. This is an extension of the Mann-Whitney test
(Chapter 15). This test also examines the difference in
medians. All cases from the k groups are ordered in a single
ranking. If the k populations are the same, the groups should
be similar in terms of ranks within each group. The rank sum
is calculated for each group. From these, the Kruskal-Wallis
H statistic, which has a chi-square distribution, is computed.
The Kruskal-Wallis test is more powerful than the k-sample
median test as it uses the rank value of each case, not merely
its location relative to the median. However, if there are a
large number of tied rankings in the data, the k-sample
median test may be a better choice.
Multivariate Analysis of Variance
Multivariate analysis of variance (MANOVA) is similar
to analysis of variance (ANOVA), except that instead of
one metric dependent variable, we have two or more.
In MANOVA, the null hypothesis is that the vectors of
means on multiple dependent variables are equal across
groups.
Multivariate analysis of variance is appropriate when
there are two or more dependent variables that are
correlated.
 SPSS Windows
One-way ANOVA can be efficiently performed using the
program COMPARE MEANS and then One-way ANOVA.
To select this procedure using SPSS for Windows click:

Analyze>Compare Means>One-Way ANOVA

N-way analysis of variance and analysis of covariance

can be performed using GENERAL LINEAR MODEL. To
select this procedure using SPSS for Windows click:

Analyze>General Linear Model>Univariate