Chapter 11: Analysis of Variance

February 14, 2022

1 11.1 The Completely Randomized Design: One-Way ANOVA

2 11.2 The Factorial Design: Two-Way ANOVA

Example 1. The management of a general merchandiser might be
brainstorming ways of improving sales of mobile electronics items.
At Arlingtons, the management decided to try selling those items
in four different in-store locations and then observe what the sales
would be in each of those locations.

The analysis of variance (ANOVA) is a technique that can be

used to compare three or more population means.
The criterion by which we classify the populations is called a
factor.
In Example 1, the factor is in-store location.
The actual different locations (in-aisle, special front, kiosk,
and expert counter) are the levels of the factor.
Different levels produce different groups.
In Example 1, the groups are the stores selling the mobile
electronics items in the aisle, the stores selling those items at
the special front location, the stores selling those items at the
kiosk, and the stores selling those items at the expert counter.
11.1 The Completely Randomized Design: One-Way
ANOVA

Completely Randomized Design

An experiment with only one factor.
Experimental units (subjects) are assigned randomly to
groups.
Analyzed by one-way ANOVA.
Assumptions in one-way ANOVA
Populations are normally distributed
Populations have equal variances
Samples are randomly and independently drawn

Hypotheses of one-Way ANOVA

H0 : µ 1 = µ 2 = . . . = µ c
H1 : Not all the population means are equal
where
n = the number of values in all groups
c = the number of groups
SSA measures differences from group to group
SSW measures random variation
SST = Sum of squares total

Note. n = n1 + n2 + . . . + nc
SSA = sum of squares among groups
c
X
SSA = nj (X j − X )2
j=1

where X j = sample mean of group j

SSW = sum of squares within groups
nj
c X
X
SSW = (Xij − X j )2
j=1 i=1
where
MSA = mean squares among groups.
MSW = mean squares within groups.

F Test for Differences Among More Than Two Means

MSA
FSTAT =
MSW
The FSTAT test statistic follows an F distribution, with c − 1
degrees of freedom in the numerator and n − c degrees of freedom
in the denominator.
Procedure for testing the hypothesis H0 : µ1 = µ2 = . . . = µk
MSA
1) Compute FSTAT =
MSW
2) If FSTAT > Fα , reject H0 , otherwise, do not reject H0 , where
Fα is the critical value from the F distribution with c − 1
degrees of freedom in the numerator and n − c in the
denominator
The results of a one-way ANOVA are usually reported in an
ANOVA summary table.
Example 2. One of the steps involved in the processing of corn
flakes for cereals involves toasting the flakes. The file ”CornFlakes”
contains the following data for corn flakes thickness (mm) for four
different toasting times (seconds).

At the 0.05 level of significance, is there evidence of a difference in

the mean thickness of the corn flakes for the different toasting
times?
Example 3. Many college and university students obtain summer
jobs. A statistics professor wanted to determine whether students
in different degree programs earn different amounts. A random
sample of 5 students in the B.A., B.Sc., and B.B.A. programs were
asked to report what they earned the previous summer. The results
(in $1,000s) are listed here.

Can the professor infer at the 5% significance level that students in

different degree programs differ in their summer earnings?
Homogeneity of Variance

With equal sample sizes, violations of this assumption do not

seriously affect inferences.
With unequal sample sizes, unequal variances can have a
serious effect on inferences.
Use the Levene test for homogeneity of variance
Levene Test for Homogeneity of Variance

Test the hypotheses

H0 : σ12 = σ22 = . . . = σc2

H1 : Not all the population variances are equal

Compute the absolute value of the difference between each

value and the median of each group.
Perform a one-way ANOVA on these absolute differences.

Example 4. Consider Example 2, at the 0.05 level of significance,

is there evidence of a difference in the variation in the thickness of
the corn flakes?
Multiple Comparisons: The Tukey-Kramer Procedure

When we conclude from the one-way ANOVA that not all the
population means are equal, we often need to know which
means are significantly different.
We can use the Tukey-Kramer multiple comparison procedure
for one-way ANOVA to answer this question.
1. For each pair of j, j ′ ∈ {1, 2, . . . , c}(j < j ′ ), compute
|X j − X j ′ |.
2. Compute the critical range for the Tukey-Kramer procedure.
3. If |X j − X j ′ | greater than the corresponding critical range, we
will conclude that µi and µj differ.
Formula for Computing Critical Range
s  
MSW 1 1
Critical range = Qα +
2 nj nj ′
where
nj = the sample size in group j
nj ′ = the sample size in group j ′
Qα = the upper-tail critical value from a Studentized range
distribution having c degrees of freedom in the numerator and
n − c degrees of freedom in the denominator (see Table E.7)

Example 5. Determine which degrees differ.

11.2 The Factorial Design: Two-Way ANOVA

Factorial Design
An experiment in which more than one factor is considered.
To analyze data from a two-factor factorial design, you use
two-way ANOVA.

There are three different tests to perform in a two-way ANOVA:

A test of the hypothesis of no difference due to factor A
A test of the hypothesis of no difference due to factor B
A test of the hypothesis of no interaction of factors A and B
r = number of levels of factor A
c = number of levels of factor B
n′ = number of values (replicates) for each cell
n = number of values in the entire experiment (n = rcn′ )
Xijk = value of the kth observation for level i of factor A and
level j of factor B
1) To test the hypothesis of no difference due to factor A
H0 : µ1.. = µ2.. = . . . = µr ..
H1 : Not all µi.. are equal

MSA
Compute FSTAT =
MSE
If FSTAT > Fα , reject H0 , where Fα is the upper-tail critical
value from an F distribution with r − 1 and rc(n′ − 1) degrees
of freedom.
2) To test the hypothesis of no difference due to factor B
H0 : µ.1. = µ.2. = . . . = µ.c.
H1 : Not all µ.j. are equal

MSB
Compute FSTAT =
MSE
If FSTAT > Fα , reject H0 , where Fα is the upper-tail critical
value from an F distribution with c − 1 and rc(n′ − 1) degrees
of freedom.
3) To test the hypothesis of no interaction of factors A and B
H0 : The interaction of A and B is equal to zero
H1 : The interaction of A and B is not equal to zero

MSAB
Compute FSTAT =
MSE
If FSTAT > Fα , reject H0 , where Fα is the upper-tail critical
value from an F distribution with (r − 1)(c − 1) and
rc(n′ − 1) degrees of freedom.
Analysis of Variance Table for the TwoFactor Factorial Design
Example 6. Return to the Arlingtons scenario
Explored how different in-store locations might affect the sales
of mobile electronics items using one-way ANOVA.
To explore the effects of permitting mobile payment methods
to buy mobile electronics items, you design an experiment
that examines this second (B) factor as it studies the effects
of in-store location (factor A) using two-way ANOVA.
Two-way ANOVA will allow you to determine if there is a
significant difference in mobile electronics sales among the
four in-store locations and whether permitting mobile
payment methods makes a difference.
To test the effects of the two factors, you conduct a 60-day
experiment at 40 same-sized stores that have similar storewide net
sales. You randomly assign ten stores to use the current in-aisle
location, ten stores to use the special front location, ten stores to
use the kiosk location, and ten stores to use the expert counter. In
five stores in each of the four groups, you permit mobile payment
methods (for the other five in each group, mobile payment
methods are not permitted). At the end of the experiment, you
organize the mobile electronics sales data by group and store the
data in the following table:
Exercises

11.5, 11.10, 11.12 (page 412-413)

11.15, 11.16, 11.19, 11.20, 11.25 (page 425)