You are on page 1of 30

# Chapter 11: Analysis of Variance and Design of Experiments

Chapter 11
Analysis of Variance and
Design of Experiments
LEARNING OBJECTIVES

The focus of this chapter is learning about the design of experiments and the analysis of
variance thereby enabling you to:
1.

Understand the differences between various experiment designs and when to use
them.

2.

3.

4.

5.

6.

## CHAPTER TEACHING STRATEGY

This important chapter opens the door for students to a broader view of statistics
than they have seen to this time. Through the topic of experimental designs, the student
begins to understand how they can scientifically set up controlled experiments in which
to test certain hypotheses. They learn about independent and dependent variables. With
the completely randomized design, the student can see how the t test for two independent
samples can be expanded to include three or more samples by using analysis of variance.
This is something that some of the more curious students were probably wondering about
in chapter 10. Through the randomized block design and the factorial designs, the
student can understand how we can analyze not only multiple categories of one variable,
but we can simultaneously analyze multiple variables with several categories each. Thus,

## Chapter 11: Analysis of Variance and Design of Experiments

this chapter affords the instructor an opportunity to help the student develop a structure
for statistical analysis.
In this chapter, we emphasize that the total sum of squares in a given problem do
not change. In the completely randomized design, the total sums of squares are parceled
into between treatments sum of squares and error sum of squares. By using a blocking
design when there is significant blocking, the blocking effects are removed from the error
effects which reduces the size of the mean square error and can potentially create a more
powerful test of the treatment. A similar thing happens in the two-way factorial design
when one significant treatment variable siphons off sum of squares from the error term
that reduces the mean square error and creates potential for a more powerful test of the
other treatment variable.
In presenting the random block design in this chapter, the emphasis is on
determining if the F value for the treatment variable is significant or not. We have de
emphasized examining the F value of the blocking effects. However, if the blocking
effects are not significant, the random block design may be a less powerful analysis of the
treatment effects. If the blocking effects are not significant, even though the error sum of
squares is reduced, the mean square error might increase because the blocking effects
may reduce the degrees of freedom error in a proportional greater amount. This might
result in a smaller treatment F value than would occur in a completely randomized
design. We have shown the repeated measures design in the chapter as a special case of
the random block design.
In factorial designs, if there are multiple values in the cells, it is possible to
analyze interaction effects. Random block designs do not have multiple values in cells
and therefore interaction effects cannot be calculated. It is emphasized in this chapter
that if significant interaction occurs, then the main effects analysis are confounded and
should not be analyzed in the usual manner. There are various philosophies about how to
handle significant interaction but are beyond the scope of this chapter. The main factorial
example problem in the chapter was created to have no significant interaction so that the
student can learn how to analyze main effects. The demonstration problem has
significant interaction and these interactions are displayed graphically for the student to
see. You might consider taking this same problem and graphing the interactions using
Row effects along the x axis and graphing the Column means for the student to see.
There are a number of multiple comparison tests available. In this text, I selected
one of the more well-known tests, Tukey's HSD, in the case of equal sample sizes. When
sample sizes are unequal, a variation on Tukeys HSD, the Tukey-Kramer test, is used.
MINITAB uses the Tukey test as one of its options under multiple comparisons and uses
the Tukey-Kramer test for unequal sample sizes. Tukey's HSD is one of the more
powerful multiple comparison tests but protects less for Type I errors than some of the
other tests.

CHAPTER OUTLINE

## 11.1 Introduction to Design of Experiments

11.2 The Completely Randomized Design (One-Way ANOVA)
One-Way Analysis of Variance
Using the Computer for One-Way ANOVA
Comparison of F and t Values
11.3 Multiple Comparison Tests
Tukey's Honestly Significant Difference (HSD) Test: The Case of Equal Sample
Sizes
Using the Computer to Do Multiple Comparisons
Tukey-Kramer Procedure: The Case of Unequal Sample Sizes
11.4 The Randomized Block Design
Using the Computer to Analyze Randomized Block Designs
11.5 A Factorial Design (Two-Way ANOVA)
Factorial Designs with Two Treatments
Applications
Statistically Testing the Factorial Design
Interaction
Using a Computer to Do a Two-Way ANOVA

KEY TERMS
a posteriori
a priori
Analysis of Variance (ANOVA)
Blocking Variable
Classification Variables
Classifications
Completely Randomized Design
Concomitant Variables
Confounding Variables
Dependent Variable
Experimental Design
F Distribution
F Value
Factorial Design

Factors
Independent Variable
Interaction
Levels
Multiple Comparisons
One-way Analysis of Variance
Post-hoc
Randomized Block Design
Repeated Measures Design
Treatment Variable
Tukey-Kramer Procedure
Tukeys HSD Test
Two-way Analysis of Variance

## SOLUTIONS TO PROBLEMS IN CHAPTER 11

11.1 a) Time Period, Market Condition, Day of the Week, Season of the Year
b) Time Period - 4 P.M. to 5 P.M. and 5 P.M. to 6 P.M.
Market Condition - Bull Market and Bear Market
Day of the Week - Monday, Tuesday, Wednesday, Thursday, Friday
Season of the Year - Summer, Winter, Fall, Spring
c) Volume, Value of the Dow Jones Average, Earnings of Investment Houses

11.2 a) Type of 737, Age of the plane, Number of Landings per Week of the plane, City
that the plane is based
b) Type of 737 - Type I, Type II, Type III
Age of plane - 0-2 y, 3-5 y, 6-10 y, over 10 y
Number of Flights per Week - 0-5, 6-10, over 10
City - Dallas, Houston, Phoenix, Detroit
c) Average annual maintenance costs, Number of annual hours spent on
maintenance

11.3 a) Type of Card, Age of User, Economic Class of Cardholder, Geographic Region
b) Type of Card - Mastercard, Visa, Discover, American Express
Age of User - 21-25 y, 26-32 y, 33-40 y, 41-50 y, over 50
Economic Class - Lower, Middle, Upper
Geographic Region - NE, South, MW, West
c) Average number of card usages per person per month,
Average balance due on the card, Average per expenditure per person,
Number of cards possessed per person

11.4

Average dollar expenditure per day/night, Age of adult registering the family,
Number of days stay (consecutive)

11.5

Source

df

Treatment
Error

2
14

14.03 1.00

Total

16

36.24

= .05

SS

MS

## Critical F.05,2,14 = 3.74

Since the observed F = 11.07 > F.05,2,14 = 3.74, the decision is to reject the null
hypothesis.

11.6

Source

df

SS

MS

Treatment
Error

4
18

26.67 1.48

Total

22

= .01

## Critical F.01,4,18 = 4.58

120.43

Since the observed F = 15.82 > F.01,4,18 = 4.58, the decision is to reject the null
hypothesis.

11.7

Source

df

SS

MS

Treatment
Error

3
12

167.5 14.0

Total

15

711.8

= .01

## Critical F.01,3,12 = 5.95

Since the observed F = 13.00 > F.01,3,12 = 5.95, the decision is to reject the null
hypothesis.

11.8

Source

df

SS

MS

Treatment
Error

1
12

64.29
43.43

64.29 17.76
3.62

Total

13

107.71

= .05

## Critical F.05,1,12 = 4.75

Since the observed F = 17.76 > F.05,1,12 = 4.75, the decision is to reject the null
hypothesis.
Observed t value using t test:
1
n1 = 7
x 1 = 29
s1 2 = 3

t =

## (29 24.71) (0)

= 4.22
3(6) + (4.238)(6) 1 1
+
7+72
7 7

Also, t =

11.9

Source
Treatment
Error

Total

2
n2 = 7
x 2 = 24.71
s22 = 4.238

F = 17.76 = 4.214

SS
583.39
972.18
1,555.57

df

MS

4 145.8475
50 19.4436
54

F
7.50

## Chapter 11: Analysis of Variance and Design of Experiments

11.10 Source

SS

df

MS

4.82 3.03
4.887

Treatment
Error

29.64
68.42

2
14

Total

98.06

16

F.05,2,14 = 3.74
Since the observed F = 3.03 < F.05,2,14 = 3.74, the decision is to fail to reject
the null hypothesis

11.11 Source

df

SS

MS

Treatment
Error

3
15

.007076
.003503

.002359
.000234

10.10

Total

18

.010579

= .01

## Critical F.01,3,15 = 5.42

Since the observed F = 10.10 > F.01,3,15 = 5.42, the decision is to reject the null
hypothesis.

11.12 Source

df

SS

Treatment
Error

2
12

180700000
11699999

Total

14

192400000

= .01

## Critical F.01,2,12 = 6.93

MS

90350000
975000

92.67

Since the observed F = 92.67 > F.01,2,12 = 6.93, the decision is to reject the null
hypothesis.

11.13 Source

df

SS

MS

Treatment
Error

2
15

29.61
18.89

14.80
1.26

Total

17

48.50

= .05

## Critical F.05,2,15 = 3.68

F
11.76

Since the observed F = 11.76 > F.05,2,15 = 3.68, the decison is to reject the null
hypothesis.

11.14 Source

df

SS

MS

Treatment
Error

3
16

456630
220770

152210
13798

11.03

Total

19

677400

= .05

## Critical F.05,3,16 = 3.24

Since the observed F = 11.03 > F.05,3,16 = 3.24, the decision is to reject the null
hypothesis.

11.15 There are 4 treatment levels. The sample sizes are 18, 15, 21, and 11. The F
value is 2.95 with a p-value of .04. There is an overall significant difference at
alpha of .05. The means are 226.73, 238.79, 232.58, and 239.82.
11.16 The independent variable for this study was plant with five classification levels
(the five plants). There were a total of 43 workers who participated in the study.
The dependent variable was number of hours worked per week. An observed F
value of 3.10 was obtained with an associated p-value of .02659. With an alpha
of .05, there was a significant overall difference in the average number of hours
worked per week by plant. A cursory glance at the plant averages revealed that
workers at plant 3 averaged 61.47 hours per week (highest number) while workers
at plant 4 averaged 49.20 (lowest number).

11.17 C = 6

= .05

MSE = .3352

q.05,6,40 = 4.23
HSD = 4.23

n3 = 8

N = 46

x 3 = 15.85

n6 = 7

x 6 = 17.2

.3352 1 1
+ = 0.896
2 8 7

## x 3 x 6 = 15.85 17.21 = 1.36

Since 1.36 > 0.896, there is a significant difference between the means of
groups 3 and 6.

11.18 C = 4

n=6

MSE = 2.389
HSD = q

11.19 C = 3

N = 24

dferror = N - C = 24 - 4 = 20

q.05,4,20 = 3.96

MSE
2.389
= (3.96)
= 2.50
n
6

MSE = 1.0

q.05,3,14 = 3.70
HSD = 3.70

n1 = 6

= .05
n2 = 5

N = 17
x1=2

x 2 = 4.6

1.00 1 1
+ = 1.584
2 6 5

x1 x 2 = 2 4.6 = 2.6
Since 2.6 > 1.584, there is a significant difference between the means of
groups 1 and 2.

## 11.20 From problem 11.6,

n2 = 5

MSE = 1.48
= .01

n4 = 5

C=5

10

N = 23

q.01,5,23 = 5.29

HSD = 5.29

1.48 1 1
+ = 2.88
2 5 5

x 2 = 10

x 4 = 16

x 3 x 6 = 10 16 = 6

## Since 6 > 2.88, there is a significant difference in the means of

groups 2 and 4.

11.21 N = 16

n=4

HSD = q

C=4

N - C = 12

MSE = 14

q.01,4,12 = 5.50

MSE
14
= 5.50
= 10.29
n
4

x 1 = 115.25

x 2 = 125.25

x 3 = 131.5

x 4 = 122.5

x 1 and x 3 are the only pair that are significantly different using the
HSD test.

11.22 n = 7

C=2

MSE = 3.62

N = 14

= .05

q.05,2,12 = 3.08

HSD = q

MSE
3.62
= 3.08
= 2.215
n
7

N - C = 14 - 2 = 12

x 1 = 29 and x 2 = 24.71

Since x 1 - x 2 = 4.29 > HSD = 2.215, the decision is to reject the null
hypothesis.

11.23 C = 4

MSE = .000234

q.01,4,15 = 5.25

n1 = 4

= .01
n2 = 6

n3 = 5

11

N = 19
n4 = 4

## x 1 = 4.03, x 2 = 4.001667, x 3 = 3.974, x 4 = 4.005

HSD1,2 = 5.25

.000234 1 1
+ = .0367
2 4 6

HSD1,3 = 5.25

.000234 1 1
+ = .0381
2
4 5

HSD1,4 = 5.25

.000234 1 1
+ = .0402
2 4 4

HSD2,3 = 5.25

.000234 1 1
+ = .0344
2
6 5

HSD2,4 = 5.25

.000234 1 1
+ = .0367
2 6 4

HSD3,4 = 5.25

.000234 1 1
+ = .0381
2
5 4

x1 x 3 = .056
This is the only pair of means that are significantly different.

## Chapter 11: Analysis of Variance and Design of Experiments

11.24 = .01

k=3

n=5

MSE
n

= 5.04

HSD = q
x 1 = 28,400

N = 15

975,000
5

x 2 = 36,900

12

N - k = 12

MSE = 975,000

= 2,225.6

x 3 = 32,800

x 1 x 2 = 8,500

x1 x 3 = 4,400
x 2 x 3 = 4,100
Using Tukey's HSD, all three pairwise comparisons are significantly different.

11.25 = .05

C=3

q.05,3,15 = 3.67

N = 18
n1 = 5

N-C = 15
n2 = 7

n3 = 6

x 1 = 7.6

x 2 = 8.8571

HSD1,2 = 3.67

1.26 1 1
+ = 1.706
2 5 7

HSD1,3 = 3.67

1.26 1 1
+ = 1.764
2 7 6

HSD2,3 = 3.67

1.26 1 1
+ = 1.621
2 7 6

## x1 x 3 = 1.767 (is significant)

x 2 x 3 = 3.024 (is significant)

MSE = 1.26

x 3 = 5.83333

## Chapter 11: Analysis of Variance and Design of Experiments

11.26 = .05
x 1 = 591

HSD = q

n=5

C=4

x 2 = 350

N = 20

13

N - C = 16

x 3 = 776

MSE = 13,798

x 4 = 563

MSE
13,798
= 4.05
= 212.75
n
5

x1 x 2 = 241

x1 x 3 = 185

x1 x 4 = 28

x 2 x 3 = 426

x 2 x 4 = 213

x 3 x 4 = 213

Using Tukey's HSD = 212.75, only means 1 and 2 and means 2 and 3 are
significantly different.

11.27

= .05
There were five plants and ten pairwise comparisons. The MINITAB
output revealed that the only pairwise significant difference was between plant 2
and plant 3. The reported confidence interval went from 22.46 to 0.18 which
contains the same sign indicating that 0 is not in the interval.

11.28 H0: 1 = 2 = 3 = 4
Ha: At least one treatment mean is different from the others
Source

df

SS

MS

Treatment
Blocks
Error

3
4
12

62.95
257.50
45.30

Total

19

365.75

= .05

## Critical F.05,3,12 = 3.49 for treatments

20.98 5.56
64.38 17.07
3.77

For treatments, the observed F = 5.56 > F.05,3,12 = 3.49, the decision is to
reject the null hypothesis.

## Chapter 11: Analysis of Variance and Design of Experiments

14

11.29 H0: 1 = 2 = 3
Ha: At least one treatment mean is different from the others
Source

df

SS

MS

Treatment 2
Blocks
3
Error
6

.001717 .000858
.076867 .025622
.003483 .000581

Total

.082067

= .01

11

F
1.48
44.10

## Critical F.01,2,6 = 10.92 for treatments

For treatments, the observed F = 1.48 < F.01,2,6 = 10.92 and the decision is to
fail to reject the null hypothesis.

11.30 Source

df

SS

MS
495.506
353.387
259.142

Treatment
Blocks
Error

5
9
45

2477.53
3180.48
11661.38

1.91
1.36

Total

59

17319.39

= .05

## Critical F.05,5,45 = 2.45 for treatments

For treatments, the observed F = 1.91 < F.05,5,45 = 2.45 and decision is to fail to
reject the null hypothesis.

11.31 Source

df

SS

MS

Treatment
Blocks
Error

3
6
18

199.48
265.24
306.59

66.493
44.207
17.033

3.90
2.60

Total

27

771.31

= .01

## Critical F.01,3,18 = 5.09 for treatments

For treatments, the observed F = 3.90 < F.01,3,18 = 5.09 and the decision is to
fail to reject the null hypothesis.

11.32 Source

df

SS

MS

Treatment
Blocks
Error

3
9
27

2302.5
5402.5
1322.5

767.5
600.3
49.0

15.66
12.25

Total

39

9027.5

= .05

## Critical F.05,3,27 = 2.96 for treatments

15

For treatments, the observed F = 15.66 > F.05,3,27 = 2.96 and the decision is to
reject the null hypothesis.

11.33 Source
Treatment
Blocks
Error

df

SS

MS

2
4
8

64.53
137.60
16.80

32.27 15.37
34.40 16.38
2.10

Total

14

218.93

= .01

## Critical F.01,2,8 = 8.65 for treatments

For treatments, the observed F = 15.37 > F.01,2,8 = 8.65 and the decision is to
reject the null hypothesis.

11.34 This a randomized block design with 3 treatments (machines) and 5 block levels
(operators). The F for treatments is 6.72 with a p-value of .019. There is a
significant difference in machines at = .05. The F for blocking effects is 0.22
with a p-value of .807. There are no significant blocking effects. The blocking
effects reduced the power of the treatment effects since the blocking effects were
not significant.

## Chapter 11: Analysis of Variance and Design of Experiments

16

11.35 The p value for columns, .000177, indicates that there is an overall significant
difference in treatment means at alpha .001. The lengths of calls differ according
to type of telephone used. The p value for rows, .000281, indicates that there is
an overall difference in block means at alpha .001. The lengths of calls differ
according to manager. While the output does not include multiple comparisons,
an examination of the column means reveals that calls were longest on cordless
phones (a sample average of 4.112 minutes) and shortest on computer phones (a
sample average of 2.522 minutes). Manager 5 posted the largest average call
length in the sample with an average of 4.34 minutes. Manager 3 had the shortest
average call length in the sample with an average of 2.275.
If the company wants to reduce the average phone call length, it would encourage
managers to use the computer for calls. However, this study might be
underscoring the fact that it is inconvenient to place calls using the computer. If
management wants to encourage more calling, they might make more cordless
phones available or inquiry as to why the other modes are used for shorter
lengths.
11.36 This is a two-way factorial design with two independent variables and one
dependent variable. It is 2x4 in that there are two row treatment levels and four
column treatment levels. Since there are three measurements per cell, interaction
can be analyzed.
dfrow treatment = 1
dferror = 16

dfcolumn treatment = 3

dfinteraction = 3,

dftotal = 23

11.37 This is a two-way factorial design with two independent variables and one
dependent variable. It is 4x3 in that there are four treatment levels and three
column treatment levels. Since there are two measurements per cell, interaction
can be analyzed.
dfrow treatment = 3
dferror = 12

dfcolumn treatment = 2
dftotal = 23

dfinteraction = 6,

11.38 Source

df

SS

MS

Row
Column
Interaction
Error

3
4
12
60

126.98
37.49
380.82
733.65

42.327
9.373
31.735
12.228

3.46
0.77
2.60

Total

79

1278.94

= .05

## Critical F.05,3,60 = 2.76 for rows

17

For rows, the observed F = 3.46 > F.05,3,60 = 2.76 and the decision is to reject the
null hypothesis.
Critical F.05,4,60 = 2.53 for columns
For columns, the observed F = 0.77 < F.05,4,60 = 2.53 and the decision is to fail
to reject the null hypothesis.
Critical F.05,12,60 = 1.92 for interaction
For interaction, the observed F = 2.60 > F.05,12,60 = 1.92 and the decision is to
reject the null hypothesis.
Since there is significant interaction, the researcher should exercise extreme
caution in analyzing the "significant" row effects.

11.39 Source

df

SS

MS

Row
Column
Interaction
Error

1
3
3
16

1.047
3.844
0.773
6.968

1.047
1.281
0.258
0.436

2.40
2.94
0.59

Total

23

12.632

= .05

## Critical F.05,1,16 = 4.49 for rows

For rows, the observed F = 2.40 < F.05,1,16 = 4.49 and decision is to fail to reject
the null hypothesis.
Critical F.05,3,16 = 3.24 for columns
For columns, the observed F = 2.94 < F.05,3,16 = 3.24 and the decision is to
fail to reject the null hypothesis.

18

## Critical F.05,3,16 = 3.24 for interaction

For interaction, the observed F = 0.59 < F.05,3,16 = 3.24 and the decision is to
fail to reject the null hypothesis.

11.40 Source
Row
Column
Interaction
Error

df

SS

MS
60.750
7.000
1.000
1.583

1
2
2
6

60.750
14.000
2.000
9.500

38.37
4.42
0.63

Total

11

86.250

= .01

## Critical F.01,1,6 = 13.75 for rows

For rows, the observed F = 38.37 > F.01,1,6 = 13.75 and the decision is to reject
the null hypothesis.
Critical F.01,2,6 = 10.92 for columns
For columns, the observed F = 4.42 < F.01,2,6 = 10.92 and the decision is to fail to
reject the null hypothesis.
Critical F.01,2,6 = 10.92 for interaction
For interaction, the observed F = 0.63 < F.01,2,6 = 10.92 and the decision is to
fail to reject the null hypothesis.

11.41 Source

df

SS

MS

Row
Column
Interaction
Error

3
1
3
24

5.09844
1.24031
0.12094
0.46750

1.69948
1.24031
0.04031
0.01948

87.25
63.67
2.07

Total

31

6.92719

= .05

## Critical F.05,3,24 = 3.01 for rows

For rows, the observed F = 87.25 > F.05,3,24 = 3.01 and the decision is to reject the
null hypothesis.

19

## Critical F.05,1,24 = 4.26 for columns

For columns, the observed F = 63.67 > F.05,1,24 = 4.26 and the decision is to
reject the null hypothesis.
Critical F.05,3,24 = 3.01 for interaction
For interaction, the observed F = 2.07 < F.05,3,24 = 3.01 and the decision is to fail
to reject the null hypothesis.

11.42 Source

df

SS

MS

Row
Column
Interaction
Error

3
2
6
12

42.4583
49.0833
4.9167
11.5000

14.1528
24.5417
0.8194
0.9583

14.77
25.61
0.86

Total

23

107.9583

= .05

## Critical F.05,3,12 = 3.49 for rows

For rows, the observed F = 14.77 > F.05,3,12 = 3.49 and the decision is to reject the
null hypothesis.
Critical F.05,2,12 = 3.89 for columns
For columns, the observed F = 25.61 > F.05,2,12 = 3.89 and the decision is to reject
the null hypothesis.
Critical F.05,6,12 = 3.00 for interaction
For interaction, the observed F = 0.86 < F.05,6,12 = 3.00 and fail to reject the null
hypothesis.

11.43 Source

df

SS

MS

Row
Column
Interaction
Error

2
3
6
24

1736.22
1078.33
503.33
607.33

Total

35

3925.22

= .05

## Critical F.05,2,24 = 3.40 for rows

868.11
359.44
83.89
25.31

20

F
34.31
14.20
3.32

For rows, the observed F = 34.31 > F.05,2,24 = 3.40 and the decision is to reject the
null hypothesis.
Critical F.05,3,24 = 3.01 for columns
For columns, the observed F = 14.20 > F.05,3,24 = 3.01 and decision is to reject the
null hypothesis.
Critical F.05,6,24 = 2.51 for interaction
For interaction, the observed F = 3.32 > F.05,6,24 = 2.51 and the decision is to
reject the null hypothesis.

11.44 This two-way design has 3 row treatments and 5 column treatments. There are 45
total observations with 3 in each cell.
FR =

MS R 46.16
= 13.23
=
MS E
3.49

p-value = .000 and the decision is to reject the null hypothesis for rows.
FC =

MSC 249.70
= 71.57
=
MS E
3.49

p-value = .000 and the decision is to reject the null hypothesis for columns.
MS I 55.27
FI =
= 15.84
=
MS E
3.49

p-value = .000 and the decision is to reject the null hypothesis for interaction.
Because there is significant interaction, the analysis of main effects is
confounded. The graph of means displays the crossing patterns of the line
segments indicating the presence of interaction.

## Chapter 11: Analysis of Variance and Design of Experiments

21

11.45 The null hypotheses are that there are no interaction effects, that there are no
significant differences in the means of the valve openings by machine, and that
there are no significant differences in the means of the valve openings by shift.
Since the p-value for interaction effects is .876, there are no significant interaction
effects which is good since significant interaction effects would confound that
study. The p-value for columns (shifts) is .008 indicating that column effects are
significant at alpha of .01. There is a significant difference in the mean valve
opening according to shift. No multiple comparisons are given in the output.
However, an examination of the shift means indicates that the mean valve
opening on shift one was the largest at 6.47 followed by shift three with 6.3 and
shift two with 6.25. The p-value for rows (machines) was .937 which is not
significant.

11.46 This two-way factorial design has 3 rows and 3 columns with three observations
per cell. The observed F value for rows is 0.19, for columns is 1.19, and for
interaction is 1.40. Using an alpha of .05, the critical F value for rows and
columns (same df) is F2,18,.05 = 3.55. Neither the observed F value for rows nor
the observed F value for columns is significant. The critical F value for
interaction is F4,18,.05 = 2.93. There is no significant interaction.
11.47 Source
df
SS
MS
F
Treatment
3
66.69
22.23 8.82
Error
12
30.25
2.52
Total
15
96.94
= .05
Critical F.05,3,12 = 3.49
Since the treatment F = 8.82 > F.05,3,12 = 3.49, the decision is to reject the null
hypothesis.
For Tukey's HSD:
MSE = 2.52

n=4

N = 16

N - k = 12

k=4

q.05,4,12 = 4.20
HSD = q

x 1 = 12

MSE
2.52
= 3.33
= (4.20)
n
4

x 2 = 7.75

x 3 = 13.25

x 4 = 11.25

## Using HSD of 3.33, there are significant pairwise differences between

means 1 and 2, means 2 and 3, and means 2 and 4.

11.48 Source

df

SS

Treatment
Error

6
19

68.19
249.61

Total

25

317.80

11.49 Source

MS

11.365
13.137

0.87

df

SS

MS

Treatment
Error

5
36

210
655

42.000
18.194

2.31

Total

41

865

SS

MS

75.46
4.66

11.50 Source

df

Treatment
Error

2
22

150.91
102.53

Total

24

253.44

= .01

## Critical F.01,2,22 = 5.72

22

16.19

Since the observed F = 16.19 > F.01,2,22 = 5.72, the decision is to reject the null
hypothesis.

x 1 = 9.200
n1 = 10
MSE = 4.66
= .01

x 2 = 14.250
x 3 = 8.714
n2 = 8
n3 = 7
C=3
N = 25
N - C = 22
q.01,3,22 = 4.64

HSD1,2 = 4.64

4.66 1 1
+ = 3.36
2 10 8

HSD1,3 = 4.64

4.66 1 1
+ = 3.49
2 10 7

HSD2,3 = 4.64

4.66 1 1
+ = 3.14
2 8 7

x 1 x 2 = 5.05 and

## Chapter 11: Analysis of Variance and Design of Experiments

23

11.51 This design is a repeated-measures type random block design. There is one
treatment variable with three levels. There is one blocking variable with six
people in it (six levels). The degrees of freedom treatment are two. The degrees
of freedom block are five. The error degrees of freedom are ten. The total
degrees of freedom are seventeen. There is one dependent variable.

11.52 Source

df

SS

MS

Treatment
Blocks
Error

3
9
27

20,994
16,453
33,891

Total

39

71,338

= .05

## Critical F.05,3,27 = 2.96 for treatments

6998.00 5.58
1828.11 1.46
1255.22

Since the calculated F = 5.58 > F.05,3,27 = 2.96 for treatments, the decision is to
reject the null hypothesis.

11.53 Source

df

SS

MS

Treatment
Blocks
Error

3
5
15

240.12
548.71
38.12

80.04
109.74
2.54

31.51
43.20

Total

23

= .05

## Critical F.05,3,5 = 5.41 for treatments

Since for treatments the calculated F = 31.51 > F.05,3,5 = 5.41, the decision is to
reject the null hypothesis.
For Tukey's HSD:
Ignoring the blocking effects, the sum of squares blocking and sum of squares
error are combined together for a new SSerror = 548.71 + 38.12 = 586.83.
Combining the degrees of freedom error and blocking yields a new dferror = 20.
Using these new figures, we compute a new mean square error, MSE =
(586.83/20) = 29.3415.
n=6

C=4

N = 24

N - C = 20

## Chapter 11: Analysis of Variance and Design of Experiments

24

q.05,4,20 = 3.96
HSD = q

MSE
= (3.96)
n

x 1 = 16.667

29.3415
= 8.757
6

x 2 = 12.333

x 3 = 12.333

x 4 = 19.833

None of the pairs of means are significantly different using Tukey's HSD = 8.757.
This may be due in part to the fact that we compared means by folding the
blocking effects back into error. However, the blocking effects were highly
significant.

11.54 Source

df

SS

MS

Treatment 1
Treatment 2
Interaction
Error

4
1
4
30

29.13
7.2825 1.98
12.67 12.6700 3.45
73.49 18.3725 5.00
110.30
3.6767

Total

39

225.59

= .05

## Critical F.05,4,30 = 2.69 for treatment 1

For treatment 1, the observed F = 1.98 < F.05,4,30 = 2.69 and the decision is to
fail to reject the null hypothesis.
Critical F.05,1,30 = 4.17 for treatment 2
For treatment 2 observed F = 3.45 < F.05,1,30 = 4.17 and the decision is to
fail to reject the null hypothesis.
Critical F.05,4,30 = 2.69 for interaction
For interaction, the observed F = 5.00 > F.05,4,30 = 2.69 and the decision is to
reject the null hypothesis.
Since there are significant interaction effects, examination of the main effects
should not be done in the usual manner. However, in this case, there are no
significant treatment effects anyway.

11.55 Source

df

SS

MS

Row
Column
Interaction
Error

3
2
6
24

257.889
1.056
17.611
54.000

Total

35

330.556

= .01

## Critical F.01,3,24 = 4.72 for rows

85.963
0.528
2.935
2.250

25

F
38.21
0.23
1.30

For the row effects, the observed F = 38.21 > F.01,3,24 = 4.72 and the decision is to
reject the null hypothesis.
Critical F.01,2,24 = 5.61 for columns
For the column effects, the observed F = 0.23 < F.01,2,24 = 5.61 and the decision is
to fail to reject the null hypothesis.
Critical F.01,6,24 = 3.67 for interaction
For the interaction effects, the observed F = 1.30 < F.01,6,24 = 3.67 and the decision
is to fail to reject the null hypothesis.

11.56 Source

df

SS

MS

Row
Column
Interaction
Error

2
3
6
24

49.3889
1.2222
1.2778
15.3333

Total

35

67.2222

= .05

## Critical F.05,2,24 = 3.40 for rows

24.6944 38.65
0.4074 0.64
0.2130 0.33
0.6389

For the row effects, the observed F = 38.65 > F.05,2,24 = 3.40 and the decision is to
reject the null hypothesis.
Critical F.05,3,24 = 3.01 for columns
For the column effects, the observed F = 0.64 < F.05,3,24 = 3.01 and the decision is
to fail to reject the null hypothesis.

26

## Critical F.05,6,24 = 2.51 for interaction

For interaction effects, the observed F = 0.33 < F.05,6,24 = 2.51 and the decision is
to fail to reject the null hypothesis.
There are no significant interaction effects. Only the row effects are significant.
Computing Tukey's HSD for rows:
x 1 = 2.667

x 2 = 4.917

n = 12

N = 36 N - k = 33

k=3

x 3 = 2.250

## MSE is recomputed by folding together the interaction and column sum of

squares and degrees of freedom with previous error terms:
MSE = (1.2222 + 1.2778 + 15.3333)/(3 + 6 + 24) = 0.5404
q.05,3,33 = 3.49
HSD = q

MSE
= (3.49)
n

.5404
= 0.7406
12

Using HSD, there are significant pairwise differences between means 1 and 2 and
between means 2 and 3.
Shown below is a graph of the interaction using the cell means by row.

11.57 Source

df

SS

MS

Treatment
Error

3
20

90.48
81.76

30.16
4.09

7.38

Total

23

172.24

= .05

## Critical F.05,3,20 = 3.10

27

The treatment F = 7.38 > F.05,3,20 = 3.10 and the decision is to reject the null
hypothesis.

11.58

Source

df

SS

MS

Treatment
Blocks
Error

2
5
10

460,353
33,524
22,197

Total

17

516,074

= .01

## Critical F.05,2,10 = 4.10 for treatments

230,176 103.70
6,705
3.02
2,220

Since the treatment observed F = 103.70 > F.05,2,10 = 4.10, the decision is to
reject the null hypothesis.

11.59 Source

df

SS

MS

Treatment
Error

2
18

9.4
185.4

4.7
10.3

0.46

Total

20

194.8

= .05

## Critical F.05,2,18 = 3.55

Since the treatment F = 0.46 > F.05,2,18 = 3.55, the decision is to fail to reject the
null hypothesis.
Since there are no significant treatment effects, it would make no sense to
compute Tukey-Kramer values and do pairwise comparisons.

11.60 Source

df

SS

MS

Row
Column
Interaction
Error

2
3
6
36

4.875
17.083
2.292
17.000

2.437
5.694
0.382
0.472

5.16
12.06
0.81

Total

47

41.250

= .05

## Critical F.05,2,36 = 3.32 for rows

28

For rows, the observed F = 5.16 > F.05,2,36 = 3.32 and the decision is to reject the
null hypothesis.
Critical F.05,3,36 = 2.92 for columns
For columns, the observed F = 12.06 > F.05,3,36 = 2.92 and the decision is to reject
the null hypothesis.
Critical F.05,6,36 = 2.42 for interaction
For interaction, the observed F = 0.81 < F.05,6,36 = 2.42 and the decision is to fail
to reject the null hypothesis.
There are no significant interaction effects. There are significant row and column
effects at = .05.

11.61 Source

df

SS

MS

Treatment
Blocks
Error

4
7
28

53.400
17.100
27.400

13.350
2.443
0.979

13.64
2.50

Total

39

97.900

= .05

## Critical F.05,4,28 = 2.71 for treatments

For treatments, the observed F = 13.64 > F.05,4,28 = 2.71 and the decision is to
reject the null hypothesis.

## Chapter 11: Analysis of Variance and Design of Experiments

29

11.62 This is a one-way ANOVA with four treatment levels. There are 36 observations
in the study. An examination of the mean analysis shows that the sample sizes are
different with sizes of 8, 7, 11, and 10 respectively. The p-value of .045 indicates
that there is a significant overall difference in the means at = .05. No multiple
comparison technique was used here to conduct pairwise comparisons. However,
a study of sample means shows that the two most extreme means are from levels
one and four. These two means would be the most likely candidates for multiple
comparison tests.

11.63 Excel reports that this is a two-factor design without replication indicating that
this is a random block design. Neither the row nor the column p-values are less
than .05 indicating that there are no significant treatment or blocking effects in
this study. Also displayed in the output to underscore this conclusion are the
observed and critical F values for both treatments and blocking. In both
cases, the observed value is less than the critical value.

11.64 This is a two-way ANOVA with 5 rows and 2 columns. There are 2 observations
per cell. For rows, FR = 0.98 with a p-value of .461 which is not significant. For
columns, FC = 2.67 with a p-value of .134 which is not significant. For
interaction, FI = 4.65 with a p-value of .022 which is significant at = .05. Thus,
there are significant interaction and the row and column effects are confounded.
An examination of the interaction plot reveals that most of the lines cross
signifying verifying the significant interaction finding.

11.65 This is a two-way ANOVA with 4 rows and 3 columns. There are 3 observations
per cell. FR = 4.30 with a p-value of .014 is significant at = .05. The null
hypothesis is rejected for rows. FC = 0.53 with a p-value of .594 is not
significant. We fail to reject the null hypothesis for columns. FI = 0.99 with a
p-value of .453 is not significant. We fail to reject the null hypothesis for
interaction effects.

11.66 This was a random block design with 5 treatment levels and 5 blocking levels.
For both treatment and blocking effects, the critical value is F.05,4,16 = 3.01. The
observed F value for treatment effects is MSC / MSE = 35.98 / 7.36 = 4.89 which
is greater than the critical value. The null hypothesis for treatments is rejected,
and we conclude that there is a significant different in treatment means. No
multiple comparisons have been computed in the output. The observed F value
for blocking effects is MSR / MSE = 10.36 /7.36 = 1.41 which is less than the
critical value. There are no significant blocking effects. Using random block
design on this experiment might have cost a loss of power.

## Chapter 11: Analysis of Variance and Design of Experiments

30

11.67 This one-way ANOVA has 4 treatment levels and 24 observations. The F = 3.51
yields a p-value of .034 indicating significance at = .05. Since the sample sizes
are equal, Tukeys HSD is used to make multiple comparisons. The computer
output shows that means 1 and 3 are the only pairs that are significantly different
(same signs in confidence interval). Observe on the graph that the confidence
intervals for means 1 and 3 barely overlap.