192 Chapter 5 Introduction to Factorial Designs

TA B L E 5 . 3
The Analysis of Variance Table for the Two-Factor Factorial, Fixed Effects Model

Source of Sum of Degrees of

Variation Squares Freedom Mean Square F0

SSA MSA
A treatments SSA a1 MSA  F0 
a1 MSE
SSB MSB
B treatments SSB b1 MSB  F0 
b1 MSE
SSAB MSAB
Interaction SSAB (a  1)(b  1) MSAB  F0 
(a  1)(b  1) MSE
SSE
Error SSE ab(n  1) MSE 
ab(n  1)

Total SST abn  1

The sums of squares for the main effects are

a y2...
SSA 
1
bn y
i1
2
i.. 
abn
(5.7)

and
b y2...
1
SSB  an y
j1
2
.j. 
abn
(5.8)

It is convenient to obtain the SSAB in two stages. First we compute the sum of squares between
the ab cell totals, which is called the sum of squares due to subtotals:
a b y2...
SSSubtotals  n1 y
i1 j1
2
ij. 
abn
This sum of squares also contains SSA and SSB. Therefore, the second step is to compute SSAB
as
SSAB  SSSubtotals  SSA  SSB (5.9)
We may compute SSE by subtraction as
SSE  SST  SSAB  SSA  SSB (5.10)
or
SSE  SST  SSSubtotals

EXAMPLE 5.1 The Battery Design Experiment

Table 5.4 presents the effective life (in hours) observed in row and column totals are shown in the margins of the
the battery design example described in Section 5.3.1. The table, and the circled numbers are the cell totals.
 5.3 The Two-Factor Factorial Design 193

TA B L E 5 . 4
Life Data (in hours) for the Battery Design Experiment
Temperature (F)
Material
Type 15 70 125 yi..

130 155 34 40 20 70
539 229 230
1 74 180 80 75 82 58 998
150 188 136 122 25 70
623 479 198
2 159 126 106 115 58 45 1300
138 110 174 120 96 104
576 583 342
3 168 160 150 139 82 60 1501
y.j. 1738 1291 770 3799  y...

Using Equations 5.6 through 5.10, the sums of squares are a b y2...
computed as follows: SSInteraction  n1 
i1 j1
y2ij. 
abn
 SSMaterial

a b n y2...  SSTemperature
SST  y
i1 j1 k1
2
ijk 
abn
 1 [(539)2  (229)2   (342)2]
 (130)2  (155)2  (74)2  4

(3799)2 (3799)2
 (60)2   77,646.97   10,683.72
36 36

a y2...  39,118.72  9613.78

SSMaterial 
1
bn 
i1
y2i.. 
abn and
1 [(998)2  (1300)2  (1501)2]
 SSE  SST  SSMaterial  SSTemperature  SSInteraction
(3)(4)

(3799)2  77,646.97  10,683.72  39,118.72

  10,683.72
36
 9613.78  18,230.75
b y2... The ANOVA is shown in Table 5.5. Because F0.05,4,27 
1
SSTemperature  an y
j1
2
.j. 
abn 2.73, we conclude that there is a signicant interaction
between material types and temperature. Furthermore,
 1 [(1738)2  (1291)2  (770)2] F0.05,2,27  3.35, so the main effects of material type and
(3)(4) temperature are also signicant. Table 5.5 also shows the P-
values for the test statistics.
(3799)2 To assist in interpreting the results of this experiment, it
  39,118.72
36 is helpful to construct a graph of the average responses at
194 Chapter 5 Introduction to Factorial Designs

TA B L E 5 . 5
Analysis of Variance for Battery Life Data

Source of Sum of Degrees of Mean

Variation Squares Freedom Square F0 P-Value
Material types 10,683.72 2 5,341.86 7.91 0.0020
Temperature 39,118.72 2 19,559.36 28.97 < 0.0001
Interaction 9,613.78 4 2,403.44 3.56 0.0186
Error 18,230.75 27 675.21
Total 77,646.97 35

each treatment combination. This graph is shown in Figure for types 1 and 2. From intermediate to high temperature,
5.9. The signicant interaction is indicated by the lack of battery life decreases for material types 2 and 3 and is
parallelism of the lines. In general, longer life is attained at essentially unchanged for type 1. Material type 3 seems to
low temperature, regardless of material type. Changing give the best results if we want less loss of effective life as
from low to intermediate temperature, battery life with the temperature changes.
material type 3 may actually increase, whereas it decreases

175

150
Average life yij.

125

100
Material type 3
75
Material type 1
50 Material type 2

25

0
15 70 125
Temperature (F)

FIGURE 5.9 Material typetemperature plot for

Example 5.1

Multiple Comparisons. When the ANOVA indicates that row or column means differ,
it is usually of interest to make comparisons between the individual row or column means to
discover the specic differences. The multiple comparison methods discussed in Chapter 3
are useful in this regard.
We now illustrate the use of Tukeys test on the battery life data in Example 5.1. Note
that in this experiment, interaction is significant. When interaction is significant, compar-
isons between the means of one factor (e.g., A) may be obscured by the AB interaction. One
approach to this situation is to fix factor B at a specific level and apply Tukeys test to the
means of factor A at that level. To illustrate, suppose that in Example 5.1 we are interested
in detecting differences among the means of the three material types. Because interaction
is significant, we make this comparison at just one level of temperature, say level 2 (70F).
We assume that the best estimate of the error variance is the MSE from the ANOVA table,