By far the easiest way to test these between-subjects effects is to run separate two-way
(Condition 3 Sex) analyses at each level of the Time variable. These four analyses will
give you all three simple effects at each Time with only minor effort. You can then accept
the Fvalues from these analyses, as I have done here for convenience, or you can pool the
error terms from the four separate analyses and use that pooled error term in testing the
mean square for the relevant effect. If these terms are heterogeneous, you would be wise
not to pool them. On the other hand, if they represent homogeneous sources of variance,
they may be pooled, giving you more degrees of freedom for error. For these effects you
don’t need to worry about sphericity because each simple effect is calculated on only one
level of the repeated-measures variable.
The within-subjects simple effects are handled in much the same way. For example,
there is some reason to look at the simple effects of Time for each Condition separately to
see whether the EC condition shows changes over time in the absence of a complete inter-
vention. Similarly, we would like to see how the BST condition changes with time. How-
ever, we want to include Sex as an effect in both of these analyses so as not to inflate the
error term unnecessarily. We also want to use a separate error term for each analysis, rather
than pooling these across Conditions.
The relevant analyses are presented in Table 14.9 for simple effects at one level of the
other variable. Tests at the other levels would be carried out in the same way. Although this
table has more simple effects than we care about, they are presented to illustrate the way in
which tests were constructed. You would probably be foolish to consider all of the tests that
result from this approach, because you would seriously inflate the familywise error rate.
Decide what you want to look at before you run the analyses, and then stick to that deci-
sion. If you really want to look at a large number of simple effects, consider adopting one
of the Bonferroni approaches discussed in Chapter 12.
From the between-subjects analysis in Table14.9a we see that at Time 1 (Pretest) there
was a significant difference between males and females (females show a lower frequency
of use). But there were no Condition effects nor was there a Condition 3 Sex interaction.
Males exceed females by just about the same amount in each Condition. The fact that there
is no Condition effect is reassuring, because it would not be comforting to find that our two
conditions differed before we had applied any treatment.
From the results in Table 14.9b we see that for the BST condition there is again a signif-
icant difference due to Sex, but there is no Time effect, nor a Time 3 Sex interaction. This
is discouraging: It tells us that when we average across Sex there is no change in frequency
of condom use as a result of our intervention. This runs counter to the conclusion that we
might have drawn from the overall analysis where we saw a significant Condition by TimeSection 14.8 Two Between-Subjects Variables and One Within-Subjects Variable 487At Sex of Subject = Male
At Sex of Subject = FemaleTime1 2 3 4Marginal Mean403020100Treatment condition
Behav. Skills Training
Educational ControlTreatment condition
Behav. Skills Training
Educational ControlMarginal Means403020100Time1 2 3 4Figure 14.3 Frequency of condom use as a function of Sex and Condition