Male S1 5 8 6
(code=1) S2 4 8 6
S3 3 5 5
S4 6 10 9
S5 3 6 5
Female S6 6 5 7
(code=0) S7 4 4 4
S8 5 6 5
S9 4 5 5
S10 3 4 5
Three main hypotheses were tested:
- whether there was a difference between males and females;
- whether there was a difference in vocabulary score between the testing occasions (Time
1 to Time3); - whether there was an interaction between sex and time.
Computer AnalysisThe sources of variance in a split-plot design can be partitioned first into sums of squares
between-subjects and sums of squares within-subjects. Then the withinsubjects sums of
squares can be further partitioned into sums of squares for treatment effects (in this
example time), sums of squares for the interaction of between-subjects by treatment
factor (timesex in this example), and the withinerror sums of squares which is actually
the sums of squares for treatment by subjects within the between subjects factor,
(timesubj(sex)). The between-subjects partition of the data has its own sums of squares
error term, subjects within the between-subjects factor, (subj(sex)) which is independent
of the within-subjects effects. This partitioning of the sums of squares for the data shown
in Table 8.15 is illustrated in Figure 8.20.
Inferences involving continuous data 339