REPEATED MEASURES DESIGN
Using each subject in a number of experimental conditions has two attractions: one statistical, one substantive. The statistical attraction is that error variability is lower because subjects are their “own controls.” Comparison of treatments A1 and A2 compares two scores from each subject. The main effect of each subject cancels in the difference, thereby freeing the comparison from the main effect of individual differences.
Using different subjects in different conditions, as in previous chapters, confounds individual differences with conditions. Although randomization resolves this confounding, it does so at the high cost of putting the individual differences in the error term. MSerror typically runs several times smaller in repeated measures designs. Confidence intervals are correspondingly shorter and power is greater. From this statistical standpoint, comparing the same subject across different conditions is most desirable.
More important is the substantive consideration that psychological process has its locus within the individual. Many investigations focus on the pattern of response across a set of stimuli, but this pattern may be irretrievably confounded with individual differences in response to given stimuli. To study response pattern, therefore, within individual comparison is desirable, perhaps essential.
Response pattern was the concern in the blame experiment of Figure 1.3, page 27 of Chapter 1, which showed the factorial graph of blame as a function of actor's intent to harm and amount of harm done. In this task, different individuals may show different judgment patterns (e.g., Figure 11.3, page 318). A group graph would hardly be meaningful; repeated measures is essential.
Analogous concern with response pattern is common in every area of psychology, most notably in perception, but also in diverse other areas from person cognition to operant psychology. Such response patterns are ideally studied with repeated measures.