Psychological process has its locus within the individual organism. Ideally, therefore, each subject would be tested under all experimental conditions. The main benefit is substantive: Comparisons across conditions are more cogent because the main effect of individual differences is factored out, not confounded with conditions, as in the independent groups designs of previous chapters. To study patterning in response, moreover, within-individual comparison may be essential.
A secondary benefit is also obtained: The error term is markedly lower, which yields shorter confidence intervals and greater power. Such repeated measures design is the focus of this chapter.
A complication with repeated measures design is loss of independence: Response to any two treatments is correlated across subjects. Because of this dependence, the error term used in previous chapters is no longer correct. Fortunately, the Anova model is readily extended to handle this dependence. Unfortunately, this extension is sensitive to the sphericity assumption on which it rests. Almost miraculously, a simple adjustment has been found that allows valid significance tests even when sphericity is violated.
All material of previous chapters transfers essentially unchanged to repeated measures design. The main novelty is that each analysis or comparison generally requires its own personal error term. Instead of a single error term, as in previous chapters, multiple error terms are often required. This complication is more than repaid, however, by the greater informativeness and greater power of repeated measures design.