Specific Interaction Contrasts: A Statistical Tool for Repeated-Measures Designs

摘要

Experimental psychologists often use multifactor repeated-measure designs in which interactions are the most important effects to be assessed. An experimenter has at least five ways to evaluate such interactions: a univariate repeated-measures analysis of variance (ANOVA), with (probably) inflated estimates of the degrees of freedom; a univariate repeated-measures ANOVA with the Greenhouse-Geisser conservative estimate of the degrees of freedom; the Greenhouse-Geisser stepwise analysis; a multivariate ANOVA; and specific interaction contrasts. We show that no matter which of the above paths is chosen, the careful experimenter must compute specific interaction contrasts (i.e., t-tests). A worked example is given. (Author)