On the Relative Power of Interaction Analysis

Abstract

It is argued that statements in the current literature suggesting that interaction effects are, in general, as easy to detect as main effects are misleading. Different effect definitions which produce different conclusions about the relative power of interaction analysis are considered for both factorial ANOVA and aptitude-treatment-interaction models. Based on what is defined as a reasonable specification of “comparable” effects, it is concluded that the power for simple main and interaction effects is, in general, lower than that for the analysis of main effects.