Moderated Multiple Regression Tests are Criterion Specific

Abstract

This article examines how the ordering of variables impacts conclusions regarding the presence of a significant interactive relationship in moderated multiple regression (MMR). It is argued that whenever the designation of a criterion is arbitrary, researchers should consider performing two MMR analyses: one in which y is treated as the criterion, x the predictor, and m the moderator and a second in which x is the criterion and y the predictor. An example illustrates that these analyses are not symmetrical and that a significant interaction may be observed in one case but not the other. A simulation further illustrates conditions under which such effects are likely to occur. Implications for researchers studying interactions are discussed.