Moderator Variables in Personality Research: The Problem of Controlling for Plausible Alternatives

Abstract

Alternative variable problems are common in personality research. Given a restriction to correlational designs, it is always possible that a particular personality variable is related to a particular behavior because both are related to some third, unmeasured construct. This being the case, researchers often use designs that involve the measurement of potential alternatives and analysis of covariance or multiple regression to statistically "control" their influence. Unfortunately, when using these techniques to control for the moderating influence of an alternative variable, some rearchers appear to use the wrong covariate. These researchers enter a linear covariate when in fact the postulation of an alternative moderator requires the entry of a covariate interaction. The authors elaborate the logic of this argument and provide an empirical example.