ASSESSING MODERATOR VARIABLES - 2 COMPUTER-SIMULATION STUDIES

The present article proposes a strategy for conceptualizing moderating relationships based on their type-strictly correlational and classica l correlational-and form-continuous, noncontinuous, logistic, and quan tum. The results of two computer simulation studies comparing the rela tive power of three statistical approaches for assessing moderator var iables are presented: moderated multiple regression, multiple regressi on with a dichotomized moderator, and correlational analysis. As predi cted, moderated multiple regression was generally found to be equal or superior to these alternative approaches at detecting moderator relat ionships. Although the alternative approaches did potentially offer gr eater statistical power under limited circumstances, this was only tru e in more extreme cases and required fairly accurate estimates of spec ific characteristics of the joint distribution of the predictor, moder ator, and outcome variables.