Appraisal of the Homogeneity of Error Variance Assumption and Alternatives to Multiple Regression for Estimating Moderating Effects of Categorical Variables

Abstract

Homogeneity of within-subgroup error variance is a necessary condition for using multiple regression to estimate moderating effects of categorical variables. A 12-year review of Academy of Management Journal, Journal of Applied Psychology , and Personnel Psychology indicates that the assumption is violated in 40% to 60% of cases. The authors reanalyze published research to illustrate how violating the assumption may change substantive conclusions. To remedy this situation, they develop and present a computer program (i.e., ALTMMR) that (a) assesses whether a data set suffers from heterogeneity of error variance and (b) computes alternative inferential statistics to the traditional multiple regression F test when heterogeneity exists. ALTMMR, which can also be used as a teaching tool, was written in Java and is executable using an Internet Web browser or as a standalone application.