School of Psychology

Historical background: Moderation

Moderation refers to the examination of the statistical interaction between two independent variables in predicting a dependent variable. One can examine a statistical interaction between two or more IVs in ANOVA or MANOVA, of course, but these IVs will be categorical in nature (e.g., gender; ethnicity, etc.). My emphasis here is on moderation within regression-based analyses (instead of ANOVA), and this type of analysis involves two independent variables (at least one of which is continuous) in predicting a dependent variable

In my basic research methods course, I teach my students that they can convert a continuous variable (e.g., socio-economic status) into a categorical variable by enacting a median split on the variable and creating a dichotomous variable. Other teachers and researchers do this too, but there is a problem with this. Aiken and West (1991) have described the use of multiple regression as a method for investigating interactions between continuous variables, and they present an excellent case as to why this is preferable to converting a continuous variable into a categorical variable so that the analysis can be done in ANOVA. In short, one loses valuable mathematical information when one converts a continuous variable into a categorical one.

However much it is advantageous to examine interactions in this regression-based mode, researchers have been frustrated by the difficulty in probing significant interactions of this type. Current statistics programmes like SPSS do not provide an easy and fool-proof way to examine the nature of the interaction. Aiken and West, in their book, provide superlative guidance in how one might wish to probe the interaction, but these computations have generally been performed by hand, and researchers find that they make mistakes and the process is laborious.

In 2003 I wrote an Excel-based programme, and separately a web-based programme, named ModGraph, which permits researchers to take output information from regression analyses and quickly and accurately create a figure (using Aiken and West's suggestions). My goal was to expedite this process, and I feel that this programme accomplishes this objective. It is not the only approach or way, however. Kris Preacher ( and Andrew Hayes ( have written a variety of macros that graph interactions quickly and flexibly. Interested users may wish to examine these options too.