The young researcher's guide to moderator variables.

Most researchers are familiar with confounding variables, and the need to define and measure these variables at the time of conducting a study. Researchers also need to be familiar with moderator variables. A moderator variable is one that influences the relationship between a predictor variable and an outcome. Expressed otherwise, if the strength (or direction) of the relationship between a predictor variable and an outcome is different at different levels of a third variable, that third variable is a moderator variable. This article uses a simple example from a published study to explain how age can be a moderator variable. Other examples of moderator variables are presented. Moderator variables should be postulated based on existing evidence and the researcher's judgement, and are studied by including a (Predictor Variable) x (Moderator Variable) interaction term in regression. The use and interpretation of this interaction is explained. Inclusion of interaction terms increases the sample size (or event count, for categorical variables) required to avoid overfitting in regression. The study of moderator variables is necessary for better theoretical understanding of a field, for treatment optimization, and to identify subgroups that require attention in research and practice.