Finding and Developing Moderators and Directional Keys by Regression Analysis.

A procedure for rapid screening of variables as potential moderators is presented and discussed. A moderator is defined as any variable which can be used to identify differentially predictable persons; or defined statistically by stating that if a predictor and a moderator are each divided into three or more categories and used as independent variables to predict a criterion, then an F test on interaction effects will determine the presence of a moderator variable. However, in sets of data based on large numbers of variables, it is difficult to discover moderators. The method proposed begins by arranging the subjects in four quadrants, according to their scores on a predictor and the criterion, and coding the hit and miss cells as dummy variables (zero and one, respectively) and then computing the point-biserial correlation of the remaining variables with this dummy variable. In addition, one of the cells coded one may be changed to minus one, to represent underprediction, in order to develop a directional key for the moderators. (Author/CTM)