Suppressor Variables in Path Models

Abstract

Suppressor variables are well known in the context of multiple regression analysis. Using several examples, the authors demonstrate that the different forms of the suppressor phenomenon described in the literature occur not only in prediction equations but also in the explanatory use of multiple regression, including structural equations models. Moreover, they show that the probability of their occurrence is relatively high in models with latent variables, in which the suppressed variable is corrected for measurement errors. Special attention will be paid to the two-wave model since this is particularly liable to the suppressor phenomenon. The occurrence of suppression in structural equations models is usually not foreseen and confronts researchers with problems of interpretation. The authors discuss definitions of the suppressor phenomenon, show how the unwary researcher can be warned against it, and present guidelines for the interpretation of the results.