Problems with detecting assumed mediation using the hierarchical multiple regression strategy

Abstract

In human resource management (HRM) and allied fields (e.g., organizational behavior, management, and industrial and organizational psychology), tests of mediation are frequently conducted using the hierarchical multiple regression (HMR) strategy of Baron and Kenny [Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173–1182]. Although previous research has identified a number of serious problems with this approach, the present study adds to the literature by identifying yet additional problems with its use in inferring the existence of mediation. Using a statistical simulation, we found that certain patterns of correlation coefficients guarantee inferences about mediation, whereas other patterns preclude such inferences. On the basis of various analyses including logistic regression and inspection of three-dimensional plots, we identified patterns of correlation coefficients needed to satisfy Baron and Kenny's [Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173–1182] conditions for inferring mediation. The same patterns have no necessary relation to actual causal connections among variables in mediation models. Moreover, as a consequence of the failure of the HMR strategy to detect mediating effects, many instances of actual mediation in HRM and allied fields may have gone undetected. In view of the foregoing, we conclude that the HMR strategy should no longer be used in testing for mediation.