Assessing Common-Metric Effect Sizes to Refine Mediation Models

Abstract

Mediation analysis tests X → M → Y processes in which an independent variable ( X) exerts an indirect effect on a dependent variable ( Y) through its influence on an intervening or mediator variable ( M). A preponderance of mediation studies, however, focuses on determining solely whether mediation effects are statistically significant, instead of focusing on what the results tell us about potential theoretical refinements in the mediation model. We argue in favor of employing a set of three standardized effect sizes based on variance proportions that allow researchers to compare their results with those of other mediation studies employing similar combinations of X, M, and Y variables. These standardized effect sizes constitute a set of common metrics signaling potential gaps in a mediation model, and as such provide useful insights for the theoretical refinement of mediation models in organizational research. We illustrate the utility of comparing these common-metric effect sizes using the examples of abusive and transformational leadership effects on employee outcomes as transmitted by social exchange quality.