Effect size measures in mediation analysis: New and Old, What is Good?

Mediation analysis plays an important role in understanding causal processes in social and behavioral sciences. Proper measures of the size of a mediated effect are central to advancing the understanding of the role of mediation/intervention in any study. Many measures of effect size (ES) have been proposed in mediation analysis. This article examines the goodness of these measures in three aspects: (1) content validity, (2) ES-consistency, and (3) range and distribution across independent conditions of permissible parameter values. In addition, by differentiating the role of the mediator from that of the predictor in accounting for the variance of the outcome variable, a new R-square measure is proposed to truthfully reflect the percentage of variance being mediated. The applicability of the new R-square is extended to inconsistent mediation as well. The article aims to build the rationale for choosing proper ES measures, and the results are expected to clarify the pros and cons of new and old ES measures of mediated effects. Logic reasoning and consideration developed in this article can be applied to studying the goodness of ES measures with more complicated mediation processes as well.