A Novel Measure of Effect Size for Mediation Analysis

Abstract

Mediation analysis has become one of the most popular statistical methods in the social sciences. However, currently available effect size measures for mediation have limitations that restrict their usage to special cases. The goal of this research was to develop a general measure of effect size for mediation analysis that was applicable to a wide variety of models while maintaining the properties of a good effect size estimate. It will be shown how modification of a currently existing effect size measure for simple three-variable mediation models results in a novel effect size measure with properties suitable for general applicability. A matrix method will be proposed for obtaining the measure from complex mediation models and its use will be demonstrated for special cases. In addition, a novel method of variance decomposition using the measure will be proposed which allows for unique and shared contributions of sets of predictors and mediators to be examined in complex mediation models. Finally, two simulation studies will be conducted to examine 1) the properties of the statistic across various population models and 2) the finite sampling properties of the statistic in a multiple mediator model.