Effect Size Measures for Mediation in Cluster-Randomized Trials

• In Event: Multilevel Modeling Methods and Applications

Abstract

Tests of mediation are associated with their own effect sizes, although a single effect size measure has not been embraced by the methodological community. MacKinnon (2008) describes three classes of effect size measures in a single-level context and Preacher and Kelley (2011) proposed two additional single-level measures. Hedges (2007) proposed effect size measures for use in cluster-randomized trials to quantify the total treatment effect. This proposed paper merges these two bodies of work to consider effect size measures of the mediated effect in cluster-randomized contexts. The paper will include a description of each index and demonstrate the application of each using empirical data. The reporting of effect sizes in multi-level mediation will support future meta-analytic work of mediation studies.

Authors

• Laura M. Stapleton, University of Maryland

• Keenan A. Pituch, The University of Texas - Austin