A Bayesian Approach to Multilevel Mediation Models With Latent Variables

• In Event: Methods for Multilevel, Categorical, and Multidimensional Data

Abstract

Large sample size requirements are a limitation of multilevel structural equation modeling (MSEM) for mediation analysis with nested data structures. Recently, there has been growing interest in Bayesian methods for mediation analysis with small sample sizes; however, to date, no study has evaluated the impact of diffuse priors in Bayesian multilevel mediation analysis with latent variables when all observed variables are measured at the within-group level. This simulation study compares Bayesian and frequentist approaches in a 1-1-1 mediation model under various design conditions. Results suggest that Bayesian methods with diffuse priors do not provide a clear fix for small sample problems.

Authors

• Kelly D Edwards, University of Virginia

• Timothy R. Konold, University of Virginia