Bayesian Analysis of Heterogeneous Mediation
69 Pages Posted: 29 Apr 2015 Last revised: 17 Nov 2018
Date Written: November 17, 2018
Abstract
Mediation analysis is used to study the relationship between stimulus and response in the presence of intermediate, generative variables. The traditional approach to mediation analysis is based on the statistical significance of coefficients in aggregate regression models that assume homogeneity of responses. Our approach shows that the results from these models are neither necessary nor sufficient to claim the presence of mediation for specific individuals. The proposed Bayesian mixture model for mediation analysis incorporates heterogeneity in mediating behavior. The Bayesian nature of the model allows researchers to compute the probability that specic respondents follow a mediated versus a general relationship between stimulus and response, where the mediating relationship reflects conditional statistical independence. The empirical investigation provides support for the proposed mixture model and demonstrates additional insights that can enhance the understanding of the mediating mechanism.
Keywords: Mediation analysis, mixture model, structural heterogeneity, conditional independence
JEL Classification: C11, C19, M31
Suggested Citation: Suggested Citation