Weighting Methods for Assessing Mediation Effect Variation in Multi-Site Trials
*Xu Qin, University of Chicago
Keywords: Causal inference, mediation mechanism, multi-site trials, mediation effect variation
Adopting the potential outcomes causal framework, this study extends weighting approaches to causal mediation analysis in multi-site randomized trials. The goal is to inform policy impact by revealing not only the prevalent causal mechanism, but also how the mechanism may vary across sites. We extend the Ratio-of-Mediator-Probability Weighting (RMPW) approach (Hong, 2010; Hong, Deutsch, & Hill, 2011) and the Inverse Probability Weighting (IPW) approach (Huber, 2014) to multi-site trials in which a continuous mediator may be nonlinearly associated with the outcome. Primary interests are in the population average and between-site variation of the direct and indirect effects. The weighting strategies allow for treatment-by-mediator interaction and greatly simplify the outcome model specification. To identify an optimal weighting procedure, we consider alternative ways of implementation and assess through Monte Carlo simulations their relative performance. We also investigate through simulations the relative strengths of the innovative weighting strategies in comparison with the widely used multilevel path analysis/SEM.