Optimal Design of Group-Randomized Trials Probing Meditation Effects With Group-Level Mediators

• In Event: Advances in Multilevel Modeling for Causal Inferences

Abstract

Literature guiding the efficient design of mediation studies has been limited to the malleable parameters considered (e.g., the between treatment equal sampling costs assumption). This study is to develop a more flexible optimal design framework for 2-2-1 mediation studies in group-randomized trials. Specifically, this study will (a) derive the optimal design parameters to maximize statistical power under a fixed budget while allowing the costs of sampling vary across both treatment conditions and levels of hierarchy; (b) develop the framework in different inferential tests (e.g., Sobel test, joint significance, and Monte Carlo confidence intervals); (c) investigate the robustness of the framework to misspecification of design parameters; and (d) implement the optimal design framework in an existing R package.

Authors

• Zuchao Shen, University of Georgia

• Walter L. Leite, University of Florida