Consequences of Parameter Values Misspecification When Planning Group-randomized Mediation Studies with an Individual-Level Mediator

• In Event: Avoid a Cluster! Methodological Considerations for Cluster and Multilevel Analysis

Abstract

When planning group-randomized studies probing mediation, efficient sample allocation is governed by several parameters including treatment-mediator and mediator-outcome path coefficients and the intraclass correlation coefficients of the mediator and outcome. Typically, these parameters are approximated in the design stage using information from prior research and are likely to deviate from the true values eventually realized in the study. This study investigates the robustness of sensitivity to misspecified parameter values in group-randomized designs with individual-level mediators. Power in these designs was found to be robust to misspecified parameter values across a variety of conditions and tests. The relative sensitivity to detect the mediated effect remained above 90% in most conditions when the incorrect parameter value ranged between 50% and 150% of the true parameter. Designs employing the Monte Carlo interval test were the most robust to misspecified parameter values but performance of the joint and Sobel test followed closely.

Authors

• Kyle Cox, University of Cincinnati

• Benjamin Kelcey, University of Cincinnati

• ZUCHAO SHEN, University of Cincinnati