Familywise Type I Error Rates and Statistical Power in Causal Mediation Analysis with Multiple Mediators

• In Event: Addressing Challenges in Causality: Free Will, Mediation and Latent Factors, Heterogeneity, and Variable Selection

Abstract

Causal mediation analyses are increasingly used to test mediators in educational research. With multiple mediators, each mediator can correspond to several causally-defined indirect effects—some may be true nulls, while others true non-nulls. Researchers may be inclined to test all such effects, raising concerns about multiple testing issues and inflated familywise Type I error rates. However, little research exists on controlling the familywise Type I error rates while maintaining power. This study develops a simulation-based power analysis method for testing interventional indirect effects in two-mediator models and evaluates the performance of multiple testing adjustments (e.g., Bonferroni). We provide user-friendly R tools to aid power analysis and sample size planning for causal mediation analysis involving multiple mediators.

Authors

• Xiao Liu, University of Texas at Austin

• Jasmine Zeng, University of Texas at Austin