Correcting for Unreliability in Power Analyses of Cluster-Randomized Trials of Mediation

• In Event: Issues in Parameter Estimation in Complex Multilevel Models

Abstract

We derive closed-form formulas to correct multilevel mediation point estimates, standard errors, and the corresponding statistical power when the mediator and outcome are subject to measurement error. The results map out how to plan for and accommodate known unreliabilities in the mediator and outcome in the design phase (e.g., how do we prospectively identify sufficient sample sizes when we know the measures of the mediator and/or outcome will not produce fully reliable estimates). The results detail how to disattenuate the observed mediation effects and how to incorporate the error variance inflation associated with the unreliability when conducting power analyses and determining sufficient sample sizes. The formulas are implemented in the R library BLINDED.

Authors

• Benjamin Kelcey, University of Cincinnati

• Kyle T. Cox, University of Cincinnati

• Nianbo Dong, University of Missouri - Columbia