Addressing Uncertainty in Power Analysis for Designing Cluster Randomized Trials

• In Event: Statistical Power Considerations in Multilevel Designs

Abstract

To design cluster randomized trials, power analysis requires knowledge of the population values of parameters, such as the effect size and intraclass correlation. Whereas these values are often unknown at the study design phase, the conventional practice is to provide researchers’ best educated guess based on prior information and belief. In the present study, we propose a hybrid classical-Bayesian (HCB) approach, which utilizes a distribution of parameter values in power analysis to represent researchers’ beliefs and uncertainty, as opposed to using a single point value that ignores uncertainty. In a simulation study, we demonstrated that the HCB approach adequately adjusts for uncertainty. In addition, we provide an R package and a Web application that implements the proposed approach.

Authors

• Winnie Wing-Yee Tse, University of Southern California
  Presenting Author

• Hok Chio Mark Lai, University of Southern California
  Non-Presenting Author