This page is optimized for a taller screen. Please rotate your device or increase the size of your browser window.

Use of Monte Carlo Simulation to Inform Design Decisions for Pairwise Cluster Randomization

William Rhodes, Lauren E. W. Olsho, Alison Sexton Ward, William D. Spector


March 2, 2015

Randomized controlled trials (RCTs) are universally recognized as the preferred way to infer treatment effects because RCTs typically minimize validity challenges while maximizing estimation efficiency. In practice, simple RCTs—where individual study subjects are assigned to treatment or control status at random—are infeasible when an intervention must be implemented at the cluster level, for example, for all patients within a hospital, or for all residents within a nursing home. An alternative in these situations is a cluster randomization (CR) design, where each cluster is randomly assigned to treatment or control status. While standard principles of randomization still apply, the CR design is less efficient than the simple RCT. In some situations, efficiency of a CR design can be improved by matching clusters, and, within each pair, assigning one cluster to treatment and the other to control status, otherwise known as a pairwise cluster randomization (PCR) design. The choice between CR and PCR is not clear, since the optimal approach depends on information about the intraclass and interclass correlation coefficients, which in practice are rarely known to the evaluator. Theory alone does not provide a concrete answer; however, a Monte Carlo simulation can provide useful evidence at the design stage. This paper demonstrates the utility of the Monte Carlo approach in the context of a planned evaluation of an intervention to reduce falls among nursing home residents and provides recommendations for researchers on key design questions, including the choice between CR and PCR, and selection of parametric or nonparametric estimators.