Using Optimal Sample Allocation to Improve the Chance of Detecting Treatment Effects in Cluster-Randomized Trials

• In Event: 34.094 - Fairmont Royal York Roundtable Session Six
  In Roundtable Session: 34.094-5 - Partially Clustered Data and Sample Size Investigations With Various Types of Multilevel Models

Abstract

Prior work has identified sample allocations that maximize statistical power with which cluster-randomized trials can detect treatment effects. Conventional frameworks produce a type of constrained optimal design because they typically assume balanced treatment assignment and that the costs of sampling units in different treatment conditions are equal. In this study, we relax cost equality assumptions and develop a framework that identifies sample allocations that optimize power in the presence of varied costs across treatment conditions and levels of hierarchy. The results show that the proposed framework can identify more efficient sample allocations (higher power) that are robust to the misspecification of design parameter values and cost structures. The solutions are implemented in R package XXX.

Authors

• Zuchao Shen, University of Georgia

• Benjamin Kelcey, University of Cincinnati