Jointly Optimal Design for Experiments Investigating Main and Moderation Effects

• In Event: Effect Estimation for Causality and Prediction: Multilevel and Single-case Designs, Machine Learning, and Distributional Assumptions

Abstract

Studies probing moderation effects in conjunction with main effects provide a richer and more comprehensive portrait of the effectiveness and equity of interventions. The present study develops an optimal design framework for moderation in multisite randomized trials (MRTs) and a jointly optimal design framework for both main and moderation effects. The framework can identify the (jointly) optimal sample allocation that maximizes statistical power under a fixed budget. Illustrative examples show that the methods work as intended. The methods have been implemented in the R package XXX (masked for peer review). The results of this study can help researchers design more cost-efficient experiments with sufficient statistical power for both effects.

Authors

• Zuchao Shen, University of Georgia

• Benjamin Kelcey, University of Cincinnati

• Zhenqiu Laura Lu, University of Georgia

• Jingyang Li, University of Georgia

• Jing Li, University of Georgia