A0270
Title: Exact variance of average treatment effect estimator in cluster RCT
Authors: Yue Fang - The Chinese University of Hong Kong (China) [presenting]
Geert Ridder - University of Southern California (United States)
Abstract: In cluster randomized controlled trials with finite populations, the variance of the average treatment effect (ATE) estimator cannot be directly estimated due to the unobserved cluster-aggregated potential outcomes. A design of two-stage sampling is examined at both cluster and unit levels, and treatment assignment is examined at the cluster level. The exact variance of the Horvitz-Thompson estimator is derived for the ATE, and a sharp upper bound is established based on a past study. Additionally, a consistent estimator is proposed for this upper bound. A key implication is that the estimator yields a narrower confidence interval compared to the clustered standard error. These findings are demonstrated through simulations and empirical applications.
