Seminar by Dr. Zhen ZHONG, Business School, The University of Hong Kong

Fisher’s randomization test (FRT) is a unique tool for statistical inference in randomized experiments, valid under the sharp null hypothesis of no individual treatment effects. However, the sharp null hypothesis has been criticized as ”uninteresting and academic” (Neyman et al. 1935), limiting the practical usage of FRT.

In this talk, I will present an extension of FRT to test conditional independence between observed outcomes and treatments given covariates, with no restriction on the variable type of treatments. Under a generalized unconfoundedness assumption, we provide causal identification for this hypothesis. Notably, our framework circumvents the need for the positive overlap assumption—a crucial barrier in observational studies with continuous treatments—and does not require the no-interference component of SUTVA.

Methodologically, our framework separates the roles of assignment and outcome models: a known or consistently estimated assignment mechanism guarantees Type I error control, while a family of outcome models, without being correctly specified, defines Bayesian power and guides the choice of the optimal test statistic. The synthesis of two classes of models through FRT yields a calibrated Bayesian procedure with desired frequentist properties. Recognizing that the generalized unconfoundedness assumption is untestable in observational studies, we develop a novel sensitivity analysis to assess the robustness of causal conclusions to unobserved confounding. The talk concludes with a re-analysis of a panel dataset, demonstrating how our methods can be integrated into a pipeline for observational causal inference.

Zhen Zhong is a Post-Doctoral Fellow at the HKU Business School, The University of Hong Kong. He received his Ph.D. in Statistics from Tsinghua University, advised by Prof. Donald B. Rubin. His research interests include causal inference, experimental design, and calibrated Bayesian methodology.