Abstract: Many questions in biomedical, environmental, and psychological research concern the effects of multiple factors and their interactions. Randomized factorial experiments are well suited to these questions, but random assignment is often infeasible or impractical. In observational studies, confounding and sparse or missing treatment combinations make factorial effects difficult to estimate. This talk presents a weighting method that uses observational data to emulate a randomized factorial experiment. The key insight is that covariate balance, while necessary, is insufficient for unbiased estimation of factorial effects; balance among the treatment factors is also essential. The method supports simultaneous estimation and inference for the effects of multiple factors and their interactions, and extends to settings with missing treatment combinations.
About the Speaker: Ruoqi Yu is an Assistant Professor of Statistics at the University of Illinois Urbana-Champaign. Her research focuses on causal inference, including design-based methods for large and complex observational studies and the integration of randomized and observational evidence. Her work is motivated by applications in public policy, health sciences, and the social sciences.