09/11/2023
By Hsien-Yuan Hsu
Date: September 13
Time: Noon to 1 p.m.
Location: Coburn Hall 275
Please RSVP to reserve your spot.
Title: We Cannot Randomize Participants in Observational Studies. What Else Can We Do to Make Causal Inference?
Speaker's Short Bio
Wei Pan, Ph.D., is an Associate Professor and Director of Health Statistics and Data Science at the Duke University School of Nursing and an Associate Professor of Population Health Sciences at the Duke University School of Medicine. His research interests are causal inference, advanced statistical modeling, data analytics, meta-analysis, and psychometrics; and their applications in the social, behavioral, and health sciences. He is an Invited Expert in the Reference Group on Health Statistics of the World Health Organization.
Abstract
We cannot randomize participants in observational studies that are likely to be plagued by selection bias. What else can we do to make causal inference? Among popular statistical techniques for reducing selection bias in observational studies, propensity score methods are widely regarded as the best practice to make valid causal inference from observational studies. However, there are still methodological challenges in propensity score methods. In this talk, we will first overview the basic concepts and caveats of propensity score methods and, then, briefly discuss two methodological developments of propensity score methods by the speaker. One is interval matching which utilizes the bootstrap technique to deal with the estimation error issue in propensity score estimation; and the other, from the robustness perspective, addresses the sensitivity of unobserved confounding in observational studies.