Donald B. Rubin, Harvard University

Techniques for Drawing Causal Inferences from Imperfect Studies

Thursday, May 15 at 7:30pm in Physics-Astronomy A110

This lecture will focus on straightforward techniques for estimating causal effects in observational studies, primarily methods of subclassification and matching using propensity scores. Medical and economic examples will be used to illustrate the technology and to show that answers from observational studies can track those from similar randomized experiments when appropriate covariates are controlled by propensity score methods. Specifically, the medical example concerns experimental and observational data on breast cancer treatments, and the economic example concerns experimental and observational data on a job-training program; if time permits, a new study on the effects of income infusion (winning the lottery) will also be discussed.