Abstract:
In the social and biomedical sciences, researchers often seek unbiased estimators of causal effects. Yet unbiased estimators are often inadmissible, in the sense that a biased alternative will always have lower expected error. This raises a natural practical question: how much is gained by using biased estimators instead? In this talk, I present an algorithm for computing minimax estimators of average-treatment effects from experimental data and compare their performance with that of the standard difference-in-means estimator.
Conor Mayo-Wilson is an associate professor in the philosophy department at the University of Washington and an affiliate faculty member in the Center for Statistics in the Social Sciences. His philosophical interests are rather broad, spanning much of epistemology (esp. formal and social), philosophy of science, decision theory, game theory, and philosophy of statistics.
