Feature Selection for Causal Effect Estimation

Tuesday, February 8 at 4:00 PM CDT

Causal Inference Seminar with Richard Hahn.

Richard Hahn is Associate Professor of Statistics at Arizona State University. Prior to that he was Associate Professor of Statistics at the University of Chicago Booth School of Business. He holds a PhD in Statistics from Duke University and an undergraduate degree in Philosophy and Economics from Columbia University. His research lies at the intersection of Bayesian machine learning, causal inference, and decision theory. Specifically, he develops novel machine learning algorithms for analyzing observational and experimental data with a focus on applications in health and behavioral science.

This paper defines the notion of a minimal control function, on the basis of which a novel regression penalty is devised that is unbiased for average treatment effects. The development of the new approach combines insights from three distinct methodological traditions for studying causal effect estimation: potential outcomes, causal diagrams, and structural models with additive errors. It is demonstrated that naive feature selection and/or regularization approaches to treatment effect estimation can exhibit severe bias for average and conditional average treatment effects.