Do causal inference methods really work?
Abstract — Over the past several decades, multiple statistical methods have been developed to infer the existence and magnitude of causal effects by analyzing observational data. These methods have been widely deployed in the social sciences and elsewhere to advance our understanding of phenomena that are difficult or impossible to study with randomized controlled trials. Theoretical analyses indicate that these methods can be effective given various assumptions, but the empirical effectiveness of these methods is surprisingly difficult to evaluate. In this talk, I will review the challenges to empirical evaluation, various approaches to such evaluation, and the results of recent implementations of these approaches. Finally, I will offer practical advice about navigating the large and rapidly growing body of methods for observational causal inference.
Bio — David Jensen is a Professor of Computer Science at the University of Massachusetts Amherst. He directs the Knowledge Discovery Laboratory and recently served as the Director of the Computational Social Science Institute. His current research focuses on causal modeling and reasoning and its applications for explainable AI, competency-aware machine learning, and robust machine learning in the face of novelty. In 2011 and 2022, he received college-level outstanding teaching awards at UMass. In 2017, one of his papers received the IEEE INFOCOM Test of Time Paper Award.
