Machine Learning and Causal Inference for Policy Evaluation

Machine Learning and Causal Inference for Policy Evaluation

 

Susan Athey

Stanford

 

ABSTRACT

This talk will review a series of recent papers that adapt machine learning methods to the problem of causal inference.  Applications include estimating heterogeneous treatment effects in randomized experiments (A/B tests) as well as observational studies; estimating and evaluating optimal treatment assignment policies (e.g. personalized medicine, targeted online content); estimating average effects of policies in high dimensions; and using surrogate outcomes to proxy for long-term outcomes in randomized experiments (e.g. the use of intermediate medical outcomes, test scores, or short-term user engagement metrics as proxies for long term health, student attainment, or lifetime user value).