Research

Long-Horizon Policy Learning with Surrogate Outcomes and Data Combination

Weiwei Wu and Jelena Bradic

(This paper will be available on arXiv soon.)

Abstract: In this work, we tackle the challenge of designing personalized treatment strategies for long-term outcomes by combining experimental and observational data. Specifically, we focus on situations where long-term outcomes are unavailable in experimental settings. At the same time, observational data, which could provide insights, is affected by hidden variables that influence both short-term and long-term outcomes. To address these issues, we develop methods that link short-term outcomes to long-term results and account for differences between experimental and observational data. These methods are supported by strong theoretical guarantees. We demonstrate the effectiveness of these methods through comprehensive simulations and a real-world application using data from a job training program in California, highlighting their practical value.