Combining experimental and historical data for policy evaluation

Li, Ting, Shi, Chengchun ORCID: 0000-0001-7773-2099, Wen, Qianglin, Sui, Yang, Qin, Yongli, Lai, Chunbo and Zhu, Hongtu (2024) Combining experimental and historical data for policy evaluation. Proceedings of Machine Learning Research, 235. pp. 28630-28656. ISSN 2640-3498

Text (li24bh) - Published Version Download (4MB)

• Scopus publication

Abstract

This paper studies policy evaluation with multiple data sources, especially in scenarios that involve one experimental dataset with two arms, complemented by a historical dataset generated under a single control arm. We propose novel data integration methods that linearly integrate base policy value estimators constructed based on the experimental and historical data, with weights optimized to minimize the mean square error (MSE) of the resulting combined estimator. We further apply the pessimistic principle to obtain more robust estimators, and extend these developments to sequential decision making. Theoretically, we establish non-asymptotic error bounds for the MSEs of our proposed estimators, and derive their oracle, efficiency and robustness properties across a broad spectrum of reward shift scenarios. Numerical experiments and real-data-based analyses from a ridesharing company demonstrate the superior performance of the proposed estimators.

Item Type:	Article
Additional Information:	© 2024 The Author(s)
Divisions:	Statistics
Subjects:	H Social Sciences > HA Statistics Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Date Deposited:	01 Oct 2024 15:54
Last Modified:	15 Oct 2024 17:06
URI:	http://eprints.lse.ac.uk/id/eprint/125588

Actions (login required)

View Item