Taylor & Francis Group
Browse
DOCUMENT
uasa_a_2008402_sm1698.pdf (355.65 kB)
TEXT
Evaluation.R (0.92 kB)
TEXT
Functions.R (15.14 kB)
.RDATA
quantTreatResults_fourCov_n_1000.RData (3.18 MB)
1/0
4 files

Fairness-Oriented Learning for Optimal Individualized Treatment Rules

dataset
posted on 2021-11-22, 19:40 authored by Ethan X. Fang, Zhaoran Wang, Lan Wang

There has recently been a surge on the methodological development for optimal individualized treatment rule (ITR) estimation. The standard methods in the literature are designed to maximize the potential average performance (assuming larger outcomes are desirable). A notable drawback of the standard approach, due to heterogeneity in treatment response, is that the estimated optimal ITR may be suboptimal or even detrimental to certain disadvantaged subpopulations. Motivated by the importance of incorporating an appropriate fairness constraint in optimal decision making (e.g., assign treatment with protection to those with shorter survival time, or assign a job training program with protection to those with lower wages), we propose a new framework that aims to estimate an optimal ITR to maximize the average value with the guarantee that its tail performance exceeds a prespecified threshold. The optimal fairness-oriented ITR corresponds to a solution of a nonconvex optimization problem. To handle the computational challenge, we develop a new efficient first-order algorithm. We establish theoretical guarantees for the proposed estimator. Furthermore, we extend the proposed method to dynamic optimal ITRs. The advantages of the proposed approach over existing methods are demonstrated via extensive numerical studies and real data analysis.

Funding

Ethan X. Fang was partially supported by NSF Grants DMS-1820702, DMS-1953196, and DMS-2015539. Zhaoran Wang was partially supported by NSF Grants ECCS-2048075, CCF-2008827, DMS-2015568, and CCF-1934931, Simons Institute (Theory of Reinforcement Learning), and gifts from Amazon, Two Sigma and J.P. Morgan. Lan Wang was partially supported by NSF Grants DMS-1952373 and OAC-1940160.

History