Preference-based Conditional Treatment Effects and Policy Learning

TL;DR

This paper introduces a preference-based framework for estimating conditional treatment effects and learning policies, allowing for flexible modeling of heterogeneous effects using outcome rankings, and provides new identifiability conditions and estimation strategies.

Contribution

It proposes the CPTE framework that unifies various preference-driven outcomes and offers new identifiability conditions and estimation methods for treatment effect estimation.

Findings

Demonstrates performance gains in synthetic experiments

Provides practical impact through semi-synthetic experiments

Introduces efficient influence-function estimators for policy optimization

Abstract

We introduce a new preference-based framework for conditional treatment effect estimation and policy learning, built on the Conditional Preference-based Treatment Effect (CPTE). CPTE requires only that outcomes be ranked under a preference rule, unlocking flexible modeling of heterogeneous effects with multivariate, ordinal, or preference-driven outcomes. This unifies applications such as conditional probability of necessity and sufficiency, conditional Win Ratio, and Generalized Pairwise Comparisons. Despite the intrinsic non-identifiability of comparison-based estimands, CPTE provides interpretable targets and delivers new identifiability conditions for previous unidentifiable estimands. We present estimation strategies via matching, quantile, and distributional regression, and further design efficient influence-function estimators to correct plug-in bias and maximize policy value. Synthetic and semi-synthetic experiments demonstrate clear performance gains and practical impact.