Sharp Bounds for Treatment Effect Generalization under Outcome Distribution Shift

2026-02-10

AI Summary

This work proposes a sensitivity analysis framework based on a constrained likelihood ratio model (parameterized by Λ ≥ 1) to address the challenge of generalizing causal effects from randomized trials to target populations when unobserved effect modifiers induce distributional shifts in outcomes. By explicitly modeling bounded distributional discrepancies between trial and target populations, the method derives sharp nonparametric bounds for the target average treatment effect and reveals that the optimal likelihood ratio exhibits a threshold structure. Leveraging this insight, an O(n log n) closed-form greedy algorithm is developed that requires only sorting samples and reallocating probability mass. Theoretical and empirical results demonstrate that the resulting bounds achieve nominal coverage whenever the true distributional shift satisfies the Λ constraint, substantially outperforming worst-case bounds while retaining informative value across diverse non-transportable scenarios.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

Generalizing treatment effects from a randomized trial to a target population requires the assumption that potential outcome distributions are invariant across populations after conditioning on observed covariates. This assumption fails when unmeasured effect modifiers are distributed differently between trial participants and the target population. We develop a sensitivity analysis framework that bounds how much conclusions can change when this transportability assumption is violated. Our approach constrains the likelihood ratio between target and trial outcome densities by a scalar parameter $\Lambda \geq 1$, with $\Lambda = 1$ recovering standard transportability. For each $\Lambda$, we derive sharp bounds on the target average treatment effect -- the tightest interval guaranteed to contain the true effect under all data-generating processes compatible with the observed data and the sensitivity model. We show that the optimal likelihood ratios have a simple threshold structure, leading to a closed-form greedy algorithm that requires only sorting trial outcomes and redistributing probability mass. The resulting estimator runs in $O(n \log n)$ time and is consistent under standard regularity conditions. Simulations demonstrate that our bounds achieve nominal coverage when the true outcome shift falls within the specified $\Lambda$, provide substantially tighter intervals than worst-case bounds, and remain informative across a range of realistic violations of transportability.

Problem

• treatment effect generalization
• outcome distribution shift
• transportability
• sensitivity analysis
• unmeasured effect modifiers

Innovation

• sensitivity analysis
• treatment effect generalization
• distribution shift
• sharp bounds
• transportability