Work

Deconfounding Scores and Representation Learning for Causal Effect Estimation with Weak Overlap

Year: 2026

Type: preprint

Abstract: Overlap, also known as positivity, is a key condition for causal treatment effect estimation. Many popular estimators suffer from high variance and become brittle when features differ strongly across treatment groups. This is especially challenging in high dimensions: the curse of dimensionality can make overlap implausible. To address this, we propose a class of feature representations called deconfounding scores, which preserve both identification and the target of estimation; the classical propensity and prognostic scores are two special cases. We characterize the problem of finding a representation with better overlap as minimizing an overlap divergence under a deconfounding score constraint. We then derive closed-form expressions for a class of deconfounding scores under a broad family of generalized linear models with Gaussian features and show that prognostic scores are overlap-optimal within this class. We conduct extensive experiments to assess this behavior empirically.

Source: arXiv (Cornell University)

DOI: 10.48550/arxiv.2604.00811

Authors: Clivio, Oscar, D'Amour, Alexander, Franks, Alexander, Bruns-Smith, David, Holmes, Chris, Feller, Avi

Institutions: Ellison Institute of Technology, Stanford University, University of California, Santa Barbara

Language: English

Topic: Advanced Causal Inference Techniques

Subfield: Statistics and Probability

Field: Mathematics

Domain: Physical Sciences

Open Access status: green

Funders: National Science Foundation, U.S. Department of Education, Li Ka Shing Foundation, Novo Nordisk, National Institutes of Health, Institute of Education Sciences, Medical Research Council, Engineering and Physical Sciences Research Council, National Institute of General Medical Sciences

Awards: Collaborative Research: DMS/NIGMS 2: Methods for Systematic Analysis of Post-transcriptional Regulation in Single Cells, G1579515842, G1974824942, Multi-group covariance estimation for metabolomic analysis with applications to neurodegenerative disease, Statistical Innovations for Balancing Weight Methods in Education Research, New Approaches to Bayesian Data Science: Tackling Challenges from the Health Sciences, G6219198695, G6631254222, HDR DSC: Collaborative Research: Central Coast Data Science Partnership: Training a New Generation of Data Scientists, NRT-HDR: Computational Research for Equity in the Legal System" (CRELS), EPSRC Centre for Doctoral Training in Modern Statistics and Statistical Machine Learning, Improving Methods for Policy Impact Evaluation with Group Panel Data in Education Research

OpenAlex ID: W7148452588

Cites:

Cited by:

Related to: 10