Confounder Detection via Treatment Intent: New Causal Study Design

A new observational study design called confounder detection via treatment intent is introduced in arXiv paper 2605.26413. The method queries human experts making treatment decisions to compare matched pairs of units, aiming to elicit unobserved variables that explain treatment allocation. This addresses the problem of unobserved confounding in causal inference from observational data, which is increasingly available but limited by missing variables affecting both treatment and outcome. The approach is motivated by the high cost, time, and ethical constraints of randomized controlled trials (RCTs), the gold standard for causal inference. The paper proposes a principled matching strategy to identify pairs for expert comparison, potentially improving causal conclusions from large-scale observational datasets.