CausalGuard: Conformal Inference under Graph Uncertainty

2026-05-20

AI Summary

This work addresses the challenge in causal effect estimation where an unknown true causal graph often leads to misspecification of adjustment sets, resulting in prediction intervals that either under-cover or are overly conservative. To tackle this issue, the authors propose CausalGuard, a novel framework that integrates large language model–guided graph priors, conditional independence test–based pruning, and Bayesian Information Criterion reweighting to generate weighted pseudo-outcomes. By combining doubly robust estimation with conformal inference, CausalGuard achieves valid marginal coverage in finite samples. Empirical evaluations across five benchmark datasets demonstrate that the method consistently attains over 90% target coverage while substantially narrowing interval width, effectively avoids invalid adjustments under stress tests, and remains robust to misspecified priors.

Abstract

Estimating treatment effects from observational data requires choosing an adjustment set, but valid adjustment depends on an unknown causal graph. Graph misspecification can cause under-coverage, while graph-agnostic conformal wrappers may regain nominal coverage only through large padding. We introduce CausalGuard, a structure-weighted conformal framework that calibrates after aggregating graph-conditional doubly robust pseudo-outcomes. Candidate DAGs are proposed from an LLM-derived edge prior, pruned by conditional-independence tests, and reweighted by Bayesian Information Criterion. A composite nonconformity score then calibrates the posterior-weighted pseudo-outcome. CausalGuard provides distribution-free finite-sample marginal coverage for this aggregated pseudo-outcome; under causal identification, overlap, conditional-mean nuisance stability, and concentration on target-aligned valid adjustment strategies, its conditional mean converges to the true Conditional Average Treatment Effect. Across five benchmarks, CausalGuard attains mean coverage above the nominal 90% level for the directly evaluable target and reduces width when graph-agnostic conformal baselines require large padding. Stress tests show that CausalGuard suppresses invalid collider adjustment and remains stable under misspecified priors when the retained candidate set is data-supported.

Problem

• causal inference
• treatment effect estimation
• graph uncertainty
• conformal inference
• adjustment set selection

Innovation

• conformal inference
• causal graph uncertainty
• structure-weighted aggregation
• doubly robust pseudo-outcome
• conditional average treatment effect