Causal Gaussian Processes for Robust Treatment Effect Evaluation with Unobserved Confounding

2026-06-19

AI Summary

This work proposes the first confounding-robust causal inference framework for continuous treatments and outcomes that requires no prior knowledge of environmental variables. Addressing the fundamental challenge that causal effects are generally non-identifiable under unobserved confounding, the method introduces a general discretization strategy over exogenous variable domains to construct a finite latent state representation capable of approximating both observational and interventional distributions to arbitrary precision. Building upon this representation, a novel Causal Gaussian Process (CGP) model is developed. Theoretical analysis establishes the universal approximation capability of the proposed approach, which substantially enhances the robustness and accuracy of continuous causal effect estimation in the presence of hidden confounders.

Abstract

The presence of confounding bias poses a key challenge in policy evaluation, as the target causal effects of actions are not identifiable (i.e., underdetermined) from observational data. On the other hand, existing confounding-robust evaluation strategies require detailed prior knowledge about the environment or apply only to discrete treatments and outcomes. This paper investigates causal effect evaluation over the continuous domain from confounded observations, while requiring only basic temporal ordering between the treatment and the outcome. We introduce a universal discretization of the exogenous domains that approximates the observational and interventional distributions of any causal model with arbitrary accuracy using a finite number of latent states. Building on this newfound universal approximation property, we develop a novel family of Causal Gaussian process (CGP) models that effectively approximate the observational and interventional distributions of any causal model with confounded observations.

Problem

• causal inference
• unobserved confounding
• treatment effect
• continuous treatments
• policy evaluation

Innovation

• Causal Gaussian Processes
• Unobserved Confounding
• Treatment Effect Estimation
• Universal Discretization
• Interventional Distribution