Testing Identifiability and Transportability with Observational and Experimental Data

2025-05-19

AI Summary

This study addresses the joint identifiability and transportability of causal effects across populations under unknown causal graphs. Motivated by the clinical challenge that experimental data from a source population often cannot be directly generalized to a target population, we propose the first Bayesian framework that does not require prior knowledge of the causal graph—unifying identifiability and transportability assessment within a single probabilistic model. Our method performs joint likelihood estimation using randomized trial data from the source domain and observational data from the target domain, grounded in mild assumptions from structural causal models. It enables probabilistic determination and joint inference of Z-conditional causal effects. Simulation results demonstrate that our approach accurately identifies transportable effects and significantly improves both unbiasedness and statistical efficiency in estimating target-population causal effects compared to single-source estimators—thereby overcoming the traditional reliance on known causal graph structures.

Technology Category

• Reasoning under Uncertainty: Causality
• Machine Learning: Causal Learning
• Intelligent Robots: State Estimation

Application Category

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

Abstract

Transporting causal information learned from experiments in one population to another is a critical challenge in clinical research and decision-making. Causal transportability uses causal graphs to model differences between the source and target populations and identifies conditions under which causal effects learned from experiments can be reused in a different population. Similarly, causal identifiability identifies conditions under which causal effects can be estimated from observational data. However, these approaches rely on knowing the causal graph, which is often unavailable in real-world settings. In this work, we propose a Bayesian method for assessing whether Z-specific (conditional) causal effects are both identifiable and transportable, without knowing the causal graph. Our method combines experimental data from the source population with observational data from the target population to compute the probability that a causal effect is both identifiable from observational data and transportable. When this holds, we leverage both observational data from the target domain and experimental data from the source domain to obtain an unbiased, efficient estimator of the causal effect in the target population. Using simulations, we demonstrate that our method correctly identifies transportable causal effects and improves causal effect estimation.

Problem

• Assessing causal effect transportability without known causal graphs
• Combining experimental and observational data for causal estimation
• Improving unbiased causal effect estimation in target populations

Innovation

• Bayesian method assesses identifiability and transportability
• Combines experimental and observational data sources
• Estimates causal effects without known causal graph