Graph neural networks improve counterfactual outcome predictions without causal graphs
Amortized Counterfactual Prediction with Graph Neural Networks
Machine LearningArtificial Intelligence
Summary
Counterfactual questions ask what might have happened if things were different, like whether a patient would have recovered with another treatment. Usually, answering these requires knowing the cause-and-effect structure, which is hard to get right. The authors propose a method that directly predicts these alternate outcomes from data using graph neural networks, without needing to identify the causal structure first. Their method, called CFGNP, performs better than previous approaches and works well even on larger problems than it was trained on.
What this means in practice
- •For healthcare data teams: Predict patient outcomes under different treatment scenarios without needing explicit causal models.
- •For credit risk analysts: Improve loan default predictions under alternative economic interventions without relying on uncertain causal graphs.
Authors
Hendrik Stumpf, Francesco Quinzan
Abstract
Counterfactual inference asks how an observed outcome might have differed under different circumstances, such as whether a patient would have recovered had they received a different treatment. Such questions arise whenever decisions depend on the outcomes of alternatives that were not observed. Answering counterfactual queries generally requires knowledge of the underlying causal structure, which is often unavailable in practice. A common approach is therefore to first estimate the causal graph from observational data and then perform counterfactual inference. However, errors in the estimated causal graph can propagate to downstream counterfactual predictions. To address this challenge, we introduce the Counterfactual Graph Neural Process (CFGNP), an end-to-end approach that answers counterfactual queries directly from observational data without first estimating a causal graph. CFGNP uses Graph Neural Networks (GNNs) to exploit the dual-graph structure by restricting cross-world message passing to corresponding factual--counterfactual variables. Across synthetic and image-based benchmarks, CFGNP achieves stronger predictive performance than existing approaches, including on the German Loan benchmark, where it reduces MSE from $4.51 \times 10^{-2}$ to $0.19 \times 10^{-2}$. Furthermore, on synthetic causal graphs, a model trained only on 10-node graphs maintains strong predictive performance on graphs with up to 30 nodes, without retraining.