Black Box Causal Inference: Effect Estimation via Meta Prediction
Lucius E. J. Bynum, Aahlad Manas Puli, Diego Herrero-Quevedo, Nhi Nguyen, Carlos Fernandez-Granda, Kyunghyun Cho, Rajesh Ranganath
TL;DR
This work introduces Black Box Causal Inference (BBCI), a meta-learning framework that treats causal effect estimation as a dataset-level prediction problem. By sampling a family of structural causal models and generating dataset–estimand pairs, BBCI trains a neural predictor to map any observed dataset to a target estimand such as the average treatment effect or conditional average treatment effect, bypassing problem-specific estimator design. The authors provide a theoretical error decomposition and demonstrate that BBCI matches or surpasses dedicated estimators across confounding, instrumental variables, and proximal causal inference settings, including mixed or real-unknown identification. The results on semi-synthetic and real data illustrate BBCI’s versatility, robustness to weak instruments and limited data, and potential to scale to new causal-identification regimes with minimal manual derivation. The work points to future directions in higher-dimensional covariates and uncertainty quantification, positioning BBCI as a practical, generalizable tool for causal effect estimation.
Abstract
Causal inference and the estimation of causal effects plays a central role in decision-making across many areas, including healthcare and economics. Estimating causal effects typically requires an estimator that is tailored to each problem of interest. But developing estimators can take significant effort for even a single causal inference setting. For example, algorithms for regression-based estimators, propensity score methods, and doubly robust methods were designed across several decades to handle causal estimation with observed confounders. Similarly, several estimators have been developed to exploit instrumental variables (IVs), including two-stage least-squares (TSLS), control functions, and the method-of-moments. In this work, we instead frame causal inference as a dataset-level prediction problem, offloading algorithm design to the learning process. The approach we introduce, called black box causal inference (BBCI), builds estimators in a black-box manner by learning to predict causal effects from sampled dataset-effect pairs. We demonstrate accurate estimation of average treatment effects (ATEs) and conditional average treatment effects (CATEs) with BBCI across several causal inference problems with known identification, including problems with less developed estimators.