Interval Estimation of Individual-Level Causal Effects Under Unobserved\n Confounding

2018/10/05 by Nathan Kallus, Kallus, Nathan, Xiaojie Mao +3 · 11 citations
Mathematics · Decision Sciences · Economics, Econometrics and Finance · #Advanced Causal Inference Techniques #Decision-Making and Behavioral Economics #Health Systems, Economic Evaluations, Quality of Life

paper · pdf · doi:10.48550/arxiv.1810.02894

Abstract

We study the problem of learning conditional average treatment effects (CATE)\nfrom observational data with unobserved confounders. The CATE function maps\nbaseline covariates to individual causal effect predictions and is key for\npersonalized assessments. Recent work has focused on how to learn CATE under\nunconfoundedness, i.e., when there are no unobserved confounders. Since CATE\nmay not be identified when unconfoundedness is violated, we develop a\nfunctional interval estimator that predicts bounds on the individual causal\neffects under realistic violations of unconfoundedness. Our estimator takes the\nform of a weighted kernel estimator with weights that vary adversarially. We\nprove that our estimator is sharp in that it converges exactly to the tightest\nbounds possible on CATE when there may be unobserved confounders. Further, we\nstudy personalized decision rules derived from our estimator and prove that\nthey achieve optimal minimax regret asymptotically. We assess our approach in a\nsimulation study as well as demonstrate its application in the case of hormone\nreplacement therapy by comparing conclusions from a real observational study\nand clinical trial.\n

Cited by

• Assessment of the conditional exchangeability assumption in causal machine learning models: a simulation study
• Intact-VAE: Estimating Treatment Effects under Unobserved Confounding
• Calibrating confounding strength in sensitivity models for weighting estimators: a comparative review and a new method
• Improving the Generation and Evaluation of Synthetic Data for Downstream Medical Causal Inference
• Assessing the robustness of heterogeneous treatment effects in survival analysis under informative censoring
• Coupling Generative Modeling and an Autoencoder with the Causal Bridge
• Neuropathic Pain Diagnosis Simulator for Causal Discovery Algorithm Evaluation
• Causal-EPIG: A Prediction-Oriented Active Learning Framework for CATE Estimation
• Causal-BALD: Deep Bayesian Active Learning of Outcomes to Infer Treatment-Effects from Observational Data
• Hölder Bounds for Sensitivity Analysis in Causal Reasoning
• Assessing Sensitivity to Unconfoundedness: Estimation and Inference

Related

• Bounds on the conditional and average treatment effect with unobserved confounding factors
• Conditional Average Treatment Effect Estimation Under Hidden Confounders
• A Partial Initialization Strategy to Mitigate the Overfitting Problem in CATE Estimation with Hidden Confounding
• Estimation of Conditional Average Treatment Effects with High-Dimensional Data
• Hölder Bounds for Sensitivity Analysis in Causal Reasoning