Identification and Debiased Learning of Causal Effects with General Instrumental Variables

2025-10-23

AI Summary

This paper addresses the challenge of identifying causal effects in the presence of unobserved confounding. We propose a nonparametric causal inference framework applicable to both multi-valued and continuous instrumental variables (IVs). Building upon the additive IV model, our approach identifies average potential outcomes and average treatment effects via a weighting function. Leveraging semiparametric efficiency theory, we derive the efficient influence function and construct a debiased machine learning estimator that achieves asymptotically normal and consistent estimation of causal effects. The method naturally extends to longitudinal data and dynamic treatment regimes. In simulations and an empirical application using the Job Training Partnership Act (JTPA) dataset, it substantially reduces finite-sample bias and improves statistical inference accuracy. To our knowledge, this is the first unified framework that establishes nonparametric identification and robust estimation for general IVs—covering both discrete and continuous instruments—under unobserved confounding.

Technology Category

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

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Instrumental variable methods are fundamental to causal inference when treatment assignment is confounded by unobserved variables. In this article, we develop a general nonparametric framework for identification and learning with multi-categorical or continuous instrumental variables. Specifically, we propose an additive instrumental variable framework to identify mean potential outcomes and the average treatment effect with a weighting function. Leveraging semiparametric theory, we derive efficient influence functions and construct consistent, asymptotically normal estimators via debiased machine learning. Extensions to longitudinal data, dynamic treatment regimes, and multiplicative instrumental variables are further developed. We demonstrate the proposed method by employing simulation studies and analyzing real data from the Job Training Partnership Act program.

Problem

• Identifying causal effects using general instrumental variables with confounding
• Developing nonparametric framework for continuous and categorical instruments
• Constructing debiased estimators via machine learning for treatment effects

Innovation

• Additive instrumental variable framework identifies causal effects
• Debiased machine learning constructs efficient estimators
• Method extends to longitudinal and dynamic treatment regimes