Instrumental Variables in Causal Inference and Machine Learning: A Survey
Anpeng Wu, Kun Kuang, Ruoxuan Xiong, Fei Wu
Abstract
Causal inference is the process of drawing conclusions about causal relationships between variables using a combination of assumptions, study designs, and estimation strategies. In machine learning, causal inference is crucial for uncovering the mechanisms behind complex systems and making informed decisions. This article provides a comprehensive overview of using Instrumental Variables (IVs) in causal inference and machine learning, with a focus on addressing unobserved confounding that affects both treatment and outcome variables. We review identification conditions under standard assumptions in the IV literature. In this article, we explore three key research areas of IV methods: Two-Stage Least Squares (2SLS) regression, control function (CFN) approaches, and recent advances in IV learning methods. These methods cover both classical causal inference approaches and recent advancements in machine learning research. Additionally, we provide a summary of available datasets and algorithms for implementing these methods. Furthermore, we introduce a variety of applications of IV methods in real-world scenarios. Lastly, we identify open problems and suggest future research directions to further advance the field. A toolkit of reviewed IV methods with machine learning (MLIV) is available at https://github.com/causal-machine-learning-lab/mliv .