Deep Learning for Causal Inference

Abstract

This review systematizes the emerging literature for causal inference using deep neural networks under the potential outcomes framework. It provides an intuitive introduction on how deep learning can be used to estimate/predict heterogeneous treatment effects and extend causal inference to settings where confounding is non-linear, time varying, or encoded in text, networks, and images. To maximize accessibility, we also introduce prerequisite concepts from causal inference and deep learning. The survey differs from other treatments of deep learning and causal inference in its sharp focus on observational causal estimation, its extended exposition of key algorithms, and its detailed tutorials for implementing, training, and selecting among deep estimators in Tensorflow 2 available at github.com/kochbj/Deep-Learning-for-Causal-Inference.

A Meta-Learner Framework to Estimate Individualized Treatment Effects for Survival Outcomes

“…Wager and Athey (2018) introduced a causal forest (CF) method by modifying the splitting rule in standard random forest algorithms to directly maximize the treatment effect within each node. More recently, Koch et al (2023) reviewed four approaches that estimate ITE through neural network architectures: TARNet learns ITE by balancing covariates between treatment and control groups through representation learning (Shalit et al, 2017); CFRNet, an extension of TARNet that incorporates an integral probability metric (Johansson et al, 2022); Dragonnet, another extension of TAR-Net with the inclusion of inverse probability weighting (Shi et al, 2019); and GANITE, which learns ITE through adversarial training (Yoon et al, 2018). Nevertheless, these approaches are formulated through a specific algorithm (e.g., trees or neural networks), lacking the flexibility to choose any desired parametric or nonparametric models for estimating ITE.…”