GCF: Generalized Causal Forest for Heterogeneous Treatment Effect Estimation Using Nonparametric Methods
Keywords: Heterogeneous Treatment Effect, Causal Inference, Double/Debiased Machine Learning, Continuous Treatment
Abstract: Heterogeneous treatment effect (HTE) estimation with continuous treatment is essential in multiple disciplines, such as the online marketplace and pharmaceutical industry. The existing machine learning (ML) methods, like forest-based modeling, either work only for discrete treatments or make partially linear or parametric assumptions that may suffer from model misspecification. To alleviate these problems, we extend causal forest (CF) with non-parametric dose-response functions (DRFs) that can be estimated locally using kernel-based Double/Debiased ML estimators. Moreover, we propose a distance-based splitting criterion in the functional space of Partial DRFs to capture the heterogeneity for continuous treatments. We call the proposed algorithm generalized causal forest (GCF) as it generalizes the use case of CF to a much broader setup. We show the effectiveness of GCF compared to SOTA on synthetic data and proprietary real-world data sets.
One-sentence Summary: We propose Generalized Causal Forest to efficiently estimate continuous treatment effects using a kernel-based DML estimator and a distance-based splitting criterion.
Supplementary Material: zip