Quasi-Oracle Estimation of Heterogeneous Treatment Effects

Abstract

Flexible estimation of heterogeneous treatment effects lies at the heart of many statistical challenges, such as personalized medicine and optimal resource allocation. In this paper, we develop a general class of two-step algorithms for heterogeneous treatment effect estimation in observational studies. We first estimate marginal effects and treatment propensities in order to form an objective function that isolates the causal component of the signal. Then, we optimize this data-adaptive objective function. Our approach has several advantages over existing methods. From a practical perspective, our method is flexible and easy to use: In both steps, we can use any loss-minimization method, e.g., penalized regression, deep neutral networks, or boosting; moreover, these methods can be fine-tuned by cross validation. Meanwhile, in the case of penalized kernel regression, we show that our method has a quasi-oracle property: Even if the pilot estimates for marginal effects and treatment propensities are not particularly accurate, we achieve the same error bounds as an oracle who has a priori knowledge of these two nuisance components. We implement variants of our approach based on both penalized regression and boosting in a variety of simulation setups, and find promising performance relative to existing baselines.

Evaluating Treatment Prioritization Rules via Rank-Weighted Average Treatment Effects

“…We start from the simulation "Setup A" in Nie and Wager (2017), in which data is generated as follows:…”

“…Causal Forest The Causal Forest prioritization rule similarly uses ensembles of trees built with recursive partitioning to assign subject priorities but, unlike random forests, the causal forest chooses partitions to minimize the CATE R-loss criterion (Nie and Wager, 2017). We use the Causal Forest implementation provided by grf.…”

Identification of Subgroups With Similar Benefits in Off-Policy Policy Evaluation

“…Imai and Ratkovic [2014] uses the LASSO to estimates the effect of treatments, Shalit et al [2017] uses neural networks and offers generalization bound for individual treatment effect (ITE). Nie and Wager [2017] proposed two step estimation procedure using double machine learning and orthogonal moments [Chernozhukov et al, 2018] that can be applied on observational data to infer HTEs, and recently, Lee et al [2020] suggests a robust partitioning algorithm by inducing homogeneity in groups. However, these methods were developed for nonsequential settings and naïvely applying them to sequential setting will result in predictions with low accuracy and high uncertainty.…”

“…In the non-sequential setting, a growing literature has focused on personalization and estimation of heterogeneous treatment effect (HTE), the individual-level differences in potential outcomes under the proposed evaluation policy versus the behaviour policy [Athey et al, 2019, Nie andWager, 2017].…”

Heterogeneous Effects of Software Patches in a Multiplayer Online Battle Arena Game

“…Many methods build upon interpretable tree-based methods, such as decision lists [15], classification and regression trees (CART) [3,26,30], and random forests [4,27]. Others follow more in line with the supervised machine learning paradigm, such as using supervised base learners, called meta-learners, which decompose HTE estimation into multiple regression or classification problems [13,17]. Representation learning using deep neural networks have also been proposed for estimating HTEs [11,24].…”