Quasi-oracle estimation of heterogeneous treatment effects
By Xinkun Nie, Stefan Wager. 2017
871 citations recorded by the source corpus.
Flexible estimation of heterogeneous treatment effects lies at the heart of many statistical applications, such as personalized medicine and optimal resource allocation. In this article we develop a general class of two-step algorithms for heterogeneous treatment effect estimation in observational studies. First, we estimate marginal effects and treatment propensities to form an objective function that isolates the causal component of the signal. Then, we optimize this data-adaptive objective function. The proposed approach has several advantages over existing methods. From a practical perspective, our method is flexible and easy to use: in both steps, any loss-minimization method can be employed, such as penalized regression, deep neural 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 when the pilot estimates for marginal effects and treatment propensities are not particularly accurate, we achieve the same error bounds as an oracle with prior knowledge of these two nuisance components. We implement variants of our approach based on penalized regression, kernel ridge regression, and boosting in a variety of simulation set-ups, and observe promising performance relative to existing baselines.
Subfields: Statistics and Probability, Causal Inference & Discovery
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Statistics and Probability
- Stefan Wager
- Susan Athey
- Victor Chernozhukov
- Edward H. Kennedy
- Whitney K. Newey
- Guido W. Imbens
- Xinkun Nie
- Rahul Singh
In Causal Inference & Discovery
- Stefan Wager
- Susan Athey
- Edward H. Kennedy
- Victor Chernozhukov
- Whitney K. Newey
- E.J. Candes
- Julie Tibshirani
- Xinkun Nie
Citations in the corpus
- Local Linear Forests (2018, methodology): “From the perspective, we can see local linear causal forests as an adapted R-learner method; note that Nie and Wager (2017) and Kennedy (2020) give quasi-oracle and double-robustness properties of R-learner estimates.”
- Double/De-Biased Machine Learning of Global and Local Parameters Using Regularized Riesz Representers (2018, background): “Prominent examples of the debiased CATE literature include Wang et al. (2010), van der Laan and Luedtke (2014), Luedtke and Van Der Laan (2016), Nie and Wager (2017), Lee et al. (2017), and most recently Kennedy (2020).”
- Learning When-to-Treat Policies (2019, background): “The problem of learning optimal decision rules is also closely related to learning heterogeneous treatment effects (Athey and Imbens, 2016; Athey, Tibshirani, Wager, et al., 2019; Chen, 2007; Künzel, Sekhon, Bickel, and Yu, 2019; Nie and Wager, 2017; Wager and Athey, 2017).”
- A Survey on Causal Inference (2020, methodology): “Different from the regular loss function adopted in X-learner, R-learner, proposed in [99] designs the loss function for CATE estimator based on the Robinson transformation [120].”
- Causal machine learning for predicting treatment outcomes (2024, neutral): “23,27,49,53,54,61,98,99).”
- Estimating Treatment Effects with Causal Forests: An Application (2019, positive): “3 Causal Forests for Observational Studies One promising avenue to heterogeneous treatment effect estimation starts from an early result of Robinson (1988) on inference in the partially linear model (Nie and Wager, 2017; Zhao, Small, and Ertefaie, 2017).”
- Conformal inference of counterfactuals and individual treatment effects (2020, background): “Under the strong ignorability assumption, a standard argument shows that Thus, the estimation problem reduces to that of estimating certain conditional expectations (e.g. Foster & Syrgkanis, 2019; Kennedy, 2020; Nie & Wager, 2017).”
- Semiparametric doubly robust targeted double machine learning: a review (2022, methodology): “We refer to Foster and Syrgkanis [2019], Kennedy [2020], Kennedy et al. [2022], Nie and Wager [2021], Semenova and Chernozhukov [2017] for some examples of such adaptation.”
- Towards optimal doubly robust estimation of heterogeneous causal effects (2020, methodology): “…on nonpara-metrics and incorporating machine learning [Athey and Imbens, 2016, Foster and Syrgkanis, 2019, Foster et al., 2011, Hahn et al., 2020, Imai and Ratkovic, 2013, Künzel et al., 2019, Nie and Wager, 2021, Semenova and Chernozhukov, 2017, Shalit et al., 2017, Wager and Athey, 2018].”
- Covariate powered cross‐weighted multiple testing (2017, methodology): “Conceptually, cross- weighting is related to cross- fitting (Schick, 1986), a method that has been successful in the fields of causal inference (Chernozhukov et al., 2017; Nie & Wager, 2020) and empirical Bayes (Ignatiadis & Wager, 2019) for estimation with high- dimensional nuisance parameters.”
- Policy Learning With Observational Data (2017, methodology): “Causal forests use the adaptive neighborhood function implied by a forest to fit a partially linear model using the method of Robinson (1988); see Nie and Wager (2020) for formal results motivating the use of local partially linear modeling for heterogeneous treatment effect estimation, and Section…”
- Bayesian Regression Tree Models for Causal Inference: Regularization, Confounding, and Heterogeneous Effects (with Discussion) (2020, positive): “As another example, Nie and Wager (2017) showed that using ps-BART (motivated by an early draft of this paper) as a component of a stacked estimator of heterogeneous treatment effects fitted using the R-learner yielded improved performance over the individual heterogeneous treatment effect estimator”
- Estimation of Conditional Average Treatment Effects With High-Dimensional Data (2019, background): “There are several papers in the broader statistics literature that have considered estimation problems related to ours (Lechner, 2019; Luedtke and van der Laan, 2016a,b; Nie and Wager, 2017; Robins, 2004; van der Laan, 2013).”