NeurIPS 2020

Robust Recursive Partitioning for Heterogeneous Treatment Effects with Uncertainty Quantification

Meta Review

The paper presents a useful and strongly grounded method for identifying heterogeneously responding subgroups in the context of causal treatment effects. The reviewers and I find this method to be novel, and believe it will be of interest to the causal inference community. While many method exists for identifying subgroups for treatment effects, the proposed method can work with black box predictors, has good theoretical guarantees, and focuses on interpretability (which is important in this task). I strongly encourage the authors to conduct the added experiments proposed by the reviewers and incorporate them into the revised version (possibly in the supplement).