# Variable importance measures for heterogeneous treatment effects

arXiv: https://arxiv.org/abs/2204.06030 · HTML: https://mlsanity.com/paper/2204.06030/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 12 Apr 2022.

## Built on by (most cited first, 10 of 10)

- [Variable importance for causal forests: breaking down the heterogeneity of treatment effects](https://mlsanity.com/paper/2308.03369/) (2023)
- [A hierarchical decomposition for explaining ML performance discrepancies](https://mlsanity.com/paper/2402.14254/) (2024)
- [Measuring Variable Importance in Heterogeneous Treatment Effects with Confidence](https://mlsanity.com/paper/2408.13002/) (2024)
- [Transparency challenges in policy evaluation with causal machine learning -- improving usability and accountability](https://mlsanity.com/paper/2310.13240/) (2023)
- [Automatic debiasing of neural networks via moment-constrained learning](https://mlsanity.com/paper/2409.19777/) (2024)
- [Overview and practical recommendations on using Shapley Values for identifying predictive biomarkers via CATE modeling](https://mlsanity.com/paper/2505.01145/) (2025)
- ["Who experiences large model decay and why?" A Hierarchical Framework for Diagnosing Heterogeneous Performance Drift](https://mlsanity.com/paper/2506.00756/) (2025)
- [Counterfactual explainability and analysis of variance](https://mlsanity.com/paper/2411.01625/) (2024)
- [From Tokens to Policy: Causal and Interpretable Heterogeneous Treatment Effects Identification](https://mlsanity.com/paper/2606.17010/) (2026)
- [Bridging extrinsic and intrinsic variable importance](https://mlsanity.com/paper/2608.00247/) (2026)

## Abstract (arXiv)

Motivated by applications in precision medicine and treatment effect heterogeneity, recent research has focused on estimating conditional average treatment effects (CATEs) using machine learning (ML). CATE estimates may represent complicated functions that provide little insight into the key drivers of heterogeneity. Therefore, we introduce nonparametric treatment effect variable importance measures (TE-VIMs), based on the mean-squared error (MSE) in predicting the individual treatment effect. More precisely, TE-VIMs represent the increase in MSE when variables are removed from the CATE conditioning set. We derive efficient TE-VIM estimators which can be used with any CATE estimation strategy and are amenable to ML estimation. We propose several strategies to calculate these VIMs (e.g. leave-one out, or keep-one in), using popular meta-learners for the CATE. We study the finite sample performance through a simulation study and illustrate their application using clinical trial data.