Measuring Variable Importance in Heterogeneous Treatment Effects with Confidence

TL;DR

This paper introduces PermuCATE, a new algorithm for reliably measuring variable importance in heterogeneous treatment effect estimation, with theoretical guarantees and empirical validation in health data.

Contribution

PermuCATE offers a statistically rigorous, lower-variance alternative to existing methods for assessing variable importance in causal inference.

Findings

PermuCATE has lower variance than LOCO in finite samples.

It provides reliable variable importance measures with increased statistical power.

Empirical results demonstrate its effectiveness on simulated and real health datasets.

Abstract

Causal machine learning holds promise for estimating individual treatment effects from complex data. For successful real-world applications of machine learning methods, it is of paramount importance to obtain reliable insights into which variables drive heterogeneity in the response to treatment. We propose PermuCATE, an algorithm based on the Conditional Permutation Importance (CPI) method, for statistically rigorous global variable importance assessment in the estimation of the Conditional Average Treatment Effect (CATE). Theoretical analysis of the finite sample regime and empirical studies show that PermuCATE has lower variance than the Leave-One-Covariate-Out (LOCO) reference method and provides a reliable measure of variable importance. This property increases statistical power, which is crucial for causal inference in the limited-data regime common to biomedical applications. We empirically demonstrate the benefits of PermuCATE in simulated and real-world health datasets, including settings with up to hundreds of correlated variables.