rdhte: Conditional Average Treatment Effects in RD Designs

2025-07-01

AI Summary

This paper addresses the problem of estimating and inferring covariate-dependent heterogeneous treatment effects under sharp regression discontinuity (RD) designs. To this end, it proposes a bias-corrected robust inference method that integrates data-driven optimal bandwidth selection, local linear estimation of the conditional average treatment effect (CATE), and construction of confidence intervals via linear combinations. The approach substantially improves statistical precision and robustness in heterogeneity analysis. The authors develop the R package `rdhte`, which seamlessly interfaces with `rdrobust`, offering automated bandwidth selection, grouped inference over multidimensional covariates, and flexible linear combination testing. Relative to existing methods, the proposed framework delivers more reliable heterogeneity detection in finite samples. It provides an open, reproducible, and extensible toolkit for empirical causal analysis—particularly for researchers seeking to uncover nuanced, context-specific treatment effects in RD settings.

Technology Category

• Reasoning under Uncertainty: Causality
• Machine Learning: Calibration & Uncertainty Quantification
• Intelligent Robots: State Estimation

Application Category

• Responsible Web: Data and user privacy-enhancing technologies for the Web
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Understanding causal heterogeneous treatment effects based on pretreatment covariates is a crucial aspect of empirical work. Building on Calonico, Cattaneo, Farrell, Palomba, and Titiunik (2025), this article discusses the software package rdhte for estimation and inference of heterogeneous treatment effects in sharp regression discontinuity (RD) designs. The package includes three main commands: rdhte conducts estimation and robust bias-corrected inference for heterogeneous RD treatment effects, for a given choice of the bandwidth parameter; rdbwhte implements automatic bandwidth selection methods; and rdhte lincom computes point estimates and robust bias-corrected confidence intervals for linear combinations, a post-estimation command specifically tailored to rdhte. We also provide an overview of heterogeneous effects for sharp RD designs, give basic details on the methodology, and illustrate using an empirical application. Finally, we discuss how the package rdhte complements, and in specific cases recovers, the canonical RD package rdrobust (Calonico, Cattaneo, Farrell, and Titiunik 2017).

Problem

• Estimates heterogeneous treatment effects in RD designs
• Provides bandwidth selection for RD analysis
• Computes linear combinations of treatment effects

Innovation

• Estimates heterogeneous RD treatment effects robustly
• Implements automatic bandwidth selection methods
• Computes linear combinations post-estimation