Regression Discontinuity Tree

Welcome to the GitHub page of Regression Discontinuity Tree. This page provides the codes for this method proposed by Reguly (2025): Discovering Heterogeneous Treatment Effects in Regression Discontinuity Designs.

Use of the codes

This repository contains codes and a description of how to estimate heterogeneous treatment effects in RDD. Codes are built under MatLab, version 2024a.

1. Add the downloaded/forked/cloned algorithm folder to your path in MatLab.
2. Open tryout.m, which gives a toy example of how to use the package.

• The main function is runTree_uni() function.
• However, you need to carefully specify your sample object as well as optCART object. The comments should guide you on how to set them.

3. tostring() function will draw you the resulting tree.

Note: This is the second version, containing nonparametric estimator.

Replication of Reguly (2025)

1. algorithm folder contains necessary functions to run regression discontinuity tree. One needs to add this folder to the path when using any other script.
2. simulations folder contain codes to replicate simulation results reported in Section 4. sharp folder contains runRDD_sharp.m file which runs the monte carlo simulations. Caution -- it takes considerable amount of time and uses multiple cores. (1day+ with Apple M1 Max chip with 64GB memory and 9 cores) Note if one does not run with 50,000 observations or reduces the MC iteration number it reduces the time significantly. After running the simulations, one can evaluate the results with evaluate_results/createTables.m.
3. empirical_application folder contains codes to replicate Section 5. It also contains a further readme file to navigate through the replication.

All feedbacks are welcomed! Please write to: agoston.reguly@uni-corvinus.hu

Python implementation

A python implementation of the parametric version of the paper is done by Firat Yaman. Check here.