On Efficient Estimation of Distributional Treatment Effects under Covariate-Adaptive Randomization

This repository contains the implementation code for reproducing results from the paper "On Efficient Estimation of Distributional Treatment Effects under Covariate-Adaptive Randomization," (ICML 2025) by Undral Byambadalai, Tomu Hirata, Tatsushi Oka, and Shota Yasui.

Simulation with Python

Environment

• Python >= 3.12
• uv >= 0.7.0

Installation

# Clone the repository
git clone https://github.com/CyberAgentAILab/dte_car.git
cd dte_car

# Setup Python environment
# See https://docs.astral.sh/uv/ for other installation
curl -LsSf https://astral.sh/uv/install.sh | sh
uv python install 3.12
uv python pin 3.12
uv venv
source .venv/bin/activate

# For running DTE estimation
uv sync

Run Simulation

You can run the simulation using simulation.py. The available arguments for this script are described below.

Arguments:

• n: Specifies the number of samples for each simulation iteration. This is a required argument and must be a positive integer.
• iterations: The number of simulation iterations.
• discrete: Indicates whether the Data Generating Process (DGP) is discrete outcome or continuous outcome. By default, the DGP is continuous. Use this flag to switch to discrete mode.
• discrete_covariates: Indicates whether the Data Generating Process (DGP) is discrete covariates or not. By default, the DGP uses continuous covariates. Use this flag to switch to discrete.

Example:

• For 1000 samples with a continuous DGP: --n 1000
• For 5000 samples with a discrete DGP: --n 5000 --discrete true

python simulation.py --n 5000 --discrete true

Empirical Application with R

Dependencies

Necessary R packages and their versions: R version 4.3.1, knitr_1.49 RColorBrewer_1.1-3 xgboost_1.7.5.1 ggplot2_3.5.1 dplyr_1.1.2 readr_2.1.4 haven_2.5.3

Data preparation

To replicate the reanalysis of the field experiment run by Attanasio et al. (2015), download the original dataset at openICPSR.
Save the whole folder in the same directory and run data_preprocess.R to prepare the dataset and save it to a CSV file called microcredit.csv.

Data analysis

• Run empirical_application_microcredit.R to calculate distributional treatment effects and save the results as figures (Figure 3 in the paper)

• functions.R contains functions needed to calculate the effects.

Contributors

• Undral Byambadalai ([email protected])
• Tomu Hirata ([email protected])