dte-ra

This repository contains R code to replicate the experimental results in "Regression Adjustment for Estimating Distributional Treatment Effects in Randomized Controlled Trials" by Tatsushi Oka, Shota Yasui, Yuta Hayakawa, and Undral Byambadalai.

Getting Started

git clone https://github.com/CyberAgentAILab/dte-ra

Installations

Install all necessary packages in R.

• R version 4.3.1
• Packages: readr_2.1.4 haven_2.5.3 fastglm_0.0.3 bigmemory_4.6.1 doParallel_1.0.17 iterators_1.0.14
foreach_1.5.2 foreign_0.8-84 patchwork_1.2.0 cowplot_1.1.3 RColorBrewer_1.1-3 ggpubr_0.6.0
gridExtra_2.3 ggplot2_3.5.1 tidyr_1.3.0 dplyr_1.1.2

Data Preparation

1. Ferraro & Price (2013): Download original data from https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN1/22633&version=1.1 and save 090113_TotWatDat_cor_merge_Price.dta file in data folder.

2. Oregon Health Insurance Experiment: Download original data from https://www.nber.org/research/data/oregon-health-insurance-experiment-data and save the folder OHIE_Public_Use_Files in data folder.

3. Run data/data_pre_process.R to save the csv files named data_ferraroprice.csv and data_oregon_12m.csv.

Analysis

experiment folder contains all R files used for analysis.

(1) experiment/functions.R: code containing all necessary functions

(2)experiment/run_simulation.R: code to run the Monte Carlo simulations and saves results as .rds files

(3)experiment/compute_stats.R: code to calculate evaluation metrics (e.g. bias, RMSE, coverage probability, average length of confidence intervals) from the saved simulation results (.rds files) and saves them as .csv files

(4)experiment/plot_figures.R: code to load the .csv files and plot figures for the simulation study

(5)experiment/experiment_water_consumption.R: code to replicate the analysis of experimental data from Ferraro & Price (2013)

(6)experiment/experiment_oregon.R: code to replicate the analysis of experimental data from Oregon Health Insurance Experiment

Steps

1. Create a folder called result to store all the results and figures.

2. To replicate the results from the Monte Carlo simulation, run the files in the following order: (1) experiment/run_simulation.R, (2) experiment/compute_stats.R, (3) experiment/plot_figures.R.

   The output will be figures appeared in Figures 1, 2 in the main text and Figures 5-14 in the Appendix.

3. Run experiment/experiment_water_consumption.R to replicate the results from the water consumption experiment.

   The output will be figures appeared in Figure 3 in the paper.

4. Run experiment/experiment_oregon.R to replicate the results from the Oregon health insurance experiment.

   The output will be figures appeared in Figure 4 in the paper.

Citation

@misc{oka2025regressionadjustmentestimatingdistributional,
      title={Regression Adjustment for Estimating Distributional Treatment Effects in Randomized Controlled Trials}, 
      author={Tatsushi Oka and Shota Yasui and Yuta Hayakawa and Undral Byambadalai},
      year={2025},
      eprint={2407.14074},
      archivePrefix={arXiv},
      primaryClass={econ.EM},
      url={https://arxiv.org/abs/2407.14074}, 
}