Adding a New Dimension to RDD! 🚀
R3D brings your regression discontinuity designs into the third dimension by handling distributional outcomes instead of just scalar responses. When your treatment effect isn’t just a single number but an entire distribution shift, R3D has you covered!
Installation 📦
# Install the development version from GitHub:
# install.packages("devtools")
devtools::install_github("Davidvandijcke/R3D")
# Or once on CRAN:
# install.packages("R3D")Features ✨
- 📏 Estimates treatment effects across the entire distribution
- 🎯 Handles both sharp and fuzzy RD designs
- 🧮 Choose between simple quantile-by-quantile or global Fréchet approaches
- 📊 Beautiful visualizations of distributional treatment effects
- 🔍 Uniform inference with multiplier bootstrap
- 🧪 Tests of effect nullity and homogeneity
Quick Example 📋
library(R3D)
# Simulate data
set.seed(123)
n <- 100
X <- runif(n, -1, 1) # Running variable
Y_list <- lapply(seq_len(n), function(i) {
# Distribution is Normal with mean depending on X
rnorm(sample(30:50, 1), mean = 2 + 2 * (X[i] >= 0))
})
# Sharp RDD with distributional outcome
fit <- r3d(X, Y_list, cutoff = 0,
method = "frechet", p = 2,
boot = TRUE, boot_reps = 200)
# Visualize the results
plot(fit)
# Examine detailed results
summary(fit)How It Works 🔧
R3D uses advanced local polynomial and Fréchet regression techniques to estimate how an entire distribution changes at a discontinuity threshold when treatment is at a higher level of aggregation than the outcome variable. Instead of just estimating E[Y|X] at the cutoff, we estimate the entire average conditional quantile function Q_Y(τ|X) at the threshold. Note the word average! Unlike traditional quantile RDD, we are now sampling random distributions, which requires a different approach to estimation and inference. For more details on this fascinating subject, check out the supporting paper by Van Dijcke (that’s me!) (2025 (that’s now!)).
Citation 📄
If you use R3D in your research, please cite:
Van Dijcke, D. (2025). Regression Discontinuity Design with Distributional Outcomes.
Working paper.Learn More 📚
Check out the full documentation for tutorials, examples, and detailed function references.
