Evaluating Heterogeneous Treatment Effects

Provides various statistical methods for evaluating heterogeneous treatment effects (HTE) in randomized experiments. The package includes tools to estimate uniform confidence bands for estimation of the group average treatment effect sorted by generic machine learning algorithms (GATES). It also provides the tools to identify a subgroup of individuals who are likely to benefit from a treatment the most "exceptional responders" or those who are harmed by it. Detailed reference in Imai and Li (2023) .

evalHTE

Overview

evalHTE provides statistical methods for evaluating heterogeneous treatment effects (HTE) in randomized experiments. The package implements the methodology developed in Imai and Li (2025) for:

• Estimating Grouped Average Treatment Effects (GATEs) with uniform confidence bands
• Identifying exceptional responders — individuals who benefit the most (or are harmed) by a treatment
• Conducting hypothesis tests for treatment effect heterogeneity and consistency
• Supporting both sample splitting and cross-validation approaches

Installation

From CRAN

install.packages("evalHTE")

Development Version from GitHub

# install.packages("devtools")
devtools::install_github("MichaelLLi/evalHTE")

Key Features

Quick Start

library(evalHTE)

# Prepare your data with outcome (y), treatment (z), and covariates
data <- data.frame(
  y = outcome_variable,
  z = treatment_indicator,
  x1 = covariate_1,
  x2 = covariate_2
)

# Step 1: Estimate heterogeneous treatment effects
fit <- estimate_hte(
  treatment = "z",
  form = y ~ z * (x1 + x2),
  data = data,
  algorithms = c("causal_forest"),
  n_folds = 5,
  ngates = 5
)

# Step 2: Evaluate the estimated HTE
est <- evaluate_hte(fit)

# Step 3: View summary results
summary(est)

# Step 4: Conduct hypothesis tests
test_results <- test_itr(est)
summary(test_results)

# Step 5: Visualize results
plot(est)                # Plot GATE estimates
plot_CI(est, alpha = 0.05)  # Plot uniform confidence intervals

Main Functions

Estimation

GATE Functions

Hypothesis Tests

Exceptional Responders

Visualization

Example with Simulated Data

library(evalHTE)
library(grf)

# Simulate data
set.seed(123)
n <- 1000
X <- matrix(rnorm(n * 5), n, 5)
colnames(X) <- paste0("x", 1:5)
tau <- X[, 1] + 0.5 * X[, 2]  # True treatment effect heterogeneity
W <- rbinom(n, 1, 0.5)        # Random treatment assignment
Y <- tau * W + rnorm(n)       # Observed outcome

# Create data frame
df <- data.frame(X, y = Y, z = W)

# Estimate HTE using causal forest
fit <- estimate_hte(
  treatment = "z",
  form = y ~ z * (x1 + x2 + x3 + x4 + x5),
  data = df,
  algorithms = c("causal_forest"),
  n_folds = 5,
  ngates = 5
)

# Evaluate results
est <- evaluate_hte(fit)
summary(est)

# Visualize GATE estimates
plot(est)

Supported Algorithms

The package supports various machine learning algorithms for estimating treatment effects:

• Causal Forest (causal_forest) via the grf package
• S-learner and T-learner meta-learners
• Any algorithm supported by caret
• User-defined custom models

Citation

If you use this package in your research, please cite:

@article{imai2025statistical,
  title={Statistical Inference for Heterogeneous Treatment Effects Discovered by Generic Machine Learning in Randomized Experiments},
  author={Imai, Kosuke and Li, Michael Lingzhi},
  journal={Journal of Business \& Economic Statistics},
  volume={43},
  number={1},
  pages={256--268},
  year={2025},
  doi={10.1080/07350015.2024.2358909}
}

Or in R:

citation("evalHTE")

Authors

• Michael Lingzhi Li (maintainer) - Harvard Business School - [email protected]
• Kosuke Imai - Harvard University - [email protected]

Contributor:

• Jialu Li - Harvard University

Related Packages

• evalITR - Evaluation of Individualized Treatment Rules
• grf - Generalized Random Forests
• caret - Classification and Regression Training

License

MIT License

References

Imai, K. and Li, M. L. (2025). "Statistical Inference for Heterogeneous Treatment Effects Discovered by Generic Machine Learning in Randomized Experiments." Journal of Business & Economic Statistics, Vol. 43, No. 1, pp. 256-268. https://doi.org/10.1080/07350015.2024.2358909