hte3

hte3 estimates conditional average treatment effects (CATE) and conditional risk ratios (CRR) in R. Fit DR-, R-, T-, and efficient plug-in (EP) learners, compare models by cross-validation, and predict effects for new observations. Use learners from sl3 or fit forests with the optional grf package.

Installation

if (!requireNamespace("remotes", quietly = TRUE)) {
  install.packages("remotes")
}

remotes::install_github("Larsvanderlaan/hte3", dependencies = TRUE)

To use grf_cate() or grf_crr(), also install grf:

install.packages("grf")

Quickstart

Create a task with your data and covariates, fit a model, and predict treatment effects. This example uses simulated data and simple mean learners.

library(hte3)
library(sl3)

data <- hte3_example_data(n = 150, seed = 1)

task <- hte_task(
  data = data,
  modifiers = c("W1", "W2", "W3"),
  confounders = c("W1", "W2", "W3"),
  treatment = "A",
  outcome = "Y",
  propensity_learner = Lrnr_mean$new(),
  outcome_learner = Lrnr_mean$new(),
  mean_learner = Lrnr_mean$new(),
  cross_fit = FALSE
)

fit <- fit_cate(
  task,
  method = "dr",
  base_learner = Lrnr_mean$new(),
  cross_validate = FALSE
)

pred <- predict(fit, data)
head(pred)
summary(fit)

Guides

Guide	What it covers
Quickstart	Fit and predict with hte_task() and fit_cate().
Treatment effects	Choose among DR-, R-, T-, and EP-learners for conditional mean differences.
Risk ratios	Estimate conditional risk ratios with fit_crr().
Causal forests	Fit models with grf_cate() and grf_crr(), with optional nuisance estimates supplied by you.
EP-learner	Choose a sieve basis size and compare EP-DR with EP-R.
Using sl3 directly	Construct tasks, train learners, and compare custom sets of models.

See the function reference for arguments and return values.

Supported outcomes and treatments

• fit_crr() and grf_crr() require a non-negative outcome.
• grf_cate() and grf_crr() currently support binary treatment only.
• For continuous treatments, use fit_cate(..., method = "r"), which assumes a partially linear effect model A * tau(X).