The study frameworks are based on the following papers:
- Wager S and Athey S. Estimation and inference of heterogeneous treatment effects using random forests. Journal of the American Statistical Association 2018. 113(523): 1228–1242.
- Athey S, Tibshirani J and Wager S. Generalized random forests. The Annals of Statistics 2019. 47(2): 1148–1178.
- Nie X and Wager S. Quasi-oracle estimation of heterogeneous treatment effects. Biometrika 2021. 108: 299–319.
The package also includes the code for reproducing all findings of
Dandl S, Hothorn T, Seibold H, Sverdrup E, Wager S, Zeileis A (2022). What Make Forest-based Heterogeneous Treatment Estimators Work? Technical report, arXiv 2206.10323. URL https://arxiv.org/abs/2206.10323.
- Empirical study: inst/empeval/paper1
- Blood loss study: demo/bloodloss.R
Dandl S, Bender A, Hothorn T (2022). Heterogeneous Treatment Effect Estimation for Observational Data using Model-based Forests. Preprint will be available soon.
- Empirical study: inst/empeval/paper2
- ALS study: for survival time and handwriting ability score as outcomes
You can install the github version, using remotes:
remotes::install_github("susanne-207/htesim")The following code immitates Setup A of Nie and Wager (2020)
# Initialize manual for data generating process
# Functions for treatment effect, prognostic effect and treatment propensity
dg <- dgp(t = tF_div_x1_x2, m = mF_sin_x1_x5, p = pF_sin_x3, model = "normal", xmodel = "unif")
# Simulate data
sdg <- simulate(dg, nsim = 1000L, dim = 12)
head(sdg) Our vignette gives a broad overview on the functionality:
- for sampling from predefined data generating processes (DGP)
- for sampling from user-defined DGP
- for replicating the empirical study of Dandl et al. (2022a, 2022b)
- for comparing the results of Dandl et al. (2022a, 2022b) with an own treatment effect estimation method on the same simulated data.
- not yet determined
GPL-2 | GPL-3