PTDBoot is an R package containing methods that enable researchers to leverage large-scale machine learning predictions for downstream statistical inference, provided that a small number of labelled calibration samples are available. The package includes functions implementing the Predict-Then-Debias Bootstrap algorithms and their extensions, as described in the paper referenced below.

You can install the development version of PTDBoot directly from R:

# install.packages("devtools")
devtools::install_github("DanKluger/PTDBoot")

Here is a tutorial that gives examples of how to use the functions in the package. The R Markdown file and processed data that were used in the tutorial can be downloaded from here.

Please report any bugs or direct any questions to dkluger@mit.edu.

If you use PTDBoot in your work, please cite the following paper:

Dan M. Kluger, Kerri Lu, Tijana Zrnic, Sherrie Wang, and Stephen Bates (2025). Prediction-Powered Inference with Imputed Covariates and Nonuniform Sampling. 2501.18577 [stat.ME] https://arxiv.org/abs/2501.18577