The valiCATE package validates machine learning predictions of Conditional Average Treatment Effects (CATEs) using the Centered-Weighted Average Treatment Effect (CWATE) and its normalized counterpart (NCWATE). These estimands unify existing validation tools — BLP, AUTOC, QINI, and AUC-HVL — as special cases of a single framework. AIPW-based estimation with cross-fitted nuisance functions provides valid inference under mild regularity conditions.
Why use valiCATE?
| Feature | Benefit |
|---|---|
| Heterogeneity detection | CWATE one-sided test: does estimated CATE variation reflect genuine treatment heterogeneity or estimation noise? |
| CATE recovery | NCWATE two-sided test: do predicted CATEs recover the true treatment effect function? |
| Unified framework | Four built-in weight functions (AUTOC, AUC-HVL, BLP, QINI) nest existing tools as special cases of a single estimand. |
🚀 Installation
To install the latest stable release from CRAN, run:
# install.packages("valiCATE") # Not yet available!Alternatively, the current development version of the package can be installed using the devtools package:
devtools::install_github("riccardo-df/valiCATE")Contributing
We welcome contributions! If you encounter issues, have feature requests, or want to contribute to the package, please follow the guidelines below.
📌 Report an issue: If you encounter a bug or have a suggestion, please open an issue on GitHub: Submit an issue
📌 Contribute code: We encourage contributions via pull requests. Before submitting, please: 1. Fork the repository and create a new branch. 2. Ensure that your code follows the existing style and documentation conventions. 3. Run tests and check for package integrity. 4. Submit a pull request with a clear description of your changes.
📌 Feature requests: If you have ideas for new features or extensions, feel free to discuss them by opening an issue.