Skip to contents

Changelog

Source: NEWS.md

ensembleHTE 0.1.1

New features

  • Y, X, and D in ensemble_hte() (and Y/X in ensemble_pred()) now accept column name(s) in addition to raw data. Pass Y as a single string, X as a character vector, or D as a single string together with data to have columns extracted automatically (e.g., Y = "hhinc_yrly_end", X = microcredit_covariates, D = "treat", data = microcredit). Similarly, prop_score in ensemble_hte() now accepts a column name string (e.g., prop_score = "prop_score").
  • Prettier startup message with clickable links (in RStudio) using the cli package. ensemble_news() and citation("ensembleHTE") are now clickable in the loading message.
  • Added cli to Imports.

Improvements

  • Significance stars now shown in summary.ensemble_hte_fit() (BLP beta1/beta2), summary.ensemble_pred_fit() (BLP intercept/slope), and restricted comparison print methods (print.gates_restricted_results(), print.gavs_restricted_results()).
  • Signif. codes legend in restricted comparison print methods updated to the standard R format (0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1).
  • print.clan_results() now computes column widths dynamically from the data, ensuring headers, values, standard errors, and separator lines are always properly aligned regardless of value magnitude.
  • ITE Distribution section removed from summary.ensemble_hte_fit() output. Individual ITE point estimates are not directly interpretable; the summary now focuses on BLP and GATES inference.
  • Prediction Summary section removed from summary.ensemble_pred_fit() output for the same reason. The returned object no longer includes prediction_summary.
  • Internal formula is now rebuilt to reflect the actual covariates used, improving consistency when Y ~ . with exclusions is passed.
  • as.data.table() conversion is now skipped when the input is already a data.table, avoiding unnecessary copies.

Bug fixes

  • Fixed clan() data subsetting when column names differ from the default internal names (Y, D, etc.), which could occur with the new column-name interface.
  • Improved error messages when Y, X, or D are accidentally NULL (e.g., from passing a non-existent column like df$wrong_name). The error now identifies which argument is NULL and suggests checking the column name.

Documentation

  • README substantially rewritten with a streamlined Quick Start using real microcredit data and the column-name interface, including an ensemble_pred() example.
  • New walkthrough vignette (vignettes/articles/microcredit-walkthrough.Rmd) demonstrating a full analysis with the microcredit dataset.
  • microcredit dataset documentation updated with Karlan & Zinman (2011) citation and DOI.

ensembleHTE 0.1.0

  • Initial CRAN release.
  • Core estimation functions:
    • ensemble_hte(): Ensemble heterogeneous treatment effect estimation with metalearners (R-learner, T-learner, S-learner, X-learner).
    • ensemble_pred(): Ensemble prediction model for general outcome prediction.
    • combine_ensembles(): Combine multiple ensemble fits from distributed computing.
  • Analysis functions:
    • gates(): Group Average Treatment Effects (GATES) estimation.
    • blp(): Best Linear Predictor of CATE.
    • clan(): Classification Analysis by covariates.
    • blp_pred(): Best Linear Predictor for prediction tasks.
    • gavs(): Group Averages for prediction tasks.
  • Comparison functions:
  • S3 methods for print, summary, and plot.
  • Comprehensive documentation and vignettes.
  • Test suite.