Augmented Balancing Weights as Linear Regression

Implements augmented balancing weights for causal inference and linear functional estimation based on David Bruns-Smith, Oliver Dukes, Avi Feller, and Elizabeth L. Ogburn (2026) . Establishes numerical equivalence between augmented balancing weight estimators and single linear models with weighted regression coefficients. Provides flexible routines for double ridge (l2 balancing), double lasso (l-infinity balancing), and generalized augmented linear outcome models. Features cross-validation procedures for tuning outcome penalty parameters, covariate balance, and Riesz loss. Supports robust influence-function-based standard errors, bootstrap confidence intervals, balance diagnostic tools, and counterfactual prediction for treatment effects such as average treatment effect (ATE) and average treatment effect on the treated (ATT), expanding upon the doubly robust estimation framework established by Robins, Rotnitzky, and Zhao (1994) and Chernozhukov, Chetverikov, Demirer, Duflo, Hansen, Newey, and Robins (2018) .

AugBalWeight: Augmented Balancing Weights as Linear Regression

AugBalWeight implements the methodology from the Journal of the Royal Statistical Society Series B paper:

Bruns-Smith, D., Dukes, O., Feller, A., & Ogburn, E. L. (2026). Augmented balancing weights as linear regression. Journal of the Royal Statistical Society Series B: Statistical Methodology, 88(3), 699–723. https://doi.org/10.1093/jrsssb/qkaf019.

Key Features

• Numerical Equivalence: Computes implied single linear regression coefficients $\hat{\beta}_{\text{aug}}$ combining outcome model and balancing weight parameters.
• Double Ridge ($\ell_2$ Balancing): Smooth decay regularization path matching ridge regression shrinkage.
• Double Lasso ($\ell_\infty$ Balancing): Soft-thresholding feature shift interpolation with double selection $I_{\text{aug}} = I_\lambda \cup I_\delta$.
• Hyperparameter Tuning: Flexible cross-validation algorithms for outcome model ($\lambda$), covariate balance ($\delta$), and Riesz loss.
• Causal Estimands: Built-in wrappers for ATE, ATT, and ATC with robust influence-function-based variance estimation.

Installation

# Install locally from source package:
install.packages("AugBalWeight_0.1.0.tar.gz", repos = NULL, type = "source")

Quick Example

library(AugBalWeight)

# Load canonical LaLonde dataset
data(lalonde_data)

# Estimate ATT (Average Treatment Effect on the Treated)
fit_att <- aug_bal_att(
  Y = lalonde_data$re78,
  Z = lalonde_data$treat,
  X = lalonde_data[, c("age", "educ", "black", "hisp", "married", "re74", "re75")],
  type = "l2"
)

summary(fit_att)
plot(fit_att)

Citation

@article{bruns2026augmented,
  title={Augmented balancing weights as linear regression},
  author={Bruns-Smith, David and Dukes, Oliver and Feller, Avi and Ogburn, Elizabeth L},
  journal={Journal of the Royal Statistical Society Series B: Statistical Methodology},
  volume={88},
  number={3},
  pages={699--723},
  year={2026},
  publisher={Oxford University Press},
  doi={10.1093/jrsssb/qkaf019}
}