Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

balnet: Pathwise Estimation of Covariate Balancing Propensity Scores

Erik Sverdrup, Trevor Hastie

Abstract

We present balnet, an R package for scalable pathwise estimation of covariate balancing propensity scores via logistic covariate balancing loss functions. Regularization paths are computed with Yang and Hastie (2024)'s generic elastic net solver, supporting convex losses with non-smooth penalties, as well as group penalties and feature-specific penalty factors. For lasso penalization, balnet computes a regularized balance path from the largest observed covariate imbalance to a user-specified fraction of this maximum. We illustrate the method with an application to spatial pixel-level balancing for constructing synthetic control weights for the average treatment effect on the treated, using satellite data on wildfires.

balnet: Pathwise Estimation of Covariate Balancing Propensity Scores

Abstract

We present balnet, an R package for scalable pathwise estimation of covariate balancing propensity scores via logistic covariate balancing loss functions. Regularization paths are computed with Yang and Hastie (2024)'s generic elastic net solver, supporting convex losses with non-smooth penalties, as well as group penalties and feature-specific penalty factors. For lasso penalization, balnet computes a regularized balance path from the largest observed covariate imbalance to a user-specified fraction of this maximum. We illustrate the method with an application to spatial pixel-level balancing for constructing synthetic control weights for the average treatment effect on the treated, using satellite data on wildfires.
Paper Structure (5 sections, 1 theorem, 21 equations, 3 figures, 1 table)

This paper contains 5 sections, 1 theorem, 21 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Regularization paths for covariate balancing propensity scores
  3. Application: Pixel-level balancing
  4. Timings
  5. Discussion

Key Result

Proposition 1

The loss $l_1(\eta)$ is strictly convex, bounded below, and has a unique minimizer if and only if the following set is empty:

Figures (3)

  • Figure 1: Regularization path for covariate balancing using data from wu2023low for treatment year 2008 and conifer forests. The plot shows percentage bias reduction (PBR) and effective sample size (ESS) as functions of $\lambda$. The right-hand axis displays the coefficient of variation of the inverse probability weighting (IPW) weights.
  • Figure 2: Standardized mean differences before and after weighting, evaluated at $\lambda_{\min}$ from the path fit, with the unweighted data corresponding to $\lambda^{\max}$, using data from wu2023low for treatment year 2008 and conifer forests.
  • Figure 3: Estimated ATT weights using penalized balance loss at $\lambda_{\min}$ and penalized maximum likelihood loss (default $\lambda_{\min}^{\textrm{ML}}$).

Theorems & Definitions (3)

  • Remark 1
  • Remark 2
  • Proposition 1: Proposition S1, tan2020regularized