Estimate Gaps Under an Intervention

Provides functions to estimate the disparities across categories (e.g. Black and white) that persists if a treatment variable (e.g. college) is equalized. Makes estimates by treatment modeling, outcome modeling, and doubly-robust augmented inverse probability weighting estimation, with standard errors calculated by a nonparametric bootstrap. Cross-fitting is supported. Survey weights are supported for point estimation but not for standard error estimation; those applying this package with complex survey samples should consult the data distributor to select an appropriate approach for standard error construction, which may involve calling the functions repeatedly for many sets of replicate weights provided by the data distributor. The methods in this package are described in the accompanying paper: .

gapclosing

Summary

Provides functions to estimate gap-closing estimands: the disparities across categories (e.g. Black and white) that persists if a treatment variable (e.g. college) is equalized. The purpose is to estimate the average outcomes that units would realize if exposed to a counterfactual treatment assignment rule.

The package will enable the user to:

Estimate treatment and outcome prediction functions with Generalized Linear Models, Generalized Additive Models, ridge regression, or random forest
Combine those in doubly-robust estimators of gap-closing estimands
Produce confidence intervals by the bootstrap
Visualize the result in plots

Installation instructions

Install from CRAN with one line: install.packages("gapclosing").

To install the latest development version,

First, install the devtools package: if(!require(devtools)) install.packages("devtools")
Then, install the gapclosing package with the command devtools::install_github("ilundberg/gapclosing").

Getting started

To get started, see the vignette. Also see the working paper for which this package is the software implementation.

Lundberg, Ian. Forthcoming. "The gap-closing estimand: A causal approach to study interventions that close disparities across social categories." Sociological Methods and Research. Draft available at https://doi.org/10.31235/osf.io/gx4y3.