Package and simulations related to "Optimally Balanced Gaussian Process Propensity Scores for Estimating Treatment Effects" by Brian Vegetabile, Dan L. Gillen, and Hal Stern
- This repository serves as a snapshot of the "gpbalancer" package as of 3/7/2019 to provide consistent and stable reproducibility of results within the paper.
- Clone or download this repository
- Using RStudio, open the file
paperOptBalGPPS.Rprojfile - Compile the package (on a Mac this can be done with "command+shift+b"
- This may require the
devtoolspackage
- This may require the
The simulations that were run for the paper are located within the ~Simulations/ folder. Each simulation was run with n_sim=1000.
01-atesim.R- Provides simulations for comparing methods for estimating the ATE02-atesim-tablesforpaper.R- Compiles the results from 01-atesim.R for building tables03-attsim-revision.R- Provides simulations for comparing methods for estimating the ATT04-attsim-tablesforpaper.R- Compiles the results from 03-attsim-mixtures.R for building tables05-timingresults.R- Compares timing results that were provided in the discussion section
Within the simulation folder are two other folders which contain the simulation results that were used to construct the tables for the paper.
- ~Simulations/ate-simresults/
2019-02-08-nonparametric_odd-atesim-results.rds2018-02-08-nonparametric_even-atesim-results.rds
- ~Simulations/att-simresults/
2019-02-11-attsim-multivariate-results-revision.rds
To recreate the application section results the file dw99_replication.R is provided that utilizes out GP methodology. A secondary analysis re-evaluates the DW99 data using the models defined in that paper, but utilizing IPW estimation. These can be recreated using reanalyzing_ds99.R
- Data for the replication is within the data folder.
- Note: Propensity score estimates that were used in this paper were too large to store on Github.