vankesteren/fair_measurement

Reproducible code repository supporting the manuscript "Fair inference on error-prone outcomes"

Fair inference on error-prone outcomes

L. Boeschoten, E.-J. van Kesteren, A. Bagheri, & D.L. Oberski

This is a reproducible code repository supporting the manuscript. By running the R scripts in this repository, all the figures and tables can be reproduced. This repository follows a CC-BY 4.0 license.

The dataset data/data_new.csv was obtained directly without making changes from the dissecting bias repository on GitLab and is subject to their license terms.

Setup

Ensure you have R and RStudio installed and run the following lines to install the dependencies

• install.packages("tidyverse", "glmnet", "patchwork", "lavaan", "remotes")
• remotes::install_github("vankesteren/firatheme")

Running the code

The structure of the folder follows loosely the structure of the Experiments section in the manuscript. Clone this repository, open the fair_measurement.Rproj file and run the following files in order:

Section	Main R file	Main output
Data preparation and feature selection	01_feature_selection.R	output/01_feature_selection.png
Fair inference on cost as a proxy of health	02_parity_correction.R	output/02_parity_correction.pdf
Fair inference on latent health	03_measurement_model.R	output/03_measurement_model.pdf
Investigating unfairness in proxies	04_differential_item_functioning.R	Table 1 (in console)

These main R files depend on the remaining R files and output files in this repository. The main figures will appear in the output/ folder.