evan-dong/demographic-prediction-argmax-bias

File outline

bisg_raking.ipynb and synthetic.ipynb are the two jupyter notebooks used to replicate our experiments.

The other files - decision_functions.py, metrics.py, viz.py, and utils.py - contain helper functions to discretize probabilities, calculate accuracy and fidelity, plotting functions for visualizations, and miscellaneous helper functions respectively.

Data

bisg_raking.ipynb uses data from two files, to be placed in the data/ folder.

Data Points

The file of data points, labels, and posteriors is data/df_agg_nc2020_dataverse.pkl, retrivable online from here. The full data citation is:

Philip Greengard and Andrew Gelman. Replication Data for: BISG: When inferring race or ethnicity, does
it matter that people often live near their relatives?, 2023. URL https://doi.org/10.7910/DVN/QIM4UF.

Map Shapefile

The shapefile used for the geographical map visualization of the counties in North Carolina is data/cb_2022_us_county_500k/cb_2022_us_county_500k.shp. These cartographic boundaries are sourced from the US Census Bureau. Specifically, use the .shp file from the 2022 counties shapefile, at the 500,000 scale zip link here.