How We Investigated Michigan Insurance Territories

This repository contains data and code to reproduce the findings featured in our story "Michigan’s 'Fair and Reasonable' Reforms Allowed Car Insurers to Charge More in Black Neighborhoods."

Our methodology is described in "How We Investigated Car Insurance Loopholes in Michigan."

We extracted the relevant territory tables from documents submitted to the Office of Insurance Rates and Forms of the Michigan Department of Insurance and Financial Services.

The code in this repository was used to produce and analyze the data for our investigation. The process is described in detail below for replication and inspection.

Notebooks

This is a summary of the notebooks in the insurer directories.

00(a)-extract-auto-table.ipynb

This notebook extracts territory tables from PDF of insurer filing to produce tabulated data files for analysis.

00b-geography-experiment.ipynb

This notebook exists for State Farm and Allstate, the two insurers using custom-made gridded territory maps. It is used to run experiments to determine the optimal geography match for our analysis.

01-analysis.ipynb

This notebook replicates and fact checks all the data figures and findings used in publication.

Data

The data produced for the analysis is stored in each insurer's outputs/ directory as INSURER_auto_clean.csv. For example, 02_allstate/outputs/allstate_auto_clean.csv for Allstate.

Those files contain the following columns:

column	description
geo_id	The census geographic identifier that matches onto the territory. Read more about how this value is selected in the methodology document.
geo_name	The census geographic area name that matches onto the territory. Read more about how this value is selected in the methodology document.
total_pop	The total population in the census geography.
white_pct	The percentage of census geography's population that is non-Hispanic/non-Latino White.
black_pct	The percentage of census geography's population that is non-Hispanic/non-Latino Black.
white_tot	The percentage of census geography's population count that is non-Hispanic/non-Latino White.
black_tot	The census geography's population count that is non-Hispanic/non-Latino Black.
generic_location_based_premium	The location-fixed or location-adjusted base rate. Read more about this value in the methodology document.
location_effect	The location-fixed or location-adjusted base rate indexed to the insurer's state median. Read more about this value in the methodology document.
geometry	The insurance territory's point or polygon geometry.
median_income	The median household income for the census geography.
density	The geography's total population divided by total area sourced from TIGER shape files from the U.S. Census Bureau.
is_along_8_mile	Boolean representing whether a gridded territory point is within a 1 mile buffer of 8 mile road. This value is only available for State Farm and Allstate cleaned data files.
is_in_detroit	This a boolean representing whether a territory is within Detroit.
is_north_8_mile	This is a boolean representing whether a gridded territory point is within a 1 mile buffer north of 8 mile road. This value is only available for State Farm and Allstate cleaned data files.
is_south_8_mile	This is a boolean representing whether a gridded territory point is within a 1 mile buffer south of 8 mile road. This value is only available for State Farm and Allstate cleaned data files.
bg_black_pct	This is the percentage of the census block group geography's population that is non-Hispanic/non-Latino Black. This value is only available for State Farm cleaned data file.
bg_black_tot	The census geography's population count that is non-Hispanic/non-Latino Black. This value is only available for State Farm cleaned data file.
bg_geo_id	The census block group geographic identifier that matches onto the territory. This value is only available for State Farm cleaned data file.
bg_median_income	The median household income for the census block group geography. This value is only available for State Farm cleaned data file.
bg_tot_pop	The total population count in census block group. This value is only available for State Farm cleaned data file.
bg_white_pct	The percentage of population that is non-Hispanic/non-Latino White in the census block group. This value is only available for State Farm cleaned data file.
bg_white_tot	The population count that is non-Hispanic/non-Latino White in the census block group geography. This value is only available for State Farm cleaned data file.
latitude	This value is the gridded territory's latitude. This value is only available in the Allstate cleaned data file.
longitude	This value is the gridded territory's longitude. This value is only available in the Allstate cleaned data file.
is_zcta_border	This boolean representing whether Allstate gridded territory point has an adjacent territory in a different ZCTA.
loc_rate_div_min_nn	This is the Allstate territory's generic_location_based_premium divided by the lowest neighboring generic_location_based_premium. This value is only available in the Allstate cleaned data file.
nn_min_val	This is the the lowest neighboring generic_location_based_premium. This value is only available in the Allstate cleaned data file.

Reproducibility

All notebooks committed have been run and display outputs.

We ran our code in a JupyterLab development environment running in a Docker container. The Docker image is derived from the Docker Stacks datascience-notebook image.

The dependencies include packages available in the datascience-notebook image plus geopandas and tabula-py. If needed, install Docker Engine to run the container.

Create and start the container by running this command in a terminal:

docker compose up

Copy and paste the output web address into an internet browser to open the JupyterLab development environment.