Is Car Insurance Racist?

Early in 2024, I was commissioned by Kinetic Invest and Investopedia to research if auto insurers in the USA unfairly charge more to non-white drivers than they do white drivers. Here’s how I did it.

Gathering Insurance Rates

The above image is a screenshot of this tool, published on carinsurance.com, which I used to gather information on the average cost of car insurance across all zip codes in America. By default it compares rates for males, aged 30, driving a 2017 Honda Accord, and looking for minimum liability coverage.

I wrote an automated user using selenium, to allow myself to use the tool to analyse my list of every US zipcode automatically. It allows you to write code which looks for specific patterns in the HTML of a website, and interact with them as if it were a regular human user. I simply look for the box where I enter the zip code, the ‘search’ button, and then the dollar amount as a result. I left the default gender, age and car values as they were.

Many modern websites have protection to stop you from overworking their servers with scrapers like this, but I didn’t encounter any on carinsurance.com. Still, I added delays to process slowly to not cause any issues to their server.

High Rates for Whom?

With a database of the average insurance rate by zip code, I accessed US demographics by zip to find who was paying the higher and lower rates.

At first glance at this point, it looks pretty damning. Areas with a low density of white people paid 68% more for their car insurance. Plotting the results, it seems clear that the two variables have a strong correlation:

Removing Deniability

You’ll hear a certain type of person claim that these rates are just, and are simply based on crime and traffic accident data. Luckily, that data is also available for free online.

I gathered crime rates from neighborhoodscout.com and accident data from this Kaggle dataset. Isolating crime rates and accident rates, non-white areas were consistently paying more than double white areas. Comparing 2 zips from Montana and Delaware which had very similar crime rates, accident rates, and population size, but one was heavily white majority, the other heavily black majority, and predictably, the black area paid over 3 times more per month.