The US is using a race-predicting algorithm to identify victims of car loan discrimination

The US government has started sending checks worth $80 million to people it believes were unfairly overcharged for car loans as a result of racial discrimination.

They key word there is “believe.” The lender in this case, Ally Financial, was prohibited by law from collecting data on its customers’ race or ethnicity.

So the government is guessing which borrowers during the period in question were black, Hispanic, or Asian using a race-predicting algorithm based on surnames and location.

Created in 2009, Bayesian Improved Surname Geocoding, or BISG, uses publicly available information on surnames and geographic location to predict an individual’s race.

While BISG is the one of the most accurate tools available right now in the absence of hard data on race demographics, it’s still imperfect. As a result, white borrowers in this case have received checks for discrimination they never experienced. People who receive checks in error don’t have to return them.

The government has used BISG to gauge demographics on health care services and mortgage applicants. Its predictions hew fairly close to self-reported data, though BISG and other race-predicting tools tend to overestimate proportions of minorities.

In Quartz’s quick and completely unscientific survey, 70% of 72 friends and colleagues who used the tool received accurate assessments of their identified race. Another 20% got inaccurate results, and 10% had surnames too rare to gauge. The Wall Street Journal (paywall) has the tool; try it yourself if you’re curious. (Side note: While “multiracial” is an option, the algorithm has a hard time deploying it correctly. At least 40% of our respondents with inaccurate answers are of a multiracial background.)

A 2013 investigation by the US Consumer Financial Protection Bureau and the US Justice Department found that dealers affiliated with Ally overcharged an estimated 235,000 minority customers via a mechanism called “dealer reserve,” in which dealers are allowed to add extra interest to a loan as a fee for setting it up.

The CFPB wants to get rid of dealer reserve, saying it’s an easy vehicle for discrimination against women and minorities. The Ally case alleged that black and Hispanic customers were overcharged up to $300 over the life of the loan. Since lenders can’t collect data on race, the government used BISG to gauge the number of potential victims as well.

In a carefully worded settlement in which the government said it did not believe the discrimination was deliberate, and Ally did not admit or deny wrongdoing, Ally agreed to pay at least $80 million to borrowers.

The Ally payout is the first in a series of similar settlements the government will soon disburse. With millions on the line, the auto industry is taking issue with the use of BISG, and the House Financial Services Committee is investigating its use in the Ally settlement.

Proponents argue that BISG is the best option available at the moment for assessing customers’ race when no hard data is available, and that seeking an alternative would eat up too much money intended for discrimination victims.

home our picks popular latest obsessions search