Discrimination Caused by Digital Algorithms


  • Algorithmic decision making has become increasingly prevalent, but it also creates the potential for subtle new forms of discrimination

  • Algorithms are also usually a black box, with almost no transparency on their inputs

  • It is difficult to show commonality in US discrimination class actions because the Supreme Court has ruled that discriminatory actions by a subset of employees are insufficient to show commonality



  • We used regressions to analyze US mortgages and examine if borrower ethnicity had a statistical relationship with their interest rate

  • We controlled for a range of credit worthiness variables, while looking exclusively at loans issued by “Fintech” lenders, who use algorithms to originate loans



  • We identified algorithmic lenders who discriminate against ethnic minorities

  • For these lenders, we showed a strong positive relationship between borrower race and the interest rate they pay, with black/latinx borrowers receiving loans with 5-7 bps higher interest rates than comparable white borrowers

  • This methodology provided a new path to show commonality in US class action discrimination cases, where there is standard centralized decision making through algorithms