How Tenant-Screening Algorithms Keep Cycles of Discrimination Going

Rylan Maksoud outside Ion Austin, the complex that paid him a settlement over pushing him out (Photo by John Anderson)

Rylan Maksoud signed a lease in October 2017 for Ion Austin's cheapest floor plan for the following year. But come Christmas, the West Campus tower's management told him that the floor plan was sold out. They gave Maksoud two options: Pay more for a different unit, or find a new apartment.

Maksoud chose to move out. And sue.

After settling for $4,500, Maksoud learned from previous Ion renters that his experience was not caused by a "software malfunction," as the property management company insisted. Year after year, Ion used algorithms generated with student income and rental data to determine just how much the lowest-paying students might spend to stay in the building.

"The problem is that they have all this data," Maksoud said. "They know exactly how much they can squeeze you [in rent], and they do." Now a University of Texas law student, Maksoud advocates at City Hall against the housing crisis.

Nearly half – 49% – of Americans say housing affordability is a major problem in their communities, according to a 2021 Pew Research Center survey. Austin's affordable housing crisis makes renters highly vulnerable to property managers' use of predictive analytics that can introduce prejudice into the tenant evaluation process.

"Whenever you program an algorithm to do something, you have to ask a very specific question. You have to feed it data to train on, and [this data] might be sourced from a bad source that has biases," said Andrew D. Selbst, a postdoctoral scholar at Data & Society Research Institute, a New York-based nonprofit. Predictive analytics draw on data from potential tenants' social media profiles, criminal history, credit scores, and other personal information.

A 2021 study by the National Bureau of Economic Research found landlords responded to white renters' applications 60% of the time, while responses to Black and Latino renters' applications were 5.6 and 2.8 percentage points lower, respectively. "The idea that you're screening out for who's likely to pay rent will immediately have a racial impact," Selbst said. "Some of that is based on historical reproduction of racism, but then you end up with a conclusion in your algorithm that says people of color are worse credit risks, are more likely to default, and have to be evicted."

Nadiyah Humber, a University of Con­nec­ticut law professor and former housing discrimination investigator under a Depart­ment of Housing & Urban Develop­ment grant in Massachusetts, said she witnessed this prejudice in a Boston-area case, when she sent "testers" – investigators posing as prospective applicants – to uncover race-based discrimination. "There was a housing provider who said to the white, male tester, 'Oh, wow, it's really great that you're considering moving here,"' Humber said. "'We don't want Black people to move into this area, because the property values will go down.'"

Two undecided court cases in Connecticut question the legality of SafeRent Solutions, a widely used tenant-screening algorithm, Humber said. One case addresses the use of criminal records in predictive analytics; the other looks at low-income housing vouchers (what's known as "source of income" discrimination). In the meantime, the Federal Housing Administration still has zero regulations in place to mitigate harms caused to renters by predictive analytics.

"Texas law doesn't provide much protection for renters," Maksoud said. "Landlords take advantage of folks because there's just not enough people vested in it. They can easily abuse their powers."

What might solutions involve? Maksoud said his work focuses on electing government officials who will address the housing crisis and on spreading general awareness across Austin. Selbst said transparency into choices made in designing algorithms, along with specific legislation, could help. "It's hard," Selbst said. "You'd also need community input into what the algorithm should be allowed to do and what counts as discrimination. None of that has been done yet."

Humber suggests that renters should be able to defend themselves against poor credit scores and that people experiencing evictions have greater right to counsel; she also supports an education campaign about the disparate impacts of tenant-screening algorithms. "A lot of people that are harmed by these technologies don't even know the prevalence of these tools," Humber said. "As opposed to just looking at numbers on a database, having landlords deny you based off of the scores alone that are infused with bias, I think there should be a way for individuals to advocate for themselves, to be able to show that they are good renters. That would be more inclusive."