How Tenant Screening Algorithms Produce Racially Disparate Housing and Environmental Outcomes

• In Event: Racial inequality in housing market processes

Abstract

Roughly 90 percent of U.S. landlords use tenant-screening background checks, and most commercial screening providers match applicants to public court records by name because those records typically lack unique identifiers such as a Social Security number. Records belonging to unrelated individuals with similar names therefore routinely appear on an applicant's report. In 2021, the Consumer Financial Protection Bureau determined that name-only matching violates the Fair Credit Reporting Act's reasonable-procedures standard, though it withdrew that advisory opinion in May 2025, and the Urban Institute has separately documented the prevalence of matching errors. Yet no prior study has estimated how such algorithmic errors shape where innocent renters live or under what conditions. 

We merge individual-level InfoUSA consumer records (2022–2024) to the universe of Wisconsin Circuit Court cases for Milwaukee County. For each renter, we construct three classes of false-positive exposure: exact first-last name matches disambiguated to a different person, names one character edit away (Levenshtein distance 1), and names two edits away. All three represent records a landlord cannot distinguish from the applicant's own without identity verification. 

Conditional on each applicant's own validated criminal record, demographics, and income, every class of algorithmic error independently predicts worse neighborhood conditions across all ten outcomes we examine: tract median rent, median household income, college-educated share, poverty and unemployment rates, Area Deprivation Index, Superfund proximity, park count, industrial toxic concentration, and fine particulate matter. Effects concentrate among approximately 43,600 Milwaukee renters with no criminal record, for whom the screening signal is pure error. An owner-placebo test exploiting the fact that homeowners bypass landlord screening confirms the mechanism operates through tenant screening rather than unobserved correlates of name similarity. 

Although per-case effects are approximately uniform across racial groups, Black, Hispanic, and Asian/Pacific Islander renters are matched to systematically more false-positive cases than White renters, even conditional on name commonality. The algorithm thus produces disparate housing and environmental outcomes through disparate exposure. Under a counterfactual zeroing all algorithmic-error channels while holding each individual's own record fixed, Black male renters bear the largest harm on every outcome: up to negative twenty-four dollars per month in tract rent, negative 2.7 percentage points in college-educated neighbors, and the highest increases in Superfund proximity and area deprivation. 

The paper supplies the empirical foundation for policy interventions still available to HUD, the CFPB, the FTC, and state attorneys general: identity-verification mandates for consumer-reporting agencies in the tenant-screening market, disparate-impact liability for screening providers whose name-only matching produces racially disproportionate outcomes, and algorithmic-auditing standards paralleling those emerging in employment and consumer credit. By demonstrating that an automated system reproduces socioeconomic and environmental inequalities long attributed to human discrimination, the paper broadens the fair-housing agenda to encompass automated decision systems in rental markets.

Author

• Xiaoou Zhan, University of Wisconsin-Madison
  Presenting Author