The use of algorithms and advanced computing across housing activities—such as mortgage underwriting, credit score assessments, tenant screening, and rent setting—has increased in recent years as technological improvements have expanded access to big data.
Today, private companies and public organizations have extensive data and information about people, neighborhoods, and housing trends at their fingertips. By aggregating this information, companies and public offices can create new algorithms that increasingly affect how people find and pay for housing. Yet many of these automations—the use of them, their construction, and their impacts—operate as “black boxes.” The algorithms affect people’s lives, but most have no insight into how they work.
Our research aims to demystify these algorithms and how they influence housing access and affordability. Through our analyses, we provide insights and tools for housing leaders to better understand how the algorithms operate, what impacts they have on people and communities, and the best way to address the barriers to housing that they produce.
Publications
Our research aims to help housing leaders better understand how algorithms affect access to and the affordability of housing. We highlight both the opportunities and the barriers that these algorithms may produce.
Tenant Screening and Housing Access Simulator
This tool visualizes the potential impact of tenant screening policies on housing access by allowing users to modify the look-back period for certain screening criteria. A look-back period is the number of years of criminal or eviction history that can be considered by screening companies or housing providers.
What are tenant screening risk scores?
Most property managers and landlords use information about an applicant’s rental history, credit and income, and criminal background when making decisions about whether to rent to someone. But as technological advances have increased the availability of public and private data, many companies have developed algorithmic models that generate applicant risk scores for landlords. Which records pose a “risk” is unclear, and incorrect, old, and minor records may lead to applicants receiving “risky” scores. Despite little evidence that risk scores accurately predict renter outcomes, landlords appear to rely on them when excluding certain applicants from housing.
How can this tool help changemakers reduce barriers to housing?
To reduce barriers to housing, some states and cities have passed laws that regulate the data included in tenant screening reports or risk scores. These laws, such as fair chance housing laws or eviction filing masking, can limit the negative effects a low-severity, inaccurate, or old record can have on an applicant. These records capture both charges and convictions, so even an accusation that never led to a conviction can count against an applicant.
This tool allows users to simulate how changes to local data protections may change the number of people defined as “risky” by tenant screening companies. Using criminal records and eviction filings from Pennsylvania as a guide, users can apply different policies that limit the records a company can consider and change the look-back (or maximum inclusion) period of data under those policies to see what effect these choices have on tenant risk scores.
The policies we include mirror the following:
- Eviction-masking policies, which seal “not guilty” evictions
- Clean slate policies, which seal low-severity criminal records such as nonconvictions and decriminalized convictions
- Fair Chance legislation, which limits how certain criminal histories, most often low-severity records such as misdemeanor convictions, can be used in tenant screening
These policies minimize the role that low-severity, inaccurate, and old records play in tenant screening decisions.
How should I interpret these data?
Because risk scores—and the algorithms that produce them—are proprietary to tenant screening companies, we developed “inferred risk scores” using court records, existing literature, and input from subject matter experts. The resulting categories—low, moderate, high, and very high risk—do not reflect how likely people in this data are to violate lease terms. Instead, they reflect how likely a company is to portray someone at a given “risk” level to a landlord. In this tool, “moderate” and “high” risk scores signal a high likelihood that a screening company will recommend rejecting an applicant.
Simulate Effects of Clean Slate Policies on Renter Risk Categorization
This tool uses data from Pennsylvania to simulate how state and local policies governing access to tenant data (e.g., eviction and criminal records) might affect “inferred risk scores” generated by tenant screening companies. The baseline assumes companies use all available records, while policy and look-back period adjustments limit the records they can consider.
About the Data
Our tool focuses on millions of court records from Pennsylvania's Administrative Office of Pennsylvania Courts, including cases filed or deposed between February 2014 and February 2024, shortly before Clean Slate 3.0 was enacted in the state, which sealed millions of records.
Millions of “low risk” individuals are not shown here, as these data only include criminal and landlord-tenant defendants. Each state categorizes crimes in different ways, meaning that comparability of misdemeanors and felonies across state lines may vary. Details on our full methodology can be found in our accompanying GitHub repository here.
In our race and ethnicity charts, note the following:
- Given incomplete data, we impute missing values for race and ethnicity. For more details, see our full methodology and evaluation here.
- “Additional groups” includes data for those who identify as multiple races or as a race not otherwise listed.
- "Hispanic" refers to those who identify as being of Hispanic, Latino, or Spanish origin.
- All other included racial groups refer to those who identify as part of that race and as non-Hispanic.
Project Credits
This data tool was developed with support from the Salesforce Foundation. We are grateful to them and to all our funders, who make it possible for Urban to advance its mission. The views expressed are those of the authors and should not be attributed to the Urban Institute, its trustees, or its funders. Funders do not determine research findings or the insights and recommendations of our experts.
We are grateful to Manu Alcalá, Alena Stern, Erika Tyagi, Jesse Janetta, and Will Englehardt for their review of this tool and our analysis.
RESEARCH Judah Axelrod, Katie Fallon, Adriana Vance, and Brendan Chen
TOOL DEVELOPMENT Rebecca John and Evy Park
PRODUCTION Samantha Cressman
EDITING Wesley Jenkins and Alex Dallman
