If considered intentionally, eviction filings can offer policymakers and housing experts a clear story of housing access in their area. Although a single eviction filing can speak to only that specific legal case, multiple filings from a tenant, against a landlord, or across a housing market can tell a broader story of housing barriers and risks.

Currently, companies that offer tenant screening services are the most likely to link multiple eviction filings. However, these screens are generally used to create higher hurdles to housing for tenants.

For policymakers and housing experts to use eviction filing data to more effectively house individuals, they need to understand the key challenges, opportunities, and implications of these data. A recent project conducted by Urban Institute researchers using a grant from the US Department of Housing and Urban Development sought to meet this need by testing how stakeholders could pair eviction filings with administrative data.

Eviction filings are problematic, but increasingly used, data sources
 

Eviction filing data recording practices are not designed with systematic use or analysis in mind. County court systems vary substantially in the quality and availability of the data they capture and publicize, making it difficult to gather consistent records across localities. Few states have assembled statewide records of eviction filings.

Additionally, eviction filings often include inaccurate or incomplete information, such as names, addresses, and at times, even the outcomes of cases. Eviction filings almost never include unique identifiers, such as Social Security numbers or dates of birth. These limitations make it challenging to reliably link these records to other court filings and datasets.

Organizations, most notably tenant screening companies, do link eviction filings to other records despite these clear limitations. To use eviction filing data effectively and appropriately, housing providers and housing support organizations will need to explore new ways of data matching so they can better target assistance and understand their local communities.

Linking Housing Choice Voucher Program administrative data to eviction filing data
 

Our project investigated how to link eviction filings to households supported by the Housing Choice Voucher (HCV) Program from 2017 to 2023. We wanted to better understand the patterns of eviction filing among households with housing subsidies, but this goal required us to determine a way to link the data together in the first place.

This approach offered a novel path because HCVs are linked to people, not units. Though some studies have linked eviction records to public housing and project-based housing vouchers, linking to HCVs was a new approach. Instead of assuming addresses and unit numbers remain consistent over time, like a study examining public housing units would, we needed to link HCVs to eviction filings by matching place (where the eviction is filed) and time (whether the resident had a voucher at the time of eviction).

Our team examined eviction filings in two states—Indiana and Virginia—because they offered different matching challenges. In Indiana, eviction filing records contained street-level addresses (which are only available in a few states), while Virginia records only included jurisdiction (e.g., city) and zip code, which is a more common reporting practice. Data were compiled and cleaned by Legal Services Corporation.

Considering how to match distinct datasets
 

For both states, names were key. Because names and addresses can be recorded in many ways, we needed to systematically make decisions to align data. For example, is a person’s first name spelled out? Is a middle name included? Are address numbers written out or numerical? If a record shows Smith Street, should Smith Avenue be considered? In addition to decisions about name styling, we needed to run addresses through a geocoder that returns an output using standardized naming conventions.

We then calculated how similar or different text strings were. Our approach (using what is known as the Jaro-Winkler distance) used the number of edits needed to make the strings identical (ranging from 0 to a perfect match of 1). We conducted spot reviews at various thresholds.

This approach taught us that lower matching scores resulted in a large share of false positives (“type 1” errors). We then identified a threshold where matched records were a much higher likelihood of being true matches (0.94). While hypothetically we would want to pay special attention to matches right around the threshold, in practice, most matches were either lower quality (lower than 0.9) or very high quality (close to 1.0). As a result, even when we only included the very-highest-quality matches, we did not drop many borderline cases (the false negatives).

Most Record Matches Were Either Very High Quality or Much Lower Than the True Match Threshold

Key takeaways for housing organizations
 

We learned three key lessons from this project that can inform the future research and outreach activities related to eviction patterns:

1. Determine the acceptable error threshold. Organizations should consider how important exact matches are versus having a broader set of good matches. This trade-off will be partially driven by the goals of a given project. For example, using matches to conduct outreach to communities with high numbers of filing records could be useful whether a household had already received an eviction filing. On the other hand, if one were conducting a program impact assessment, it’d be important to limit false matches.
2. Geography is a critical input. A matching approach based on street names requires street names from different datasets to use the same or similar naming conventions. But on-the-ground examples surface challenges with this approach. Residents of housing properties with multiple buildings may have some mail sent to a central property management address and other mail sent to a unit address on a different street. Our matching approach cannot by itself identify whether, for example, 100 Elm Street is a match with 4600 10th Avenue. AI-assisted tools could help identify these situations and produce matches, but the tools would need to be operationalized and reviewed carefully.
3. Matching is an art as well as a science. Our research found few threshold cases when reviewing matched outputs, but community context matters and can affect the types of name similarity or frequency. With some names more common within certain communities, such as ethnic and racial groups, these locations may face a higher likelihood of false positives. This potential for bias highlights the importance of local context and local knowledge.

Our project was designed as a proof of concept, but we believe people working on evictions can use our work to understand how eviction filings may affect people in different circumstances. We will be uploading our code on GitHub, which will lay out our steps and processes, with the hope that it will be of use to others working on these issues.

This project was funded from a grant from the US Department of Housing and Urban Development. The findings and conclusions in this piece are those of the authors and do not represent the official position or views of the U.S. Department of Housing and Urban Development.