Study: UB researchers create early warning system to monitor housing evictions

Human service organizations play an important role in connecting people to housing, health care, food access and other essential services. Yet forecasting community needs can be difficult in times of rapid change or crisis due to limited resources and restricted access to data.

For instance, during the COVID-19 pandemic, concerns about a potential surge in evictions exposed gaps in the data used to assess housing instability.

“The fear of an ‘eviction tsunami’ during COVID‑19 was very real because the sudden economic disruption made it hard for many households to keep up with rent,” said Maria Rodriguez, MSW, PhD, assistant professor in UB’s Department of AI and Society. “Unfortunately, many community‑based organizations couldn’t determine which households were most at risk because detailed housing data is frequently locked behind paywalls and not easy to access. Without that visibility, it was challenging to prepare for a wave of evictions that, at the time, seemed likely.”

A new case study by researchers at the University at Buffalo and three partner universities looks at whether publicly available data and open-source tools can help predict eviction filings during crises like the COVID-19 pandemic. The collaboration brought together experts in social work, computer science and data science.

Public datasets often lack detail, rely on multiple sources and aren’t consistently updated. These factors can affect accuracy and sometimes lead to higher-than-actual estimates. However, unlike proprietary systems that require costly licenses or restricted access, these resources are available to human service organizations at no cost.

“When the pandemic began, we were deeply concerned about the potential impact of evictions on homelessness as well as the resource constraints agencies were facing,” said Rodriguez, who led the study. “This led us to examine how effectively open-source data and statistical tools could anticipate trends in eviction filings. Even with limitations, these approaches can help human service professionals predict emerging challenges and respond faster when communities are under strain.”

Rodriguez was joined on the research team by Kenneth Joseph, PhD, associate director of UB’s Department of AI and Society and associate professor in UB’s Department of Computer Science and Engineering, and Jan Voltaire Vergara, a recent graduate of UB’s Department of Computer Science and Engineering. Additional co‑authors include Erin Dohler, PhD, and Amy Wilson, PhD, of the University of North Carolina at Chapel Hill; John Phillips, PhD, of the University of Minnesota Duluth; and Melissa Villodas, PhD, of George Mason University. 

The study, which was published Aug. 9 in the Journal of Technology in Human Services, grew out of weekly conversations among the research team during the early months of the COVID-19 pandemic. As economic disruption increased the risk of eviction and homelessness, the team focused on New York’s Bronx County, one of the communities hardest hit by both the foreclosure crisis and the first wave of infections.

To reflect the challenges faced by many small human service agencies, the team used open-source tools to analyze publicly available datasets on eviction filings, demographics and employment trends. They then built statistical forecasting models from this ZIP code-level data to project eviction filings in Bronx County from January 2020 through July 2021 and compared those projections with what occurred during the same period.

Across all models, projected eviction filings were about 2.6 to 3.3 times higher than what was actually reported. These findings suggest eviction trends could have been far worse without government protection policies, such as a temporary eviction moratorium.

“The results show that open-source tools can provide a quick, big-picture analysis of public data and generate meaningful insights, which is an important capability for organizations responding to a crisis,” Rodriguez said. “Even when projections exceed actual outcomes, they can help agencies prepare, allocate resources and advocate for support before conditions worsen.”

The analysis also reinforces patterns that emerged during the pandemic: Communities with higher proportions of people of color experienced some of the largest employment declines early on. This highlights how open-source tools can help organizations see where inequities are most concentrated and prioritize services accordingly in the communities most affected.

Rodriguez added that human service organizations will need stronger data skills to get the most out of these tools.

“Our work demonstrates that open‑source tools can help forecast housing instability, while also underscoring why stronger public data systems matter if we want more equitable outcomes,” she said. “Human service agencies don’t necessarily require a team of data scientists, but having someone who can manage and analyze data is important. With the right skills, even small teams can use open data to make better decisions during future crises.”