To some, the terms “logistic regression,” “cluster algorithm” or “management information system” might seem a little wearisome. To the University of Washington (UW) students, who spent the summer crunching numbers on family homelessness in the Puget Sound region, they are tools for social change.
“Data contains a great deal of information that is not really available any other way,” said Bryna Hazelton, a research scientist at the UW eScience Institute. “To understand whether something works in the way you want it to be working, you really need data, and you need to be able to analyze it.”
At the end of August, the eScience Institute wrapped up its new 10-week Data Science for Social Good (DSSG) program — modeled after similar programs at the University of Chicago and the Georgia Institute of Technology — in which teams of students and data scientists worked closely with organizations on data projects aiming to answer questions of public policy and social good.
One team focused on family homelessness in King, Pierce and Snohomish counties, building off a partnership between the Bill and Melinda Gates Foundation, the nonprofit Building Changes and county stakeholders, who have been working together under an initiative to cut family homelessness in the region in half by 2020.
The team used data from the Homeless Management Information System to look at factors that might predict whether a homeless family finds permanent housing. Four student fellows, two high school interns with the UW Alliances for Learning and Vision for Underrepresented Americans program, eScience Institute data scientists Hazelton and Ariel Rokem, and Gates Foundation research leads Neil Roche and Anjana Sundaram round out the team.
According to annual point-in-time counts, there are more than 4,000 homeless families with children on any given night in the Tri-County area.
“There’s a lot of discussion about a lot of different approaches to helping solve homelessness that really runs the gamut,” Hazelton said. “There’s really a need to understand what actually works.”
While analysis typically looks at success rates in individual programs, the team focused instead on trajectories: a family’s path of travel through the social service system and how that affected the outcome — whether the family had a “successful exit” into permanent housing.
They found, for example, that families who went solely to emergency shelter were largely unsuccessful, while success rates for those who went from emergency shelter to transitional housing — one of the most common paths — increased twofold.
There are limitations to the data: It only tracks families that have enrolled in the system, and because programs are typically full, that number is restricted by capacity rather than need. Data also doesn’t follow families once they leave the system.
Given that, the data is far from providing concrete conclusions. Still, students said, they can offer stakeholders new ways of understanding, viewing and collecting data. With complicated, “noisy” information that varies from county to county, just determining whether two people are part of the same family can become a tall order. But the students were able to offer creative solutions as they put their heads together five days a week on the sixth floor of the UWphysics and astronomy tower.
“Identifying problems in the data that prevent analysis from happening, that’s really important feedback,”
Hazelton said. “Data does not come in an easy, ready-to-go package.”
Student Chris Suberlak had the idea of using an algorithm typically used for clustering galaxies in astronomy to help group homeless families. At one point, the team helped a county uproot a significant error in its data system. And in response to stakeholder’s request for a visual representation of how families move through the system, the students created a Sankey diagram — a flow chart often used by engineers to track energy.
“We believe we’ve come up with things that will help the county make better data-driven decisions,” student Joan Wang told the audience at their final presentations on Aug. 20. The next day, students presented their findings to the Gates Foundation and representatives from the three counties.
Though students may have only scratched the surface in their 10-week stint, their work has implications for the future. Potentially, it will allow policymakers to answer questions about how families typically move through programs, what barriers exist for finding permanent housing and what paths commonly lead there.
“If data is available, we can make rational decisions,” Rokem said. “It’s more of a social and political question to say ‘How important is it to society that we know how many families are homeless? What kinds of actions can we take within the system that actually help?’”