How Data is Helping Homeless Shelters Curb The Cycle of Readmission

One of the many tragedies of homelessness in New York City is that passing through the shelter system might leave a family in as precarious a position as they were in before they arrived—with many families returning to the shelter not long after they leave. “To me, success means stabilizing families first and foremost, so that they do not have to return three years or five years after they leave,” says Raysa Rodriguez, Vice President of Policy and Planning at Win (Women In Need) shelters, the largest provider of its kind, serving over 10% of the city’s homeless families. “Unfortunately, in New York, this is not true of many families.”

New York City’s homelessness is at its most severe since the Great Depression, with more than 127,000 people seeking refuge in the city’s shelter system at some point during the 2016 fiscal year. According to the Coalition for the Homeless, the primary cause is not as much unemployment, disability, mental illness or drug addiction as it is a lack of affordable housing—even for families with a breadwinner. Families, in particular, often find themselves homeless as a result of eviction, and 17% of families with children return to shelters within 30 days of leaving one.

How might data analytics not only help to design more effective shelters, but also predict whether a family is at risk of being readmitted after they leave? These are some of the questions that the Urban Intelligence Lab at NYU’s Center for Urban Science + Progress (CUSP), in collaboration with Win, hopes to explore. Under the leadership of Dr. Constantine Kontokosta, Assistant Professor of Urban Informatics, Director of the Urban Intelligence Lab, and Deputy Director of CUSP, a team of researchers, supported by funding from the MacArthur Foundation, is integrating numerous datasets provided by Win and the City’s Department of Homeless Services to develop comprehensive predictive models that might be used to help Win – and shelters across the City and country – to provide better services to the homeless families they support.

The datasets, which are anonymized, include the demographic, historical, familial and medical data for about 6,000–7,000 families, dating back to 2012. “One of the challenges of this project is getting the right data in the right form,” says Kontokosta. “We’re working with a lot of legacy systems used by the Department of Homeless Services and other agencies, and some of the data we have—notes by caseworkers, for example—exist only in handwritten form.” The question, then, is how to digitize and create common standards and metrics across datasets.

New York City’s homelessness is at its most severe since the Great Depression

To understand how valuable such a project might be, it’s worth considering how useful both quantitative and qualitative data is in helping us to imagine the life of a typical Win client who we’ll call Theresa. Theresa is in her 30s, with two children, about average for Win clients. Like the majority of clients passing through Win’s doors, Theresa is a single mom. She and her boys ended up at the shelter after being evicted by their landlord. Along with 53% of Win clients, Theresa does not have a high school diploma, setting her at a disadvantage when looking for work. She works as a retail assistant, placing her among the 46% of Win clients who are employed. Like many of her fellow Win clients, she works more than one job, sometimes picking up shifts at a call center over nights and weekends. Despite working overtime, Theresa makes little more than $1400 a month—about average for Win clients who are earning—and barely above the federal poverty line.

With long hours at work and a grueling school schedule and commute, staying in a shelter is hard on the whole family. Theresa’s sons are physically and emotionally exhausted, they struggle to concentrate and sometimes pick fights with the other kids in their class. Their performance at school is both significantly poorer than it was when they were living in their own home, and when compared to other children their age. At the shelter, emotions run high and Theresa sometimes finds herself in altercations with other clients. When Theresa’s caseworker meets with her each week, he finds her difficult to engage with because she is so frustrated by her situation. Theresa and her boys stay at the shelter for more than a year, forming part of 25% of families that live in shelters for more than nine months. Eventually, Theresa and her sons are set to move into a cheap one-bedroom that she found through a friend. But before long, Theresa finds that even by juggling two jobs, she cannot keep up with the bills and supporting her sons financially and emotionally through another destabilizing move. Within a month the family of three finds itself back at Win, making them part of 17% of families with children who return to the shelter system within 30 days of leaving.

For CUSP, risk of readmission to the homeless shelter is a key issue they aim to explore through data analytics. “One of the biggest challenges for shelters and homeless families is that readmission becomes a permanent cycle,” says Kontokosta. A recent report by the Institute for Children, Poverty and Homelessness suggests that the growing population of the shelter system is primarily caused by clients who stay long term and have repeated entries, as opposed to new clients. “The data can help us to identify red flags that indicate when a family is likely to be admitted again, which means the shelter can provide targeted support and intervention. The analysis can be used as an additional tool to support and enhance the on-the-ground knowledge and expertise of case workers.” In September, Kontokosta and his team presented their proposal and preliminary findings at the Bloomberg Data For Good Exchange Conference in New York City. In addition, a paper on their work has been accepted by the Journal of Technology in Human Services.

The collaboration between CUSP and Win, originally convened by LOOM Media—a one-stop solution for corporate marketing programs tied to city service and innovation—is also working on predictive analytics for facility maintenance and operational efficiency, and developing a “Smart Shelter” road map for integrating technology into shelter design. In the coming months, they will be testing new programs that aim to expand wifi access inside the shelters and install quality-of-life sensors that monitor air pollution and noise.

“We might not be able to control the lack of affordable housing and rental subsidies in the city, but if we can isolate the elements that we can control, there is a lot we can learn about what we are doing well, where we can improve and what some of the predictable outcomes might be,” says Rodriguez. Factors within Win’s control include staff quality and training (is Theresa’s caseworker sufficiently trained to help her?), security and shelter operations (are altercations within the shelter making Theresa feel unsafe?), child services (are Theresa’s sons getting the support they need?), referrals to community services (is the family sufficiently supported and prepared when they leave the shelter?) Rodriguez considers these questions and says, “Do these influence client success or not? Learning more about this is why this partnership is so exciting.”

Image courtesy of Win

For NYU In Brooklyn, 2017