Many people living on the street today have been there for years, but they had never previously imagined they’d end up there for so long—until they did. Misfortune is unpredictable—arriving with an abrupt job loss, mental breakdown, the loss of a spouse, or an unplanned pregnancy. Researchers are now attempting to wield big data and predictive analysis to help tackle chronic homelessness by stopping it before it starts.
The Economic Roundtable’s (ERT) analytic model for homeless risk assessment combines emerging technological tools to systematically assess social vulnerability among people who have recently become homeless, and to target services for prevention for the people who are most prone to falling into long-term homelessness. The aim is to triage services for people at especially high risk of spiraling into persistent homelessness—particularly among youth and the working poor—to deliver the help they need before a temporary crisis becomes a permanent one.
The ERT model is based on the understanding that homelessness is not just a social problem but a social process. Social-service agencies can use the model to connect individuals to resources that provide more intense, sustained intervention than just a night in a shelter or a temporary food voucher. For some that might mean a job training program; for others, drug rehab. The model was developed against the backdrop of a roiling debate over the ethics of “predictive” social science in public policy. Social analysis can tell us a lot about how long-term homelessness happens, but can it be prevented if we can foresee it months or years in advance?
ERT is currently testing its model on Los Angeles County’s vast and diverse homeless population, which encompasses everyone from serial couch surfers to Skid Row residents. The group received some pushback from L.A. County officials, stemming from complaints that ERT’s project exceeded the scope of its original authorization to use the county’s public data, and the local government has not officially adopted the tool. But ERT hopes the program can be applied on a smaller scale to help change the way homeless services work.
Given the limitations of the current social safety net, researchers have focused on the roughly 40 percent of homeless individuals who suffer “persistent homelessness” (homeless for at least twelve straight months or for at least two episodes within three years). The predictive analysis model maps the pathway to chronic homelessness for certain high-risk groups who could most benefit from preventive measures—like the fast-food worker who suddenly got laid off, or the drug user who breaks down after an overdose, or the college student forced to choose between rent and tuition. ERT’s model for predicting future long-term homelessness with a reported accuracy rate that is several times higher than just random selection. Once they are identified, prevention can range from something as simple as transitional housing for an evicted family or as complex as intensive therapy for veterans with post-traumatic stress disorder (PTSD).
When assessing L.A.’s newly homelessness population, ERT uncovered key destabilizing triggers that often foreshadow chronic homelessness. For many of the homeless working poor, social factors associated with long-term homelessness include serious mental illness, disability and chronic health problems, HIV/AIDS, and barriers to employment like substance abuse or family-care burdens. A history of incarceration is also linked to chronic homelessness, as is PTSD and alcohol abuse. Black workers are more than twice as likely as Latinx or white workers to become persistently homeless. Among youth, persistent homelessness is also linked to having lived in foster care, incarceration, and being black. Not surprisingly, ERT found a direct correlation between the time spent in homelessness in adolescence and risk of long-term homelessness during young adulthood—which reflects the destructive cycle the predictive model aims to break. These homeless subgroups are, at least statistically, primed for becoming “stuck” in a spiral of destabilization, trauma, and displacement—but they can steer clear of that risk if they get help before crisis strikes.
The early-intervention model for homelessness is unique in the emerging and controversial field of predictive analysis. Other prominent uses of predictive technology have centered on criminal law and policing, often facing criticism for fueling racial profiling of marginalized communities. But predictive analysis to prevent long-term homelessness is not intended to “filter out” people for negative treatment, but rather to triage and streamline services to maximize effectiveness. According to Dan Flaming, executive director of ERT, the homelessness early-intervention system is designed to ensure that there “is not a negative feedback loop, as with policing models that reinforce and validate biased assumptions.” Instead, the front-end approach contrasts with the current one-size-fits-all welfare system, in which “service providers are not able to differentiate someone who will be chronically homeless three years from now from someone who will make it out of homelessness in three months with very little help.” As for the potential for bias or other abuses of data, ERT emphasizes that the system would only be used to help prioritize people for certain services, so there would not be criminal justice implications. Nonetheless, Flaming did note that a predictive analysis system has inherent data-confidentiality implications, and “[p]eople who carry out the screening have an ethical and legal responsibility to protect individuals’ privacy.”
Predictive models prioritize people for individualized social services. For individuals coping with substance abuse, immediate access to treatment can help them avoid further deterioration on the streets. A young single parent may simply need subsidized child care or training to get back in the job market. A formerly incarcerated youth might need help with expunging a legal record to gain access to decent jobs. Former foster-care youth struggling with trauma can be routed to transitional support services as they move toward independence.
ERT’s predictive model is not being applied by local authorities on a wide scale, but the group is working toward a pilot phase through collaboration on apprenticeship programs run by the California Federation of Labor. Predictive analysis will be integrated into the program to identify prospective trainees among at-risk homeless youth. Flaming notes that since the initiative aims to recruit youth who would likely otherwise be overlooked, opening a pathway from homelessness to a career “would provide inclusion and representation for extremely marginalized and frequently excluded individuals in these training programs.”
The feedback loop, in fact, could flip the traditional charity paradigm upside down for many of those who are typically framed as “at risk.” If the system can reach people early on in their housing crisis, service providers can catch them before they hit rock-bottom, and communities can ensure that a temporary stumble does not become a permanent quagmire.
Michelle Chen is a contributing editor at Dissent and co-host of the Belabored podcast.