WHEN IT COMES TO COMBATING HOMELESSNESS, the most effective tool may be preventing it in the first place. After all, it’s easier and cheaper to keep someone in an apartment than to get that same person off the streets and back into housing. That holds particular sway in Los Angeles, where nearly 60,000 people live without shelter.
Yet acting before someone becomes homeless is not as simple as it sounds and prompts some pointed questions: What does “prevention” actually entail? Is it just giving money to someone in need, or is more required? Who should receive scarce prevention resources?
Finding the answers is complicated by the fact that it’s impossible to tell just by looking at someone whether that person will wind up on the streets. In practice, local officials often cast a wider net than is required to stave off homelessness.
As Phil Ansell, director of the L.A. County Homeless Initiative, said in an interview, “Investing in homelessness prevention has been challenging, because it generally takes assisting 10 families or individuals to actually prevent one family or individual from becoming homeless.”
APPEARANCES MAY DECEIVE, but data are helpful. In fact, according to a team at UCLA, a person’s data, thoughtfully examined, can help experts predict, with a high degree of certainty, whether that person is likely to become homeless.
The case is laid out in “Predicting and Preventing Homelessness in Los Angeles,” which was released late last year by the UCLA California Policy Lab and the University of Chicago Poverty Lab. The 14-page report details how researchers examined records of service interactions from seven L.A. County agencies to anticipate who would become homeless. While the report looked at past records, researchers hope that honing the research and refining the response could make future homeless prevention efforts more effective and efficient.
Prevention is not a new concept. Chicago and New York City have developed prevention programs to help keep at-risk people in housing, and Los Angeles has undertaken similar work. In these instances, the process starts when someone calls a hotline or visits a government agency and asks for aid. That is followed by a screening process.
Beth Horwitz, vice president of strategy and innovation at All Chicago, which handles the city’s response, said qualified individuals can receive funds within three to five days of asking for help.
“Predicting and Preventing Homelessness in Los Angeles” lays the groundwork to help a different pool of people, says Janey Rountree, one of the report’s authors and the executive director of the California Policy Lab.
“What’s different about our work,” Rountree said, “is that it allows the county to be more proactive and to identify people who maybe are not going to walk in the door and raise their hand and say they are at risk for whatever reason.” Rountree said.
The “risk list”
RESEARCHERS UTILIZED COUNTY DATA compiled from 2012 to 2016. They examined the interactions that 1.9 million people had with agencies such as the Department of Health Services, the Sheriff’s Department and the Department of Public Social Services.
Through a computer-aided process of determining which interactions were the most critical, the team sought to predict who would experience a homelessness spell in 2017. While the material was not accompanied by names, researchers compiled a “risk list” based on the interactions.
Rountree said the team did not enter with presumptions about what factors would be most meaningful in anticipating homelessness, but treated all interactions equally at the beginning. During the data analysis the computer considered everything to determine what interactions were actually predictive. The team learned that frequently, low-income individuals who lost housing often lacked a community or family safety net.
The report said that of the 3,000 people deemed at highest risk, almost 46% became homeless in 2017, and that those individuals were 27 times more likely than the average county client to lose housing. The research also holds in a broader context; if expanded to the 19,600 people most at risk (or the top 1% of county clients), 35% experienced a homelessness spell in 2017.
The report posited that effectively serving that 1% would prevent 6,900 homeless spells a year.
Till von Wachter, another of the report’s authors and faculty director of the California Policy Lab, said this kind of work has been facilitated by advances in data collection and data science. He also noted that it takes time to determine which interactions are the most meaningful; for example, the number of visits to an emergency room may be less important than what caused those visits.
Ultimately, von Wachter said analyzing and honing data creates a “predictive probability” that a person will lose their housing in a set period.
“That is essentially a summary measure of all the services that the person receives weighted by how likely each of the services would contribute to a new homelessness spell,” said von Wachter, who is also a professor of economics at UCLA. “And the result is a single number for each individual that is the probability of becoming homeless.”
Having that number allows the creation of a ranked risk list. It can count the 3,000 people most at risk, or 19,600 people, or any desired figure.
Rountree said one of the most predictive features was the level of poverty in the area where someone receives services. Yet she noted that losing housing is a result of more than just being poor, that a catalytic event often leads to someone winding up on the streets.
This could be a mental health issue, a medical matter, a job loss or some- thing else. She added that the event usually happens quickly, and a bad six-month period can be devastating.
“While you and I might survive [the event] without losing our housing,” she said, “for people living in that type of extreme poverty, it’s often a triggering event and destabilizes their housing.” she said.
The need for intervention
ONCE THOSE AT RISK ARE IDENTIFIED, changing circumstances can affect their situation. The risk list needs to remain fresh, with continual new streams of data that allow professionals to know which county clients are in the most danger of becoming homeless — potentially even before the person recognizes the threat.
“A risk list calculated a year or six months earlier will be less useful than a risk list calculated a couple of months before,” von Wachter said. “We will be using data that [is] very timely to refresh the risk list on a regular basis.”
The report’s authors and the county are working together to create a pilot program to apply the predictive capabilities in a real-world setting. The funding is coming from Measure H, the quarter-cent sales tax approved by county voters in 2017 to pay for homeless services. The aim is to have a pilot project in place by the end of this year.
Yet having data that identify risk means little if there is no effective intervention — in this case county government must be able to take the information and respond quickly.
Ansell said this will be accomplished by establishing a County Centralized Homelessness Prevention Unit, with staff dedicated to working with the risk list and directing help where required.
What helps most?
THEN THERE IS AN IMPORTANT QUESTION: What form does help take? The researchers and Ansell say it will be varied and tailored to the individual or family at risk. It could be monetary, or it could involve connecting a client to legal aid if someone is facing eviction. It could mean enrolling individuals in a county program they are qualified for but have not utilized.
Ansell says there are a wealth of possibilities. “Our current prevention efforts focus heavily on short-term financial assistance, either payment of back due rent, or current rent, or paying to turn the utilities back on that have been turned off,” he said. “By intervening sooner, in advance of an immediate risk of homelessness, we anticipate that the range of assistance that will be most appropriate will vary more greatly than is currently the case with families and individuals who are at imminent risk of homelessness.”
Moving from a study to actual homelessness prevention is a big step. Horwitz of All Chicago said success in her city has come from building a network of strategic partners and having people on the “front lines” who can quickly aid individuals experiencing a housing crisis.
She said the All Chicago team will be closely watching what happens in Los Angeles. She called the California Policy Lab work “an important step forward” but also acknowledged concerns.
“Predictive analytics … can come at a cost — they are built using algorithms on general trends but may not be true about the individual, and if not used carefully, can end up being used to pathologize certain groups,” she said.
The Los Angeles team recognizes that identifying people at risk, and intervening before they lose housing, is a new and untested approach.
“It’s worth noting no one has ever done this before,” said Rountree, pointing specifically to the computer-aided, data-focused predictive component. She added that the pilot program is “where presumably we learn a lot about how to reach out to these people and what their needs are.”
The irony is that, if local efforts succeed, it may be impossible to know precisely who was helped on an individual basis, and only raw data will tell the case. After all, it’s easy to identify someone who falls into homelessness. Pointing out a person who avoided homelessness because they were given the right help at the right time is much more difficult.