Might Data Help the Destitute? Analytics to Address Homelessness

Abigail Rose Lindner
Abigail Lindner
Regent University

Homelessness is a worldwide social and economic challenge; the United Nations Human Settlements Program estimates that the homeless population, defined as those individuals not in a permanent housing situation, is about 100 million. Because of the lack of data and political sensationalism in the matter, homelessness does not receive the amount of discourse and treatment that it should. However, in the past five years, data and social science researchers have explored the application of data technology and predictive analytics to confront this human tragedy.

The Chair for the Institute of Global Homelessness, Dame Louise Casey, lists a three-pronged approach to addressing homelesseness, which may be summarized as Prevention, Provision, and Reformation (Beckwith, 2020). Behind each, the force of data and modern technologies has been rallied to re-evaluate the policies undergirding homelessness programs.


This first prong of Casey’s approach uses predictive analytics to determine what factors aggravate the risk of homelessness and what public policies ameliorate the risks. Like most social problems, homelessness does not follow a linear path; it involves a “complex interplay of individual, interpersonal, and socioeconomic factors” (Fowler, Hovmand, Marcal & Das, 2019). Mashariki (2018) emphasizes an urban analytics approach that layers data on poverty, unemployment, affordable housing, individual health, substance abuse, and other factors to create a more robust picture of homelessness.

In Los Angeles County in California, where about 141,000 people experience some amount of homelessness every year, researchers from the California Policy Lab, the LA County Homeless Initiative, and others collaborated to analyze the 85 million service utilization records from the County’s Enterprise Linkages Project. The goal: predict return to homelessness, or RTH, and first-time homelessness, or FTH, based on inpatient and outpatient visits with state service agencies, such as the Department of Health Services and the Department of Mental Health. A preliminary model produced from this project indicated that “[t]he majority of single adults who will experience [FTH] or a [RTH] are already clients of mainstream County agencies” (Wachter, Bertrand, Pollack, Rountree & Blackwell, 2019). By targeting the 1% with the highest risk for homelessness, the county could prevent nearly 6,900 instances of homelessness in a year.

In New York, the Allegheny County Department of Human Services introduced an algorithm in the late 2010s to likewise identify high-risk individuals, as determined by use of mental health services, emergency rooms, and substance abuse services or interaction with the criminal justice system (Castillo, Zamorana, Jaramilla & Gonzalo, 2020). In both California and New York, these metrics are used because consistent research in the past two decades has demonstrated that, among the homeless in developed countries, an interaction between mental health, physical health, substance abuse, family conflict, and crime can predict the risk of homelessness. Structural disadvantage, including specific policy and group-based discrimination, is another consideration. This algorithm, judged to be accurate and non-discriminatory for dimensions like race and age, enables Allegheny County to identify high-risk individuals and allocate resources accordingly, confronting homelessness before it happens.


The second prong of Casey’s approach acknowledges that prevention does not help the already homeless. The insight gained from predictive analytics provides direction for public response to homelessness. As homelessness is not a homogenous condition, a heterogeneity of systematic responses is demanded. In 2015, the Built for Zero movement at Community Solutions spear-headed a data-driven approach to meeting the needs of homeless populations in, as of writing, more than 80 communities in the United States. Whereas previously agencies acted separately and relied on census data to measure homelessness, Built for Zero creates a database of comprehensive, real-time, multi-agency data for each single adult experiencing homelessness in a given community so that solutions can be individualized and program success based on individual change rather than organizational outcomes (Broom, 2019). Provisions include food, shelter, and medical treatment that care for the impacted individuals in the short-term.

The construction of these databases assumes knowledge about the identities and locations of homeless individuals. The truth, however, is that this population, by dint of lacking permanent residence, is among the hardest to track. The introduction of mobile and GIS technologies, such as the Homeless Outreach Portal in Los Angeles County, has facilitated the collation of real-time data on homelessness in recent decades (Kelkar, Frey, Suriya & Engel, 2019). In the United Kingdom, the Homeless Data England project, operating within the Ministry of Housing, Communities & Local Government, is working on a national scale to improve data on homelessness program outcomes and homeless population (Wu, Man, Taylor & Aldridge, 2020).


The third prong of Casey’s approach draws from the insights from the first prong and the measurable results from the second to inform the restructuring of the entire homelessness system; it moves from a short-term and immediate care framework to long-term and sustainable solution. Any hope of eradicating homelessness - eradication in the sense that supply of services for homelessness exceeds demand - requires a coordination among agencies at a number of action levels, not just the federal, state, or local governments.

Data and digital technology are essential for this multi-stakeholder coordination. The Principal at Community Solutions, Beth Sandor, names five traits that communities that have reduced homelessness share in common: a commitment to a shared target across all programs and investments, data and feedback on system progress and human outcomes, a dedicated team with the capacity to work in and change the system, proven practices that can be tested and adapted in every community, and access to flexible resources to quickly respond to new information or needs (Beckwith, 2020).


Figure 1: A homeless person in downtown Toronto on January 3, 2018. Photo by Christopher Katsarov/The Canadian Press.

Consider this broader “ecosystem approach” to homelessness (Kelkar et al., 2019) as applied to housing policy. As the unifying factor among the diverse individuals of the homeless population is want for a permanent housing situation, it makes sense that a key focus is on providing housing.

Early experimental studies in the late 20th century and large-scale studies in the United States and Canada in the 2010s suggest that a housing first philosophy, wherein housing is arranged quickly and without preconditions, results in better social stability than the treatment first philosophy, wherein an individual receives housing after demonstrating readiness or service plan compliance (Fowler et al. 2019). The National Alliance to End Homelessness emphasizes that the Housing First model, often accused of being “one-size-fits-all”, in fact recognizes that one size does not fit all. A Housing First model provides flexibility in that it does not prescribe services, but rather provides the one service that every person who is homeless needs (National Alliance to End Homelessness, 2019). Consequently, Housing First actually goes beyond the shelter question of homelessness.

Fowler et al. (2019) describes a complex systems perspective of Housing First and homeless services, with a top layer of homeless and residential services that screen for need, middle layers that move in a staircase model from shelters to temporary housing to permanent supportive housing, and a bottom layer that supports investigation of household instability and creation of safety nets. This complex systems view acknowledges the multiplicity of actors, including nonprofits and landlords, and the nonlinearity of their actions. Via a series of computer simulations of policy experiments for homelessness, the researchers found that homelessness declines the most when Housing First and the first prong of prevention are combined and that broad-scale prevention works best in the context of rights-based housing policies (Fowler et al., 2019).


Figure 2: Policy experiments showing the impact of housing first and prevention efforts on the number of people in homeless assistance. Graphic from Fowler et al. (2019).

Cho (2014) considers Housing First a “whole-system orientation”, not a single program, because its aims - reduction of homelessness frequency, across-the-board permanent housing, and care and support for housing maintenance and quality of life improvement - require a shared commitment. There must be engagement among “a variety of programs and services including homeless outreach, emergency shelter, permanent supportive housing, affordable housing, rapid re-housing, along with case management supports, health care, income supports, employment services, and more.” This reinvention achieves that third prong of system reformation.

Homelessness affects all of us, no matter how well-to-do we may be. While awareness has risen, research in the U.K. discovered a disproportionately small amount of action is subsequently undertaken by the public, in part because of misunderstandings about homelessness and its prevention, belief that homelessness is an individual problem, and a fatalistic sense that a single person’s actions will do basically nothing to help so large an issue (Teixeira, 2017). The work being done to organize, clarify, and analyze the data on this population is helping to change this story.



  1. Beckwith, Ebony. (2020). The keys to ending homelessness: Data and technology. Salesforce, https://www.salesforce.com/blog/data-technology-homelessness-conversation/.
  2. Broom, Douglas. (2019). How three US cities are using data to end homelessness. World Economic Forum, https://www.weforum.org/agenda/2019/03/united-states-cities-using-data-to-end-homelessness/.
  3. Castillo, C., Zamorana, M., Jaramilla, G. & Gonzalo, S. (2020). Algorithmic impact assessment of the predictive system for risk of homelessness developed for the Allegheny County. Eticas Research and Consulting, https://www.alleghenycountyanalytics.us/wp-content/uploads/2020/08/Eticas-assessment.pdf.
  4. Cho, Richard. (2014). Four clarifications about Housing First. United States Interagency Council on Homelessness, https://www.usich.gov/news/four-clarifications-about-housing-first/.
  5. Fowler, P., Hovmand, P., Marcal, K. & Das, S. (2019). Solving homelessness from a complex systems perspective: Insights for prevention responses. Annual Review of Public Health 40. 465-486. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6445694/.
  6. Kelkar, M., Frey, R., Suriya, N. & Engel, S. (2019). Addressing homelessness with data analytics: A data-driven approach to homelessness. Deloitte Insights, https://www2.deloitte.com/us/en/insights/industry/public-sector/homelessness-data.html.
  7. Mashariki, Amen. (2018). The case for a data-driven approach to tackle homelessness. Civicist, https://civichall.org/civicist/the-case-for-a-data-driven-approach-to-tackle-homelessness/.
  8. National Alliance to End Homelessness. (2019). What Housing First really means. National Alliance to End Homelessness, https://endhomelessness.org/what-housing-first-really-means/.
  9. Teixeira, Ligia. (2017). Why we need to change the way we talk about homelessness. European Journal of Homelessness 11(2). 77-91.
  10. Wachter, T., Bertrand, M., Pollack, H., Rountree, J. & Blackwell, B. (2019). Predicting and Preventing Homelessness in Los Angeles. California Policy Lab,
  11. Wu, M., Man, L., Taylor, R. & Aldridge, S. (2020). Homelessness Data England: Providing the missing link in homelessness research. ADR UK,