Pathways into Homelessness

• In Event: Preventing Future and Subsequent Homelessness

Abstract

We investigate the determinants of homelessness by documenting complete life trajectories leading to a first spell. We use a panel of linked administrative records from health care, criminal justice, education, and social assistance for the full population of British Columbia, Canada. Future homeless individuals face severe disadvantage beginning in childhood: over half had contact with child welfare services, and fewer than a quarter graduated from high school. Criminal charges, emergency room visits, substance use treatment, and other acute events rise sharply in the months before entry. Conditioning on rich individual histories, both childhood disadvantage and acute adult shocks strongly predict homelessness entry out-of-sample. We identify a highest-risk group that is 100 times more likely to become homeless than the general population. These findings offer the most comprehensive picture to date of the risk factors for homelessness, and show that entry is predictable from data collected at points of contact with government services. This predictability makes targeted early intervention potentially feasible and cost-effective. Lastly, we document novel time trends in homelessness, such as rising rural homelessness and longer homelessness spells, and test their associations with rising housing costs.

Our main contribution is to identify risk factors of homelessness with greater detail, frequency, and comprehensiveness than any prior study. The Australian Journey Home project (Cobb-Clark, Herault, Scutella, and Tseng, 2016; McVicar, Moschion, and van Ours, 2015; Moschion and van Ours, 2019; Scutella, Johnson, Moschion, Tseng, and Wooden, 2012; Wooden, Bevitt, Chigavazira, Greer, Johnson, Killackey, Moschion, Scutella, Tseng, and Watson, 2012, among others) produced pioneering work on trajectories into homelessness. But the Journey Home studies rely on self-reported surveys from approximately 1,700 individuals drawn from welfare agency files. Our administrative data cover the full population of British Columbia and avoid the recall and reporting concerns inherent in self-reported survey data. Another strand of prominent recent work links US Census and American Community Survey data to tax records, food benefits, housing assistance, and Medicare/Medicaid coverage (Meyer et al., 2021, Meyer, Wyse, and Logani, 2023, Meyer, Wyse, and Corinth, 2023, Meyer, Wyse, Meyer, Grunwaldt, and Wu, 2024, Meyer, Wyse, and Williams, 2025, Meyer, Wyse, Meyer, Grunwaldt, and Wu, 2025). Our data differ from theirs in three important dimensions. First, we observe flows into homelessness rather than a stock at a single point in time. This lets us study first entries, timing events precisely relative to onset. Second, we link individuals to childhood records, revealing disadvantages that originate decades before housing loss. Third, our high-frequency administrative data across health, criminal justice, and social assistance reveal severe adverse events accumulating rapidly in the months before entry, a pattern difficult to detect without high-frequency data around first entry. Our work also advances the pioneering predictive models of the California Policy Lab (e.g., von Wachter, Bertrand, Pollack, Rountree, and Blackwell, 2019, Blackwell, Caprara, Rountree, Santillano, Vanderford, and Battis, 2024) and Allegheny County (Castillo, Zamorano, Jaramillo, and Gonzalo, 2020), which demonstrate that prediction of homelessness from administrative records is feasible and actionable.

Author

• Jeffrey Hicks, University of Toronto
  Presenting Author