What drives the rate?
Reading this carefully matters. These are correlations, not causes. Two huge caveats: (1) Funding per capita correlates POSITIVELY with the rate because HUD sends more money to places with worse problems, not because money creates homelessness. (2) High-cost coastal areas are both rich AND high-rate, which suppresses the income/poverty signal. The strongest signals here are about HOUSING MARKET CONDITIONS (rent burden, renter share), not jobs or income.
All correlations, ranked
Spearman correlation is more robust to outliers and skew (NYC etc.). Pearson included for reference. Sample restricted to "Higher confidence" + "Possible undercount" CoCs from the data quality audit (excludes the 29 likely-undercount CoCs).
Top 6 drivers - scatter plots
Each dot is one CoC. Color by category: blue = Major City, green = Other Urban, orange = Suburban, red = Rural.
What this tells us
Housing markets dominate. The strongest correlate is renter share - places where more people rent (rather than own) have systematically higher rates. Rent burden (% of income going to rent) and median rent burden are next. Vacancy rate negatively correlates: tighter rental markets = more homelessness. This is consistent with Colburn & Aldern's "Homelessness Is a Housing Problem" (2022).
Unemployment is essentially uncorrelated (Spearman -0.01). This sounds counterintuitive but is a robust finding in the literature. Many people experiencing homelessness work; many unemployed people are not homeless. Joblessness alone is rarely the proximate cause.
Funding signal is reverse-causal. CoCs with more federal funding per capita have HIGHER rates because HUD's CoC Program allocates by need - high-need areas get more money. The cleaner signal is funding per homeless person, which is mildly NEGATIVE: better-funded systems may prevent some homelessness, but the effect is small.
What we couldn't measure here: climate (would need NOAA station-to-CoC mapping), crime (FBI UCR is city-level not CoC-level), state right-to-shelter laws (only NYC, MA, DC fully have it - too small for correlation), jail/hospital exits, predatory lending. These need separate datasets and separate analyses.
Methodology
Target: PIT homelessness rate per 10,000 (UCSF BHHI, 2019). Covariates from UCSF BHHI ACS-derived data (2018 estimates). Funding from HUD CoC Program Awards (2024). HIC beds from HUD HIC (2024). Data quality tier excluded "Likely undercounting" CoCs (n=361 vs 384 total). Eviction data was too sparse to use (n=2 - Eviction Lab coverage issue for our year).