BETA · Background Dashboard

Homelessness Is a
Housing Problem

Alameda County HCD Model Explorer

An interactive structural-regression simulator applying Colburn & Aldern (2022) to Alameda County — with 2024 HUD Point-in-Time (PIT) data, geographic comparables, and county-calibrated confidence intervals.

Background dashboard only — not for reporting purposes. Produced by Alameda County HCD for staff education and orientation only. Data, regression coefficients, and confidence intervals are preliminary and should not be used for policy decisions, citation, or external reporting.
📖 About this dashboard & methodology

Homelessness Is a Housing Problem — Alameda County HCD Model Explorer

This interactive dashboard is produced by Alameda County Housing and Community Development (HCD) and is inspired by the structural regression framework developed by Gregg Colburn and Clayton Page Aldern in Homelessness Is a Housing Problem (University of California Press, 2022). The regressions, visualizations, and data have been recreated and updated independently by Alameda County HCD using 2024 HUD Point-in-Time (PIT) count data and current American Community Survey (ACS) housing estimates. This work is reproduced here for public education purposes with attribution and discussion, but without direct consultation or review by the original authors. Readers are encouraged to consult the book directly for the full methodology and findings.

The book's central finding, confirmed across decades of Continuum of Care (CoC)-level data, is that rent levels and rental vacancy rates explain the overwhelming majority of variation in homelessness rates between cities, while poverty, mental illness, drug use, unemployment, and climate explain virtually none of it once housing market conditions are controlled for.

The model is not a claim that social vulnerability does not matter to individuals. It is a claim that those vulnerabilities are expressed more severely where housing markets are tightest. A person with a substance use disorder is far more likely to lose their housing in San Francisco (vacancy 3%, median rent $2,700) than in Indianapolis (vacancy 8%, median rent $960), not because of any difference in their individual circumstances, but because the housing market provides no buffer. Alameda County HCD has used this framework to design and fund housing production programs, including the Scalable Housing Infill Funding Toolkit (SHIFT) missing middle initiative and a senior ADU program, oriented toward expanding supply elasticity and moderating rent growth at scale.

How to use this simulator

  • Adjust Median gross rent and Rental vacancy rate to generate a structural prediction for Alameda County
  • Move the social control sliders to highlight metros with similar poverty, unemployment, or climate profiles — their predicted rates remain unchanged by those variables
  • Toggle City ordinary least-squares (OLS) vs. County OLS to see how model calibration affects the prediction
  • Toggle 2019 vs. 2024 to compare the pre-COVID baseline scatter with current data
  • Scroll or pinch to zoom the map; hover circles for metro name and rate

Alameda County defaults

  • Median gross rent: $2,200/mo — ACS 2022–23 5-yr estimate, Oakland-Fremont metro
  • Rental vacancy: 5.5% — Bay Area vacancy has run 3–6%; 5.5% reflects the 2023–24 period as pandemic-era tightness eased slightly
  • County OLS calibration — recommended for county CoCs; city model systematically overpredicts counties (see diagnostic panel)
  • 2024 PIT actual: 9,450 — Alameda County CoC (CA-502), January 25, 2024; Oakland: 5,485 (58%)
  • At defaults, County OLS predicts ~9,490 — within 40 people of actual, and well within the 90% confidence interval

Core model

Calibration:
ln(HRi) = α + β₁·ln(Renti) + β₂·Vacancyi + εi
3.33
intercept α
+1.45
β₁ log rent ✓
−0.08
β₂ vacancy ✓
0.23
R² counties
0.115
SE mean (log)
Calibrated on 35 county CoCs · Colburn & Aldern (2022) adapted
Why the actual PIT count falls outside the City OLS confidence interval — and how the County OLS fixes it
1
The 90% CI shown is for the conditional mean, not an individual observation

The range ±1.645 × SEmean answers "where would the average CoC with these housing conditions land?" — not where any single CoC falls. A full 90% prediction interval (±1.645 × RMSE) is 4–7× wider and encompasses individual CoC scatter. The CI is the right frame for a structural policy argument; the PI is the right frame for asking whether Alameda's count is structurally explained.

2
City-calibrated model systematically overpredicts county CoCs

The city model's intercept (α=3.39) and rent coefficient (β₁=1.76) were estimated on 50 city-level CoCs. Cities concentrate services and visible street homelessness, producing structurally higher per-capita rates than county-level CoCs for the same rent level. Applying city coefficients to a county CoC introduces upward bias of roughly 20–25%.

3
Fix: County OLS calibration

County-level regression (α=3.33, β₁=1.45, β₂=−0.08) gives a point estimate of ~9,490 at Alameda's default inputs — within 40 people of the actual 9,450. The 90% CI spans ~7,860–11,450, containing the actual count. Toggle "County OLS" above to apply this calibration throughout.

At $2,200 rent / 5.5% vacancy: City OLS predicts ~11,540 (CI 10,140–13,150); actual 9,450 is below range. County OLS predicts ~9,490 (CI 7,860–11,450); actual 9,450 is squarely within range.

Structural variables SIGNIFICANT

Median gross rent $2,200
$700$3,200/mo
Rental vacancy rate 5.5%
1% tight14% slack

Model output — Alameda County

i
About the 90% confidence interval The range shown is a 90% CI for the conditional mean — where the average CoC with these housing conditions would land. Computed as point estimate ± 1.645 × SEmean in log space, where SEmean = RMSE ÷ √n. A full prediction interval capturing individual-CoC scatter would be 4–7× wider and is more appropriate for asking whether a single CoC's count is structurally explained.
↗ Colburn & Aldern (2022) ↗ Penn State STAT 501: CI vs. PI in regression
Predicted rate
Model-predicted population
PIT count — actual

📊 Rent vs. homelessness scatter charts

Rent vs. homelessness rate — PIT / ACS

◉ Blue ring = Alameda County (home)  ·  Filled blue dot = nearest structural comparable  ·  Light blue ring = nearest social-variable match  ·  ★ Gold ring = highlighted CA jurisdiction (selector below)  ·  Hover for full statistics.
Median gross rent — 2024 ACS est.
Cities   R² ≈ 0.57 · 2024
Counties   R² ≈ 0.23 · 2024
Highlight jurisdictions for comparison

Highlight jurisdictions for comparison

Counties
Click any to add a gold ring to that point on the scatter. Hover for full statistics.
2024 mode: Cross-section from 2024 HUD AHAR PIT counts and ACS 2022–23 5-yr median gross rent estimates. Notable 2024 outliers — Seattle (+23% from COVID-era growth), New York (+53% driven by asylum-seeker shelter surge), Honolulu (Maui fire displacement) — reflect non-structural shocks. The book's original analysis used 2007–2019 pooled panel data; the 2019 cross-section shown is the final-year slice of that dataset.
🗺 Geographic distribution map

Geographic distribution of homelessness rates

Layer:
Circle size = CoC population. Shade = homelessness rate per 1,000 (lighter = lower, darker = higher). Scroll or pinch to zoom, drag to pan. = Alameda County.

Variable reference table
Variable Role Definition / source 2024 Alameda est. Effect (R²)
R² = bivariate OLS, per-capita PIT count (cities sample). Sources: HUD 2024 & 2019 AHAR; ACS 2022–23 & 2018–19 5-yr; BRFSS; NSDUH; NOAA; Saiz (2010); Colburn & Aldern (2022). 2019 scatter uses final cross-section from the book's 2007–2019 panel dataset.