Datasets and spatial tools for United States homelessness research. COCHomeless
packages HUD Continuum of Care (CoC) Point-in-Time (PIT) homeless counts, CoC
boundaries, a Census-to-CoC crosswalk, and county-level homeless estimates,
covering 2007–2025.
# install.packages("remotes")
remotes::install_github("SSDALab/COCHomeless", build_vignettes = TRUE)
library(COCHomeless)sf is required (spatial objects). tidycensus, ggplot2, tigris,
imputeTS and a few others are only needed to rebuild data or run the vignettes.
All series run 2007 through 2025 (annual). The 2025 PIT counts are HUD's
official release; hud2025_prelim keeps the earlier Community Solutions estimate
for reference. In 2021 HUD waived the unsheltered count for many CoCs during
COVID-19, so 2021 understates homelessness — the vignettes show an ARIMA
reconstruction.
1. CoC level (HUD administrative areas)
| Object | What |
|---|---|
hud2007 … hud2025 |
PIT "Overall Homeless" count per CoC, by year |
coc2007 … coc2025 |
CoC boundary polygons (national sf), by year |
coc_mergers |
CoC merger crosswalk over time (code lineage) |
pit_us |
national totals split into sheltered / unsheltered |
pit_coc |
per-CoC sheltered / unsheltered / total, 2007–2025 |
pit_coc_detail |
per-CoC by shelter type × subpopulation (veterans, chronic, youth, families, gender, race, age) |
hud2025_prelim |
preliminary 2025 estimate (superseded by hud2025) |
CoC boundaries (coc20XX) also carry that year's total/sheltered/unsheltered.
2. Census ↔ CoC crosswalk (links Census tracts/counties to CoCs)
| Object | What |
|---|---|
tract_coc2007 … tract_coc2025 |
tract → CoC hard assignment, by year |
tract_coc_wt2007 … tract_coc_wt2025 |
tract ↔ CoC area weights, by year |
county_coc2007 … county_coc2025 |
county ↔ CoC weights (both directions), by year |
Tract geography uses 2010-census tracts for 2007–2019 and 2020-census tracts for 2020–2025.
3. County-level estimates (CoC counts disaggregated to counties)
| Object | What |
|---|---|
homeless |
county estimates count07…count25 + ACS covariates (imputed) |
homeless_na |
same, before imputation (missing counties as NA) |
sp_homeless |
homeless as sf, also carrying sheltered<yy>/unsheltered<yy> |
county_pit |
county sheltered / unsheltered / total, 2007–2025 (long) |
county_pit_detail |
county by shelter type × subpopulation (gender, race, age, …) |
counties, states |
full-US sf base layers |
area |
county land area (km²) |
After installing with build_vignettes = TRUE:
browseVignettes("COCHomeless")- CoC Point-in-Time homeless counts (
coc-pit-counts) — national & state maps, per-CoC trends, sheltered/unsheltered series with ARIMA reconstruction of 2021. - County-level homeless estimates (
county-estimates) — county-vs-CoC comparison, per-capita maps, US/CA/WA trends, a 2025 summary grid. - Census-to-CoC crosswalk (
census-coc-crosswalk) — using the crosswalk andcoc_mergers.
Vignette sources are in vignettes/; build scripts that regenerate
every dataset from source are in data-raw/.
County estimates disaggregate CoC PIT counts by population density and impute counties with no CoC, following Almquist, Helwig & You (2020). Please cite:
Almquist, Z. W., Helwig, N. E., & You, Y. (2020). Connecting Continuum of Care point-in-time homeless counts to United States Census areal units. Mathematical Population Studies, 27(1), 46–58. https://doi.org/10.1080/08898480.2019.1636574
The Census-to-CoC crosswalk follows Tom Byrne's
HUD-CoC-Geography-Crosswalk;
please credit it when using the tract_coc* / county_coc* data. Run
citation("COCHomeless") for full details.
ACS pulls need a Census API key in ~/.Renviron (CENSUS_API_KEY); never commit it.