U.S. County Health Workforce Index

Published weights & robustness

Each product-line score is a weighted average of national percentile ranks; the composite is the plain mean of the four product scores. Weights are judgment-based (not fitted) and published in full:

Robustness check (published with every build): an equal-weight variant of the index reproduces the published ranking at Spearman ρ 0.9733, and 40 of the top 50 counties appear in both variants.

~50% of each product's weight sits on market scale (65+ population) plus incumbent provider count because the verified buyer personas (BD teams, executives, acquirer diligence) scale with incumbents — each provider is a prospective buyer. Material weight on BOTH count and whitespace deliberately implements the U-shape: competitive metros need competitive intel while large-senior thin-supply counties attract entrants. Income carries only 10% because Medicare funds most home-health and hospice care; QCEW wages stay at or below 15% because county-industry cells are frequently suppressed. Weights are judgment, not fitted, and will be revisited against real conversion data post-launch; the sensitivity block publishes how much the ranking depends on them.

Methodology & data vintages

Components are converted to national percentile ranks (0–100, average-rank ties) computed over counties with published values only. 65+ growth uses ACS 5-year 2013-2017 vs 2019-2023 (non-overlapping vintages, 6-year midpoint gap), reported as the 6-year change (exponent 1/6).

Missing data: Missing or suppressed component values are imputed to the 50th percentile for scoring and flagged in completeness.imputed; any rendered rank or score for a county with imputed components must carry the qualifier 'scored on X of Y components'. Provider counts of zero and the absence of any HPSA designation are real observations, not imputations. Raw inputs that are unavailable publish as null and must render as 'not available' — never as a fabricated 0.

Input vintages (verbatim from the build)

• •ACS 5-year 2013-2017 — Census ACS 5-year 2013-2017 legacy sequence files (B01001, sequence 0002, per-state All_Geographies_Not_Tracts_Block_Groups) (cached)
• •ACS 5-year 2019-2023 — Census ACS 5-year 2019-2023 table-based summary file (B01001 sex-by-age bands summed for 65+, B19013 median household income) — the identical published estimates the Census API serves; pulled from www2.census.gov because the API requires CENSUS_API_KEY (cached)
• •Downloaded 2026-07-23 (CMS publishes on its stated refresh cycle) — CMS Provider Data Catalog 6jpm-sxkc (home health agencies) (cached)
• •Downloaded 2026-07-23 (CMS publishes on its stated refresh cycle) — CMS Provider Data Catalog yc9t-dgbk (hospice general information) (cached)
• •Downloaded 2026-07-23 (CMS publishes on its stated refresh cycle) — CMS Provider Data Catalog 4pq5-n9py (nursing home provider information incl. certified beds and average residents per day) (cached)
• •2023 Gazetteer — Census 2023 Gazetteer county file (names + USPS state) (cached)
• •Downloaded 2026-07-23 (HRSA refreshes nightly) — HRSA BCD_HPSA_FCT_DET_{PC,DH,MH}.csv, deduped by (hpsa_id, county_fips); Designated + Proposed-for-Withdrawal rows (cached)
• •QCEW 2024 annual averages — BLS QCEW annual averages, NAICS 6216 home health care services, county, private ownership, avg annual pay (cached)
• •QCEW 2024 annual averages — BLS QCEW annual averages, NAICS 623 nursing and residential care facilities, county, private ownership, avg annual pay (cached)
• •HUD-USPS crosswalk unavailable in this build (no HUD_USPS_API_TOKEN); ZIPs approximated by their 2020 ZCTA and assigned the county with the largest overlapping land area (Census 2020 ZCTA-to-county relationship file). — census-zcta520-county20 (cached)

Index version cdi-1.0.0, generated 2026-07-23. Every published score is reproducible from the published integer component percentiles and the weights above.