Data & downloads: every number on this site

How to check our work

Each data page names the source for every record it shows. If you want to go further, download the file below and search for the row. If a number on this site is not in one of these files, it is a bug — please tell us.

The current data was collected on 2026-09-21. The full record of what was searched, found and deliberately excluded is in RESEARCH_LOG.md, and every source URL opened is listed in sources.md.

Don't want to download anything?

Search the data lets you find a column by name across every panel below, see how populated it actually is, and filter rows on the server — plus a JSON endpoint at /data/query.json for scripts.

Download the data

43 files

trials.csv Registered Alzheimer's clinical trials (ClinicalTrials.gov API v2 pull) 3.8 MB · 7816 rows

mortality_us.csv US Alzheimer's and dementia deaths by year, sex, age, race (CDC WONDER) 1.6 MB · 6474 rows

biomarkers.csv Diagnostic-accuracy table for blood, CSF and imaging biomarkers 81 KB · 95 rows

genetics_gwas.csv Genome-wide association study loci for Alzheimer's disease 373 KB · 807 rows

panel_state.csv Analysis-ready panel: one row per US state 19 KB · 51 rows

panel_county.csv Analysis-ready panel: one row per US county 2.5 MB · 3143 rows

panel_country.csv Analysis-ready panel: one row per country 62 KB · 217 rows

timeseries_national.csv US national time series (prevalence, deaths, costs, funding) 842 B · 13 rows

timeseries_country.csv Country-year long-format time series 2.1 MB · 65946 rows

panel_state_meta.json Column dictionary for panel_state.csv 38 KB · 67 rows

alz_cognitive_state.json CDC BRFSS cognitive-decline and caregiving measures by state, 2015-2022 (collected from data.cdc.gov) 419 KB · 16 rows

owid_dementia_timeseries.csv Dementia deaths and age-standardised death rate per 100,000 by country and year, 2000-2021, 143 countries (IHME Global Burden of Disease via Our World in Data). Modelled estimates, not counted deaths. 94 KB · 3146 rows

correlation_studies.json Catalogue of correlation studies: exposure, outcome, feasibility and confounder for each 243 KB · 7 rows

correlation_results.json Computed correlations for every study the data currently supports 52 KB · 3 rows

correlation_report.json Full correlation analysis: 257 coefficients across counties, states, countries, studies, genetics and drugs, with baseline, partial-control and sign-stability checks for each (JSON) 887 KB · 12 rows

panel_county_meta.json Column dictionary for panel_county.csv 150 KB · 144 rows

panel_country_meta.json Column dictionary for panel_country.csv 24 KB · 49 rows

biomarkers.json Biomarker accuracy records (JSON) 126 KB · 95 rows

genetics.json Genetics records (JSON) 167 KB · 135 rows

prevention_trials.json Prevention trial records (JSON) 205 KB · 109 rows

caregiving_interventions.json Caregiver intervention trial records (JSON) 100 KB · 70 rows

caregiving_burden.json Caregiver burden statistics (JSON) 44 KB · 74 rows

imaging.json Imaging and diagnostic test records (JSON) 94 KB · 41 rows

economics.json Cost and cost-effectiveness records (JSON) 153 KB · 211 rows

policy.json Policy, legislation and funding records (JSON) 362 KB · 302 rows

care_settings.json Care-setting and workforce records (JSON) 1.1 MB · 1920 rows

digital_health.json Digital health tool records (JSON) 87 KB · 46 rows

disparities.json Disparities records (JSON) 75 KB · 75 rows

treatments.json Treatment records (JSON) 67 KB · 23 rows

risk_factors.json Modifiable risk factor records (JSON) 41 KB · 31 rows

datasets.json Catalogue of public source datasets (JSON) 16 KB · 18 rows

research.json Research summaries (JSON) 90.2 MB · 27380 rows

research_score.json Every study's score, component by component, with the wording behind each one (JSON) 124.8 MB · 7 rows

library_gaps.json What this library does and doesn't cover: source mix, identifier/author/year coverage, peer-review and preprint counts (JSON) 26 KB · 19 rows

score_charts.json The figures on /research/how-we-score, as data plus the SVG drawn from it (JSON) 143 KB · 19 rows

risk_charts.json The figures on /risk-factors and /life-course, as data plus the SVG drawn from it (JSON) 35 KB · 2 rows

news.json News items (JSON) 333 KB · 347 rows

stats.json Key facts shown on the home page (JSON) 20 KB · 48 rows

sources.md Index of every source URL the research pipeline opened 9 KB

panel_state_timeseries.csv State-year panel, 1999-2024: Alzheimer's deaths and age-adjusted death rate by US state (CDC WONDER / NCHS), the file behind /trends 96 KB · 1326 rows

panel_country_timeseries.csv Country-year panel: modelled dementia deaths and rates by country (IHME Global Burden of Disease via Our World in Data), the file behind /trends 164 KB · 3146 rows

county_dementia_outcomes.csv County-year dementia prevalence, incidence and mortality counts among Medicare beneficiaries, with suppression flags, the file behind /county 185 KB · 3207 rows

RESEARCH_LOG.md Research log: what was searched, found, and excluded 15 KB

US national time series

13 years

One row per year. Blank cells mean no sourced value for that year — we do not interpolate or carry a value forward to fill a gap.

timeseries_national.csv — prevalence, deaths, costs, caregivers and research funding by year.
year	alz prevalence 65plus pct	alz count 65plus	alz deaths total	alz deaths per 100k	total dementia care cost usd	medicare medicaid combined spend usd	medicaid spend total usd	unpaid caregivers count	unpaid caregiver hours millions	unpaid caregiver cost usd billions	research funding nih adrd usd	fda approved alz drugs cumulative	us population 65plus
1993	—	—	—	—	—	—	—	—	—	—	—	1	—
1996	—	—	—	—	—	—	—	—	—	—	—	2	—
2000	—	—	—	—	—	—	—	—	—	—	—	3	—
2001	—	—	—	—	—	—	—	—	—	—	—	4	—
2003	—	—	—	—	—	—	—	—	—	—	—	5	—
2010	—	—	—	—	—	—	—	—	—	—	—	5	40,300,000
2019	—	—	—	—	—	—	—	—	—	—	2,398,000,000	5	—
2020	—	—	—	—	—	—	—	—	—	—	2,869,000,000	5	55,800,000
2021	—	—	—	—	—	—	—	—	16,000	271.6	3,251,000,000	6	—
2022	—	6,500,000	120,122	36	—	—	—	—	18,000	339.5	3,514,000,000	6	—
2023	10.8	6,700,000	—	—	345,000,000,000	222,000,000,000	—	11,475,000	18,400	346.6	3,761,000,000	7	—
2024	—	6,900,000	—	—	360,000,000,000	—	—	—	—	—	3,859,000,000	8	—
2025	—	7,200,000	—	—	384,000,000,000	—	72,000,000,000	11,926,000	19,161	413.5	3,882,000,000	8	—

This table is wider than a phone screen — swipe it sideways to see the rest.

US Alzheimer's deaths by year

27 years

Death certificates, counted — not modelled. The age-adjusted rate is the one to compare across years, because it removes the effect of the population simply getting older.

National Alzheimer's and dementia deaths. Measured Source: CDC WONDER D76 Underlying Cause of Death 1999-2020 (API)
Year	Alzheimer's deaths	All dementia deaths	Crude rate per 100k	Age-adjusted rate per 100k
1999	44,536	74,386	15.96	16.47
2000	49,558	83,694	17.61	18.05
2001	53,852	92,514	18.9	19.29
2002	58,866	102,105	20.47	20.81
2003	63,457	110,569	21.87	22.06
2004	65,965	114,271	22.53	22.6
2005	71,599	129,573	24.23	23.99
2006	72,432	151,432	24.28	23.69
2007	74,632	156,042	24.78	23.84
2008	82,435	178,960	27.11	25.79
2009	79,003	177,523	25.75	24.17
2010	83,494	196,371	27.04	25.11
2011	84,974	212,876	27.27	24.73
2012	83,637	223,404	26.64	23.78
2013	84,767	234,242	26.81	23.52
2014	93,541	239,753	29.34	25.44
2015	110,561	245,926	34.4	29.43
2016	116,103	249,605	35.93	30.29
2017	121,404	261,914	37.27	31.04
2018	122,019	266,957	37.3	30.52
2019	121,499	271,872	37.02	29.85
2020	134,242	303,984	40.74	32.45
2021	119,399	279,704	35.98	30.96
2022	120,122	292,881	36.04	28.9
2023	114,034	286,865	34.05	27.75
2024	116,022	299,192	34.11	27.09
2025	116,854	—	34.36	27.26

This table is wider than a phone screen — swipe it sideways to see the rest.

Analysis-ready panels

3 panels

One row per state, county or country, with as many sourced columns as we could fill. We list the columns we could not fill as well, because a column of blanks in someone else's dataset is the kind of thing that quietly wrecks an analysis.

State panel

67 columns · 51 rows · 19 KB · 4 modelled · 3 not populated

Download panel_state.csv Column dictionary (JSON)

panel_state.csv — what each column holds and where it came from.
Column	What it is	Unit / year	Basis	Source
`state`	USPS 2-letter state abbreviation (50 states + DC) Identifier column, not a data value.	text	Not populated	N/A (identifier)
`state_name`	Full state name Identifier column, not a data value.	text	Not populated	N/A (identifier)
`alz_prevalence_65plus_pct`	Estimated percentage of the population age 65+ living with Alzheimer's dementia Modeled estimate based on 2020 age/sex/race population distributions, not a direct case count or registry; published in the 2025 F&F report.	percent 2020	Modelled estimate	Alzheimer's Association 2025 Alzheimer's Disease Facts and Figures, Table 5 (data from Dhana et al., state/county prevalence model)
`alz_count_65plus`	Estimated number of people age 65+ living with Alzheimer's dementia Report gives this in thousands (1 decimal); value here is that figure x1000, rounded to nearest integer. Same modeled-estimate caveat as alz_prevalence_65plus_pct.	count 2020	Modelled estimate	Alzheimer's Association 2025 Alzheimer's Disease Facts and Figures, Table 5
`alz_deaths_total`	Number of deaths with Alzheimer's disease as the underlying cause of death, by state 2022 is the most recent year with state-level mortality data available as of the 2025 report. US total for this column is 120,122 (rate 36.0/100k).	count 2022	Measured	Alzheimer's Association 2025 Facts and Figures, Table 6, citing National Center for Health Statistics death certificate data
`alz_deaths_per_100k`	Unadjusted (not age-adjusted) annual Alzheimer's death rate per 100,000 population Report explicitly presents UNADJUSTED rates (not age-adjusted) so that larger/older-population states aren't penalized in the comparison; do not treat as age-standardized.	rate per 100k 2022	Measured	Alzheimer's Association 2025 Facts and Figures, Table 6 (NCHS)
`alz_prevalence_rank`	State rank by alz_prevalence_65plus_pct, 1 = highest prevalence percentage Computed rank, not a value published in the source report. Ties broken by original table order.	rank (1=highest) 2020	Measured	Derived by this project from Table 5 percentages (not published directly by Alzheimer's Association)
`pop_total`	Total resident population July 1, 2024 estimate.	count 2024	Measured	U.S. Census Bureau, Population Estimates Program, Vintage 2024
`pop_65plus`	Population age 65 and older Not a single published Census count; combines 2024 population total with 2023 ACS age-share, so treat as an approximation, not an official Census 65+ count.	count 2023	Measured	Derived: 2024 Census population estimate x 2023 ACS pct_65plus share
`pct_65plus`	Percent of population age 65+	percent 2023	Measured	U.S. Census Bureau, ACS 2023 1-Year Estimates, Geographic Comparison Table GCT0103
`pct_85plus`	Percent of population age 85+	percent 2023	Measured	U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT0104
`pct_female`	Percent of population female Used because Census's own ranking GCT series does not publish a female-share table directly.	percent 2024	Measured	KFF State Health Facts, Distribution by Gender (based on ACS 1-Year Estimates)
`pct_race_black`	Percent of population Black or African American alone	percent 2023	Measured	U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT0202
`pct_race_hispanic`	Percent of population Hispanic or Latino (any race)	percent 2024	Measured	KFF State Health Facts, Distribution by Race/Ethnicity (ACS 1-Year Estimates)
`median_age`	Median age of population	years 2023	Measured	U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT0101

Show all 67 columns (15 shown so far)

This table is wider than a phone screen — swipe it sideways to see the rest.

County panel

143 columns · 3143 rows · 2.5 MB · 30 modelled · 8 not populated

Download panel_county.csv Column dictionary (JSON)

panel_county.csv — what each column holds and where it came from.
Column	What it is	Unit / year	Basis	Source
`fips`	5-digit county FIPS code (2-digit state + 3-digit county), zero-padded string Join key for all rows. Sourced from the CHR file's '5-digit FIPS Code' column; spot-checked against known FIPS for LA County (06037), Manhattan/New York County (36061), Harris County TX (48201).	code 2024 geography vintage	Measured	US Census Bureau / County Health Rankings
`county`	County (or county-equivalent) name	text 2024	Measured	County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release
`state`	2-letter USPS state abbreviation	text 2024	Measured	County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release
`state_name`	Full state name	text 2024	Measured	Derived (static FIPS/USPS state name lookup)
`county_type_metro`	1 = metro county, 0 = nonmetro county Derived from USDA Rural-Urban Continuum Code (RUCC) 2023: codes 1-3 = metro -> 1; codes 4-9 = nonmetro -> 0. Blank for 8 counties where RUCC 2023 has no matching FIPS (Connecticut legacy county codes, retired -- see README).	binary (1/0) 2023	Measured	USDA Economic Research Service, Rural-Urban Continuum Codes 2023
`population`	Total county population Two different source vintages are blended: PLACES population (Census 2023 vintage) where available, else CHR population estimate (vintage varies, generally 2021-2022). Not itself a PLACES model estimate -- it is CDC's pass-through of the Census county population estimate.	count (persons) 2023 (PLACES) / see caveat	Modelled estimate	CDC PLACES 2025 release TotalPopulation field (Census Bureau 2023 county population estimate), with County Health Rankings 2024 'Population raw value' used as fallback for the handful of counties absent from PLACES
`alz_prevalence_pct_65plus`	Alzheimer's/dementia diagnosed prevalence, % of population 65+ This column is intentionally empty for every row -- NOT SOURCED. Re-investigated 2026-09-27 (see data/county_outcomes_README.md for the full trail): CMS's Mapping Medicare Disparities (MMD) Tool still only exposes this through an interactive web map with per-query CSV export, confirmed to be a client-rendered SPA with no discoverable bulk API; CMS's older county-level Chronic Conditions Excel files (2007-2018, which included Alzheimer's) have since been retired by CMS in favor of the MMD Tool. CDC's Alzheimer's Disease and Healthy Aging Data Portal is confirmed state/national-only (fetched directly). A real peer-reviewed county estimate exists (Dhana et al. 2023, Alzheimer's & Dementia 19:4388-4395), but the per-county numbers are behind a paywall with no public data table found -- using the map image or press-release examples to fill other counties would be fabrication, so it was not used. Do not fill this column with a proxy (e.g. PLACES COGNITION, which measures self-reported cognitive difficulty in the general adult population, not diagnosed Alzheimer's/dementia in the 65+ population) without clearly relabeling it.	percent (0-100)	Not populated	NOT SOURCED
`alz_count_estimate`	Estimated count of county residents with Alzheimer's/dementia This column is intentionally empty for every row -- NOT SOURCED. Re-investigated 2026-09-27 (see data/county_outcomes_README.md for the full trail): CMS's Mapping Medicare Disparities (MMD) Tool still only exposes this through an interactive web map with per-query CSV export, confirmed to be a client-rendered SPA with no discoverable bulk API; CMS's older county-level Chronic Conditions Excel files (2007-2018, which included Alzheimer's) have since been retired by CMS in favor of the MMD Tool. CDC's Alzheimer's Disease and Healthy Aging Data Portal is confirmed state/national-only (fetched directly). A real peer-reviewed county estimate exists (Dhana et al. 2023, Alzheimer's & Dementia 19:4388-4395), but the per-county numbers are behind a paywall with no public data table found -- using the map image or press-release examples to fill other counties would be fabrication, so it was not used. Do not fill this column with a proxy (e.g. PLACES COGNITION, which measures self-reported cognitive difficulty in the general adult population, not diagnosed Alzheimer's/dementia in the 65+ population) without clearly relabeling it.	count (persons)	Not populated	NOT SOURCED
`dementia_mortality_per_100k`	Age-adjusted death rate, all dementia (ICD-10 F01,F03,G30), per 100,000 population This column is intentionally empty for every row -- NOT SOURCED. Re-confirmed 2026-09-27 (see data/county_outcomes_README.md), for two independent reasons: (1) CDC WONDER's own API documentation (wonder.cdc.gov/wonder/help/wonder-api.html) states outright that 'only national data are available for query by the API. Queries for mortality and births statistics from the National Vital Statistics System cannot limit or group results by any location field, such as Region, Division, State or County' -- this is a stated policy, not a fixable parameter mistake. (2) Independently, a live GET and a live hand-built XML POST to https://wonder.cdc.gov/controller/datarequest/D76 today both returned HTTP 403 Akamai 'Access Denied' pages, so even the interactive UI is unreachable by automation. The interactive WONDER web UI does expose county-level mortality (subject to CDC's own <10-death suppression), but only to a human in a browser. A future job with real-browser access to wonder.cdc.gov's UI (or a CDC data-use agreement / NCHS Research Data Center application for a restricted-use county mortality file) could source this manually per county.	rate per 100,000	Not populated	NOT SOURCED
`dementia_deaths_total`	Total deaths with underlying cause = dementia This column is intentionally empty for every row -- NOT SOURCED. Re-confirmed 2026-09-27 (see data/county_outcomes_README.md), for two independent reasons: (1) CDC WONDER's own API documentation (wonder.cdc.gov/wonder/help/wonder-api.html) states outright that 'only national data are available for query by the API. Queries for mortality and births statistics from the National Vital Statistics System cannot limit or group results by any location field, such as Region, Division, State or County' -- this is a stated policy, not a fixable parameter mistake. (2) Independently, a live GET and a live hand-built XML POST to https://wonder.cdc.gov/controller/datarequest/D76 today both returned HTTP 403 Akamai 'Access Denied' pages, so even the interactive UI is unreachable by automation. The interactive WONDER web UI does expose county-level mortality (subject to CDC's own <10-death suppression), but only to a human in a browser. A future job with real-browser access to wonder.cdc.gov's UI (or a CDC data-use agreement / NCHS Research Data Center application for a restricted-use county mortality file) could source this manually per county.	count (deaths)	Not populated	NOT SOURCED
`alz_deaths_total`	Total deaths with underlying cause = Alzheimer's disease (ICD-10 G30) This column is intentionally empty for every row -- NOT SOURCED. Re-confirmed 2026-09-27 (see data/county_outcomes_README.md), for two independent reasons: (1) CDC WONDER's own API documentation (wonder.cdc.gov/wonder/help/wonder-api.html) states outright that 'only national data are available for query by the API. Queries for mortality and births statistics from the National Vital Statistics System cannot limit or group results by any location field, such as Region, Division, State or County' -- this is a stated policy, not a fixable parameter mistake. (2) Independently, a live GET and a live hand-built XML POST to https://wonder.cdc.gov/controller/datarequest/D76 today both returned HTTP 403 Akamai 'Access Denied' pages, so even the interactive UI is unreachable by automation. The interactive WONDER web UI does expose county-level mortality (subject to CDC's own <10-death suppression), but only to a human in a browser. A future job with real-browser access to wonder.cdc.gov's UI (or a CDC data-use agreement / NCHS Research Data Center application for a restricted-use county mortality file) could source this manually per county.	count (deaths)	Not populated	NOT SOURCED
`pct_65plus`	Percent of population age 65 and older CHR pass-through of ACS 5-year data; not independently re-pulled from ACS 2020-2024 for consistency with other CHR-sourced demographic shares (pct_female fallback) in this file.	percent (0-100) ~2018-2022 (ACS vintage embedded in CHR 2024)	Measured	County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release
`pct_85plus`	Percent of population age 85 and older ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. Computed as (male 85+ + female 85+) / total population from table B01001.	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_female`	Percent of population that is female ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B01001; CHR '% Female raw value' used as fallback for the 8 Connecticut counties absent from the ACS pull.	percent (0-100) 2020-2024 (ACS) or CHR vintage for the 8 CT counties missing ACS	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_white_nh`	Percent of population that is White alone, Not Hispanic or Latino ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B03002.	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_black`	Percent of population that is Black or African American alone, Not Hispanic or Latino ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B03002. Excludes Hispanic Black residents (counted under pct_hispanic instead) -- this is the standard non-Hispanic race breakdown, not 'Black alone regardless of ethnicity.'	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_hispanic`	Percent of population that is Hispanic or Latino (any race) ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B03002.	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_asian`	Percent of population that is Asian alone, Not Hispanic or Latino ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B03002.	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`median_age`	Median age of county population ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B01002.	years 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`population_density_per_sqmi`	Population per square mile of land area	persons per square mile 2023 (population) / 2024 (land area)	Measured	Derived: population (see 'population' column) divided by ALAND_SQMI from the 2024 Census Gazetteer file
`median_household_income`	Median household income ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B19013; CHR 'Median Household Income raw value' used as fallback for the 8 Connecticut counties absent from the ACS pull.	US dollars (nominal, ACS inflation-adjusted to survey-period dollars) 2020-2024 (ACS) or CHR vintage for 8 CT counties	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_below_poverty`	Percent of all people (all ages) with income in the past 12 months below the poverty level ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B17001 (total population poverty status, not just children). Differs from CHR's 'Children in Poverty' measure, which was NOT used here.	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_bachelors_plus`	Percent of population 25+ with a bachelor's degree or higher ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B15003 (bachelor's + master's + professional + doctorate, divided by population 25+).	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`unemployment_rate`	Unemployment rate (unemployed / civilian labor force) ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B23025; CHR 'Unemployment raw value' used as fallback for the 8 Connecticut counties absent from the ACS pull. Not seasonally adjusted; differs methodologically from BLS LAUS unemployment rates.	percent (0-100) 2020-2024 (ACS) or CHR vintage for 8 CT counties	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_uninsured`	Percent of civilian noninstitutionalized population with no health insurance coverage CHR pass-through of Census SAHIE data.	percent (0-100) ~2021 (Small Area Health Insurance Estimates vintage embedded in CHR 2024)	Measured	County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release
`pct_no_vehicle`	Percent of households with no vehicle available ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B08201.	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_single_person_households`	Percent of households that are a single person living alone (nonfamily householder living alone) ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B11001 (B11001_008 / B11001_001).	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_living_alone_65plus`	Percent of the 65+ population living alone ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B09021 (population 65+ 'lives alone' living-arrangement category / total population 65+ in that table's universe).	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_diabetes`	Percent of adults 18+ with diagnosed diabetes MODELLED (small-area) estimate, not a direct survey measurement. Derived from BRFSS 2023 self-reported survey responses using a validated multilevel regression and poststratification (MRP) small-area model, not a full county census. Values for counties in Kentucky, Pennsylvania, and Loving County TX are blank in this release -- see README missingness table. Interpret correlations with caution: because these are model outputs (not raw counts), they can be smoother / more spatially autocorrelated than a true measured rate, which can inflate correlations with other smoothed covariates (e.g., other PLACES measures or CHR measures that are also modelled).	percent (0-100) 2023 (BRFSS)	Modelled estimate	CDC PLACES: County Data (GIS-Friendly Format), 2025 release
`pct_obesity`	Percent of adults 18+ with obesity (BMI 30+) MODELLED (small-area) estimate, not a direct survey measurement. Derived from BRFSS 2023 self-reported survey responses using a validated multilevel regression and poststratification (MRP) small-area model, not a full county census. Values for counties in Kentucky, Pennsylvania, and Loving County TX are blank in this release -- see README missingness table. Interpret correlations with caution: because these are model outputs (not raw counts), they can be smoother / more spatially autocorrelated than a true measured rate, which can inflate correlations with other smoothed covariates (e.g., other PLACES measures or CHR measures that are also modelled).	percent (0-100) 2023 (BRFSS)	Modelled estimate	CDC PLACES: County Data (GIS-Friendly Format), 2025 release
`pct_physical_inactivity`	Percent of adults 18+ reporting no leisure-time physical activity MODELLED (small-area) estimate, not a direct survey measurement. Derived from BRFSS 2023 self-reported survey responses using a validated multilevel regression and poststratification (MRP) small-area model, not a full county census. Values for counties in Kentucky, Pennsylvania, and Loving County TX are blank in this release -- see README missingness table. Interpret correlations with caution: because these are model outputs (not raw counts), they can be smoother / more spatially autocorrelated than a true measured rate, which can inflate correlations with other smoothed covariates (e.g., other PLACES measures or CHR measures that are also modelled).	percent (0-100) 2023 (BRFSS)	Modelled estimate	CDC PLACES: County Data (GIS-Friendly Format), 2025 release
`pct_smoking`	Percent of adults 18+ who are current smokers MODELLED (small-area) estimate, not a direct survey measurement. Derived from BRFSS 2023 self-reported survey responses using a validated multilevel regression and poststratification (MRP) small-area model, not a full county census. Values for counties in Kentucky, Pennsylvania, and Loving County TX are blank in this release -- see README missingness table. Interpret correlations with caution: because these are model outputs (not raw counts), they can be smoother / more spatially autocorrelated than a true measured rate, which can inflate correlations with other smoothed covariates (e.g., other PLACES measures or CHR measures that are also modelled).	percent (0-100) 2023 (BRFSS)	Modelled estimate	CDC PLACES: County Data (GIS-Friendly Format), 2025 release
`pct_hypertension`	Percent of adults 18+ with diagnosed high blood pressure MODELLED (small-area) estimate, not a direct survey measurement. Derived from BRFSS 2023 self-reported survey responses using a validated multilevel regression and poststratification (MRP) small-area model, not a full county census. Values for counties in Kentucky, Pennsylvania, and Loving County TX are blank in this release -- see README missingness table. Interpret correlations with caution: because these are model outputs (not raw counts), they can be smoother / more spatially autocorrelated than a true measured rate, which can inflate correlations with other smoothed covariates (e.g., other PLACES measures or CHR measures that are also modelled).	percent (0-100) 2023 (BRFSS)	Modelled estimate	CDC PLACES: County Data (GIS-Friendly Format), 2025 release
`pct_depression`	Percent of adults 18+ ever told they have a depressive disorder MODELLED (small-area) estimate, not a direct survey measurement. Derived from BRFSS 2023 self-reported survey responses using a validated multilevel regression and poststratification (MRP) small-area model, not a full county census. Values for counties in Kentucky, Pennsylvania, and Loving County TX are blank in this release -- see README missingness table. Interpret correlations with caution: because these are model outputs (not raw counts), they can be smoother / more spatially autocorrelated than a true measured rate, which can inflate correlations with other smoothed covariates (e.g., other PLACES measures or CHR measures that are also modelled).	percent (0-100) 2023 (BRFSS)	Modelled estimate	CDC PLACES: County Data (GIS-Friendly Format), 2025 release
`pct_binge_drinking`	Percent of adults 18+ reporting binge drinking MODELLED (small-area) estimate, not a direct survey measurement. Derived from BRFSS 2023 self-reported survey responses using a validated multilevel regression and poststratification (MRP) small-area model, not a full county census. Values for counties in Kentucky, Pennsylvania, and Loving County TX are blank in this release -- see README missingness table. Interpret correlations with caution: because these are model outputs (not raw counts), they can be smoother / more spatially autocorrelated than a true measured rate, which can inflate correlations with other smoothed covariates (e.g., other PLACES measures or CHR measures that are also modelled).	percent (0-100) 2023 (BRFSS)	Modelled estimate	CDC PLACES: County Data (GIS-Friendly Format), 2025 release
`pct_fair_poor_health`	Percent of adults 18+ reporting fair or poor self-rated general health MODELLED (small-area) estimate, not a direct survey measurement. Derived from BRFSS 2023 self-reported survey responses using a validated multilevel regression and poststratification (MRP) small-area model, not a full county census. Values for counties in Kentucky, Pennsylvania, and Loving County TX are blank in this release -- see README missingness table. Interpret correlations with caution: because these are model outputs (not raw counts), they can be smoother / more spatially autocorrelated than a true measured rate, which can inflate correlations with other smoothed covariates (e.g., other PLACES measures or CHR measures that are also modelled).	percent (0-100) 2023 (BRFSS)	Modelled estimate	CDC PLACES: County Data (GIS-Friendly Format), 2025 release
`pct_checkup_annual`	Percent of adults 18+ who report visiting a doctor for a routine checkup within the past year MODELLED (small-area) estimate, not a direct survey measurement. Derived from BRFSS 2023 self-reported survey responses using a validated multilevel regression and poststratification (MRP) small-area model, not a full county census. Values for counties in Kentucky, Pennsylvania, and Loving County TX are blank in this release -- see README missingness table. Interpret correlations with caution: because these are model outputs (not raw counts), they can be smoother / more spatially autocorrelated than a true measured rate, which can inflate correlations with other smoothed covariates (e.g., other PLACES measures or CHR measures that are also modelled).	percent (0-100) 2023 (BRFSS)	Modelled estimate	CDC PLACES: County Data (GIS-Friendly Format), 2025 release
`primary_care_physicians_per_100k`	Primary care physicians per 100,000 population CHR pass-through of HRSA Area Health Resources Files; computed here as CHR's population-to-provider ratio raw value inverted and rescaled to per-100k.	rate per 100,000 ~2022 (Area Health Resources Files vintage embedded in CHR 2024)	Measured	County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release
`mental_health_providers_per_100k`	Mental health providers per 100,000 population CHR pass-through of CMS NPI file counts; computed here as CHR's population-to-provider ratio raw value inverted and rescaled to per-100k.	rate per 100,000 ~2023 (CMS National Provider Identification vintage embedded in CHR 2024)	Measured	County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release
`hospital_beds_per_1000`	Certified hospital beds per 1,000 population Counts only Medicare-certified short-term/critical-access hospitals with active certification (PGM_TRMNTN_CD = 00) in this CMS extract; excludes VA, psychiatric-only, and non-Medicare-certified facilities. Blank means no active Medicare-certified hospital was found in this county in the extract -- it usually means the county genuinely has none (common for small rural counties whose residents use a neighboring county's hospital), but a small share could be extract gaps.	beds per 1,000 population 2026 snapshot (facility roster) / population as dated above	Measured	CMS Provider of Services File - Quality Improvement and Evaluation System (QIES), current quarterly snapshot pulled 2026-09-21
`nursing_home_beds_per_1000`	Nursing home / skilled nursing facility beds per 1,000 population This column is intentionally empty for every row -- NOT SOURCED. The CMS Provider of Services QIES extract used for hospital_beds_per_1000 does not include skilled nursing facilities in this dataset's category codes. CMS's 'Long-Term Care Facility Characteristics' dataset covers SNFs but only has City/State/ZIP, no county FIPS, and would require a ZIP-to-county crosswalk (with the usual many-to-many ZIP/county ambiguity) that was out of scope for this job.	beds per 1,000 population	Not populated	NOT SOURCED
`pct_65plus_with_disability`	Percent of population 65+ with any disability ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B18101 (ages 65-74 and 75+, both sexes, 'with a disability' summed over 'total' for those age bands).	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`physicians_per_1000`	All physicians (any specialty) per 1,000 population This column is intentionally empty for every row -- NOT SOURCED. CHR and PLACES only publish primary-care-physician and mental-health-provider counts (already captured in the two columns above), not an all-specialty physician count. A true all-physician county rate would require HRSA's Area Health Resources File (county) or a CMS NPPES-derived count, neither of which could be pulled and correctly deduplicated by county within this job's scope.	rate per 1,000	Not populated	NOT SOURCED
`pm25_avg_ugm3`	Average daily fine particulate matter (PM2.5) concentration CHR pass-through of CDC/EPA Air Quality System modelled daily PM2.5 concentration estimates -- itself a MODELLED (not directly monitored-everywhere) estimate produced by CDC's Environmental Public Health Tracking Network, since ground monitors do not exist in every county.	micrograms per cubic meter (ug/m3) ~2021 (CDC/EPA Environmental Public Health Tracking vintage embedded in CHR 2024)	Modelled estimate	County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release
`drinking_water_violations`	Whether the county had a community water system with a health-based drinking water violation CHR pass-through of EPA Safe Drinking Water Information System (SDWIS). This is a binary presence/absence flag, not a violation count or rate.	binary (1 = had a violation, 0 = did not) ~2022 (embedded in CHR 2024)	Measured	County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release
`pct_lead_exposure_risk`	Percent of housing stock / population at elevated risk of lead exposure This column is intentionally empty for every row -- NOT SOURCED. No machine-readable, county-level, nationally complete bulk dataset for lead exposure risk could be located within this job's scope (CDC's childhood blood lead surveillance data is state-level and has large reporting gaps; HUD/EPA housing-age-based lead risk indices were not available as a clean bulk download).	percent (0-100)	Not populated	NOT SOURCED
`social_association_index`	Number of membership associations (civic, religious, business, political, labor, etc.) per 10,000 population CHR pass-through of Census County Business Patterns. Already expressed per-10,000; no further conversion applied.	count per 10,000 population ~2022 (County Business Patterns vintage embedded in CHR 2024)	Measured	County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release
`pct_veterans`	Percent of civilian population 18+ who are veterans ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B21001.	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_foreign_born`	Percent of population that is foreign-born ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B05002.	percent (0-100) 2020-2024	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`pct_broadband`	Percent of households with a broadband internet subscription of any type ACS 5-year estimates are themselves modelled/smoothed over a 5-year rolling window (2020-2024), not a single-year census count; small counties carry wide margins of error not included in this file. Connecticut's 8 legacy counties (FIPS 09001-09015) have no data here because the Census Bureau retired those FIPS codes in favor of 9 new Connecticut planning regions for the 2020-2024 vintage -- see README. From table B28002; CHR 'Broadband Access raw value' used as fallback for the 8 Connecticut counties absent from the ACS pull.	percent (0-100) 2020-2024 (ACS) or CHR vintage for 8 CT counties	Modelled estimate	US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API)
`norc_bene_cnt_2020`	Count of Medicare beneficiaries meeting NORC Dementia DataHub inclusion criteria, county total (all ages, all insurance types), 2020 -- the denominator for the two rate columns below Joined additively from data/county_dementia_outcomes.csv (see that file's meta for full methodology). All-age Medicare beneficiary population, not total county population. Blank where suppressed (true value < 11 per CMS rule) -- see norc_bene_cnt_2020_suppressed; blank is NOT zero. 15 of 3,143 panel counties have no match in the source (11 Alaska boroughs/census areas with no corresponding SSA county code, Kalawao County HI, and all 3 Delaware counties -- Delaware's source row is a state-level total that cannot be attributed to any single county, so it was excluded rather than misattributed; see build script).	count (persons) 2020	Measured	NORC Dementia DataHub (NORC at the University of Chicago), 2020 Public Use File
`norc_bene_cnt_2020_suppressed`	1 if norc_bene_cnt_2020 was suppressed at the source (blank, true value < 11); 0 if a real value is present; blank if the county had no NORC match at all Distinguishes genuine suppression from no-match: blank in this column means the county had no NORC row at all (see norc_bene_cnt_2020 caveat), not that a value was suppressed.	boolean (0/1) or blank 2020	Measured	Derived from NORC Dementia DataHub suppression flag
`norc_prev_cnt_2020`	Count of Medicare beneficiaries with prevalent diagnosed AD/ADRD in 2020, county total, broadest NORC case definition (DEMENTIACAT blank = 'All Determinations' = Highly likely + Likely + Possible combined) DIAGNOSED cases identified from Medicare claims, NOT a modelled disease-prevalence estimate -- diagnosis rates for identical underlying symptoms vary by place, so a county with a low count/rate may under-diagnose rather than have less disease. This is a distinct measurement from the panel's own alz_count_estimate column (which remains NOT SOURCED and untouched by this join). Blank where suppressed (true value < 11); blank is NOT zero. See data/county_dementia_outcomes_meta.json for full case-definition and suppression methodology.	count (persons) 2020	Modelled estimate	NORC Dementia DataHub (NORC at the University of Chicago), 2020 Public Use File
`norc_prev_cnt_2020_suppressed`	1 if norc_prev_cnt_2020 was suppressed at the source; 0 if present; blank if the county had no NORC match 24 of the 3,128 matched counties are suppressed at prev_cnt.	boolean (0/1) or blank 2020	Measured	Derived from NORC Dementia DataHub suppression flag
`norc_inc_cnt_2020`	Count of Medicare beneficiaries newly (first-time) diagnosed with AD/ADRD in calendar year 2020, county total, broadest NORC case definition Single calendar-year first-time-diagnosis count from Medicare claims. Blank where suppressed (true value < 11); blank is NOT zero.	count (persons) 2020	Measured	NORC Dementia DataHub (NORC at the University of Chicago), 2020 Public Use File
`norc_inc_cnt_2020_suppressed`	1 if norc_inc_cnt_2020 was suppressed at the source; 0 if present; blank if the county had no NORC match	boolean (0/1) or blank 2020	Measured	Derived from NORC Dementia DataHub suppression flag
`norc_mort_cnt_2020`	Count of all-cause deaths in 2020 among beneficiaries defined as prevalent AD/ADRD cases, county total ALL-CAUSE mortality among the diagnosed-prevalent population, NOT deaths with an ICD-10 dementia/Alzheimer's cause-of-death code -- do not conflate with the panel's own (NOT SOURCED) dementia_deaths_total/alz_deaths_total columns, which were defined as cause-of-death counts and remain untouched by this join. Blank where suppressed (true value < 11); blank is NOT zero.	count (deaths) 2020	Measured	NORC Dementia DataHub (NORC at the University of Chicago), 2020 Public Use File
`norc_mort_cnt_2020_suppressed`	1 if norc_mort_cnt_2020 was suppressed at the source; 0 if present; blank if the county had no NORC match	boolean (0/1) or blank 2020	Measured	Derived from NORC Dementia DataHub suppression flag
`norc_dementia_prevalence_rate_2020`	norc_prev_cnt_2020 / norc_bene_cnt_2020 * 100 -- diagnosed dementia prevalence rate among all-age Medicare beneficiaries Blank whenever prev_cnt or bene_cnt is suppressed -- never computed from an imputed numerator or denominator. Denominator is ALL-AGE Medicare beneficiaries (includes under-65 disability/ESRD beneficiaries), not restricted to 65+, so this is NOT directly comparable to the panel's own alz_prevalence_pct_65plus column (which remains NOT SOURCED and untouched) despite the similar name.	percent (0-100) 2020	Measured	Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (dementiadatahub.org/resources/outcome-measures.html)
`norc_dementia_incidence_rate_2020`	norc_inc_cnt_2020 / norc_bene_cnt_2020 * 100 -- diagnosed dementia incidence rate (first-time diagnoses in 2020) among all-age Medicare beneficiaries Blank whenever inc_cnt or bene_cnt is suppressed.	percent (0-100) 2020	Measured	Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula
`norc_dementia_mortality_rate_2020`	norc_mort_cnt_2020 / norc_prev_cnt_2020 * 100 -- share of diagnosed prevalent dementia cases who died in 2020 (a case-fatality rate among the diagnosed population) IMPORTANT: this is NOT a population mortality rate per 100,000 residents and is NOT comparable to the panel's own (NOT SOURCED) dementia_mortality_per_100k column, despite both being 'dementia mortality' -- it is the percent of already-diagnosed prevalent cases who died that year, i.e. a case-fatality rate, per the DataHub's own denominator guidance. Blank whenever mort_cnt or prev_cnt is suppressed.	percent (0-100) 2020	Measured	Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (mortality and COVID measures are divided by prev_cnt, not bene_cnt, per the codebook)
`pct_low_access_to_store_2019`	Percent of county population living more than 1 mile (urban) / 10 miles (rural) from the nearest supermarket, supercenter or large grocery store USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	percent (0-100) 2019	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`pct_low_income_low_access_2019`	Percent of county population that is both low-income and low-access to a supermarket or large grocery store USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	percent (0-100) 2019	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`pct_seniors_low_access_2019`	Percent of county population age 65+ living low-access to a supermarket or large grocery store USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	percent (0-100) 2019	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`grocery_stores_per_1000pop_2020`	Grocery stores per 1,000 population USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	stores per 1,000 population 2020	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`convenience_stores_per_1000pop_2020`	Convenience stores per 1,000 population USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	stores per 1,000 population 2020	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`supercenters_club_stores_per_1000pop_2020`	Supercenters and club stores per 1,000 population USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	stores per 1,000 population 2020	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`specialized_food_stores_per_1000pop_2020`	Specialized food stores (meat markets, produce markets, etc.) per 1,000 population USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	stores per 1,000 population 2020	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`dollar_stores_per_1000pop_2020`	Dollar stores per 1,000 population USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	stores per 1,000 population 2020	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`fastfood_restaurants_per_1000pop_2020`	Fast-food restaurants per 1,000 population USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	stores per 1,000 population 2020	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`fullservice_restaurants_per_1000pop_2020`	Full-service restaurants per 1,000 population USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	stores per 1,000 population 2020	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`snap_authorized_stores_per_1000pop_2023`	SNAP-authorized retail stores per 1,000 population USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	stores per 1,000 population 2023	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`wic_authorized_stores_per_1000pop_2022`	WIC-authorized retail stores per 1,000 population USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	stores per 1,000 population 2022	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`snap_benefits_per_capita_2022`	Average annual SNAP benefits redeemed per capita USDA Economic Research Service, Food Environment Atlas 2025 release. Joins on county_fips against panel_county.csv:fips at 3,143/3,143 (100%, including all 8 Connecticut legacy counties -- this file does not have the CT FIPS-scheme problem that affects the CMS/BLS columns added in the same join). Blank cells reflect gaps in USDA's own file (commonly for rare store categories like supercenters in small counties); USDA's public documentation does not distinguish a true zero from a suppressed or unreported cell in this release, so blank is left blank rather than read as zero.	US dollars per capita 2022	Measured	USDA Economic Research Service, Food Environment Atlas (2025 release)
`num_tracts`	Number of census tracts in the county USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pop2010_total`	Total population, 2010 Census (denominator for several shares in this group) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`ohu2010_total`	Total occupied housing units, 2010 Census USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_group_quarters_total`	Total group-quarters population, 2010 Census USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_tracts_urban`	Percent of the county's census tracts classified as urban USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_low_access_pop_halfmi`	Percent of population living low-access to a supermarket/large grocery store (urban: >0.5 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_low_income_low_access_pop_halfmi`	Percent of population that is both low-income and low-access (urban: >0.5 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_kids_low_access_halfmi`	Percent of population under 18 living low-access (urban: >0.5 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_seniors_low_access_halfmi`	Percent of population 65+ living low-access (urban: >0.5 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_housing_units_no_vehicle_low_access_halfmi`	Percent of housing units with no vehicle that are also low-access (urban: >0.5 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_housing_units_snap_low_access_halfmi`	Percent of SNAP-household housing units that are also low-access (urban: >0.5 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_low_access_pop_1mi`	Percent of population living low-access to a supermarket/large grocery store (>1 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_low_income_low_access_pop_1mi`	Percent of population that is both low-income and low-access (>1 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_kids_low_access_1mi`	Percent of population under 18 living low-access (>1 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_seniors_low_access_1mi`	Percent of population 65+ living low-access (>1 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_housing_units_no_vehicle_low_access_1mi`	Percent of housing units with no vehicle that are also low-access (>1 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_housing_units_snap_low_access_1mi`	Percent of SNAP-household housing units that are also low-access (>1 mile) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_low_access_pop_10mi`	Percent of population living low-access to a supermarket/large grocery store (rural: >10 miles) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_low_income_low_access_pop_10mi`	Percent of population that is both low-income and low-access (rural: >10 miles) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_low_access_pop_20mi`	Percent of population living low-access to a supermarket/large grocery store (rural: >20 miles) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`pct_low_income_low_access_pop_20mi`	Percent of population that is both low-income and low-access (rural: >20 miles) USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`tract_lowincome_pop_total`	Total low-income population summed across the county's tracts USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`tract_kids_pop_total`	Total population under 18 summed across the county's tracts USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`tract_seniors_pop_total`	Total population 65+ summed across the county's tracts USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`tract_housing_units_no_vehicle_total`	Total housing units with no vehicle available, summed across the county's tracts USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`tract_housing_units_snap_total`	Total SNAP-recipient housing units, summed across the county's tracts USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`poverty_rate_popweighted_avg_pct`	Population-weighted average tract poverty rate USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`median_family_income_popweighted_avg_usd`	Population-weighted average tract median family income, US dollars USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	percent (0-100) 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_lila_1and10`	Number of tracts flagged low-income-and-low-access under the 1-and-10-mile definition USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_lila_halfand10`	Number of tracts flagged low-income-and-low-access under the half-and-10-mile definition USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_lila_1and20`	Number of tracts flagged low-income-and-low-access under the 1-and-20-mile definition USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_lila_vehicle`	Number of tracts flagged low-income-and-low-access under the vehicle-access definition USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_low_income`	Number of tracts classified low-income USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_hunv_flag`	Number of tracts flagged for housing units with no vehicle and low access USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_la1and10`	Number of tracts flagged low-access under the 1-and-10-mile definition USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_lahalfand10`	Number of tracts flagged low-access under the half-and-10-mile definition USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_la1and20`	Number of tracts flagged low-access under the 1-and-20-mile definition USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_latracts_half`	Number of tracts flagged low-access at the half-mile threshold USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_latracts1`	Number of tracts flagged low-access at the 1-mile threshold USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_latracts10`	Number of tracts flagged low-access at the 10-mile threshold USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_latracts20`	Number of tracts flagged low-access at the 20-mile threshold USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`num_tracts_latracts_vehicle20`	Number of tracts flagged low-access at 20 miles among households with no vehicle USDA Economic Research Service, Food Access Research Atlas, 2019 vintage (initial release April 2021), aggregated from census tract to county. Joins on county_fips against panel_county.csv:fips at 3,141/3,143 (99.9%). The 2 unmatched are Alaska's Chugach (02063) and Copper River (02066) Census Areas, which post-date this 2019 file's pre-split Valdez-Cordova (02261) -- a real, separate join gap from the Connecticut FIPS issue that affects the CMS/BLS columns in this same batch; all 8 Connecticut legacy counties ARE present and populated in this file, left null rather than guessed for the 2 Alaska areas.	count 2019 data, initial release April 2021	Measured	USDA Economic Research Service, Food Access Research Atlas (2019 vintage, county aggregation)
`cms_nh_certified_beds_total`	Total CMS-certified nursing home beds in the county CMS Care Compare, Nursing Home Provider Info, August 2026 snapshot, aggregated to county by facility address (14,691 facilities, name-normalized county match). Joins on county_fips against panel_county.csv at 2,826/3,143 (89.9%). Connecticut is blank under EITHER FIPS scheme (legacy county or post-2022 planning region) -- confirmed by direct set-difference, not assumed: this CMS file has zero Connecticut rows at all, because Connecticut provider addresses no longer carry a legacy county FIPS to build one from. This is a genuine absence in the CMS source, not a fixable join gap -- no crosswalk exists for it, including at the state level (panel_state.csv's cms_nh_* columns are also blank for CT; see data/dataset_expansion.json 'KNOWN LIMITATION' entry). Blank for other non-CT counties this Care Compare extract does not cover, too -- blank is never filled with 0.	count 2026 (August 2026 Care Compare snapshot)	Measured	CMS Care Compare, Nursing Home Provider Info
`cms_nh_n_facilities`	Number of CMS-certified nursing home facilities in the county CMS Care Compare, Nursing Home Provider Info, August 2026 snapshot, aggregated to county by facility address (14,691 facilities, name-normalized county match). Joins on county_fips against panel_county.csv at 2,826/3,143 (89.9%). Connecticut is blank under EITHER FIPS scheme (legacy county or post-2022 planning region) -- confirmed by direct set-difference, not assumed: this CMS file has zero Connecticut rows at all, because Connecticut provider addresses no longer carry a legacy county FIPS to build one from. This is a genuine absence in the CMS source, not a fixable join gap -- no crosswalk exists for it, including at the state level (panel_state.csv's cms_nh_* columns are also blank for CT; see data/dataset_expansion.json 'KNOWN LIMITATION' entry). Blank for other non-CT counties this Care Compare extract does not cover, too -- blank is never filled with 0.	count 2026 (August 2026 Care Compare snapshot)	Measured	CMS Care Compare, Nursing Home Provider Info
`cms_nh_beds_per_1000_pop`	CMS-certified nursing home beds per 1,000 population (beds divided by panel_county.csv's own population column) CMS Care Compare, Nursing Home Provider Info, August 2026 snapshot, aggregated to county by facility address (14,691 facilities, name-normalized county match). Joins on county_fips against panel_county.csv at 2,826/3,143 (89.9%). Connecticut is blank under EITHER FIPS scheme (legacy county or post-2022 planning region) -- confirmed by direct set-difference, not assumed: this CMS file has zero Connecticut rows at all, because Connecticut provider addresses no longer carry a legacy county FIPS to build one from. This is a genuine absence in the CMS source, not a fixable join gap -- no crosswalk exists for it, including at the state level (panel_state.csv's cms_nh_* columns are also blank for CT; see data/dataset_expansion.json 'KNOWN LIMITATION' entry). Blank for other non-CT counties this Care Compare extract does not cover, too -- blank is never filled with 0.	beds per 1,000 population 2026 (August 2026 Care Compare snapshot)	Measured	CMS Care Compare, Nursing Home Provider Info
`cms_ma_total_medicare_beneficiaries`	Total Medicare beneficiaries (Original Medicare + Medicare Advantage), county total, June 2026 CMS Medicare Monthly Enrollment, June 2026 (most recent complete month; source file covers 2013-06 through 2026-06). Joins on county_fips against panel_county.csv at 3,135/3,143 (99.7%); the 8 unmatched are Connecticut. This CMS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column.	count (beneficiaries) June 2026	Measured	CMS Medicare Monthly Enrollment
`cms_ma_original_medicare_beneficiaries`	Original (fee-for-service) Medicare beneficiaries, county total, June 2026 CMS Medicare Monthly Enrollment, June 2026 (most recent complete month; source file covers 2013-06 through 2026-06). Joins on county_fips against panel_county.csv at 3,135/3,143 (99.7%); the 8 unmatched are Connecticut. This CMS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column.	count (beneficiaries) June 2026	Measured	CMS Medicare Monthly Enrollment
`cms_ma_advantage_beneficiaries`	Medicare Advantage and other plan beneficiaries, county total, June 2026 CMS Medicare Monthly Enrollment, June 2026 (most recent complete month; source file covers 2013-06 through 2026-06). Joins on county_fips against panel_county.csv at 3,135/3,143 (99.7%); the 8 unmatched are Connecticut. This CMS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column.	count (beneficiaries) June 2026	Measured	CMS Medicare Monthly Enrollment
`medicare_advantage_penetration_pct`	Medicare Advantage enrollment as a percent of total Medicare beneficiaries, June 2026 (= advantage_beneficiaries / total_medicare_beneficiaries * 100, from CMS's own published counts) CMS Medicare Monthly Enrollment, June 2026 (most recent complete month; source file covers 2013-06 through 2026-06). Joins on county_fips against panel_county.csv at 3,135/3,143 (99.7%); the 8 unmatched are Connecticut. This CMS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column.	percent (0-100) June 2026	Measured	CMS Medicare Monthly Enrollment
`naics62_health_socasst_estabs`	Number of health care and social assistance establishments (NAICS 62, broad sector), 2024 annual average BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment, 2024 annual averages. Joins on county_fips against panel_county.csv at 3,121/3,143 (99.3%); 22 unmatched = Connecticut's 8 legacy counties plus 14 non-CT counties BLS itself suppresses or omits from this extract (HI 15005; MT 30103; NE 31005, 31007, 31009, 31075, 31085, 31103, 31117; NV 32009, 32011; ND 38087; TX 48173, 48301 -- confirmed by direct set-difference against panel_county.csv, not estimated). This BLS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column. BLS does not publish a county-level 'Total, all ownerships' row in this file; the _estabs and _total_annual_wages / _avg_annual_pay_usd columns here are a direct sum of the 4 published ownership-sector rows (Federal/State/Local government + Private), so any county flagged by _any_ownership_cell_suppressed = 'Y' has a total that is a floor, not a complete count. _estabs and _emplvl_annual_avg are BLS's own 4-quarter averages, not a point-in-time snapshot; _total_annual_wages is a true annual sum; _avg_annual_pay_usd = total_annual_wages / emplvl_annual_avg and inherits that averaging.	count 2024 annual averages	Measured	BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment
`naics62_health_socasst_emplvl_annual_avg`	Average annual employment level at health care and social assistance establishments (NAICS 62), 2024 BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment, 2024 annual averages. Joins on county_fips against panel_county.csv at 3,121/3,143 (99.3%); 22 unmatched = Connecticut's 8 legacy counties plus 14 non-CT counties BLS itself suppresses or omits from this extract (HI 15005; MT 30103; NE 31005, 31007, 31009, 31075, 31085, 31103, 31117; NV 32009, 32011; ND 38087; TX 48173, 48301 -- confirmed by direct set-difference against panel_county.csv, not estimated). This BLS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column. BLS does not publish a county-level 'Total, all ownerships' row in this file; the _estabs and _total_annual_wages / _avg_annual_pay_usd columns here are a direct sum of the 4 published ownership-sector rows (Federal/State/Local government + Private), so any county flagged by _any_ownership_cell_suppressed = 'Y' has a total that is a floor, not a complete count. _estabs and _emplvl_annual_avg are BLS's own 4-quarter averages, not a point-in-time snapshot; _total_annual_wages is a true annual sum; _avg_annual_pay_usd = total_annual_wages / emplvl_annual_avg and inherits that averaging.	count 2024 annual averages	Measured	BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment
`naics62_health_socasst_total_annual_wages`	Total annual wages paid by health care and social assistance establishments (NAICS 62), 2024, US dollars BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment, 2024 annual averages. Joins on county_fips against panel_county.csv at 3,121/3,143 (99.3%); 22 unmatched = Connecticut's 8 legacy counties plus 14 non-CT counties BLS itself suppresses or omits from this extract (HI 15005; MT 30103; NE 31005, 31007, 31009, 31075, 31085, 31103, 31117; NV 32009, 32011; ND 38087; TX 48173, 48301 -- confirmed by direct set-difference against panel_county.csv, not estimated). This BLS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column. BLS does not publish a county-level 'Total, all ownerships' row in this file; the _estabs and _total_annual_wages / _avg_annual_pay_usd columns here are a direct sum of the 4 published ownership-sector rows (Federal/State/Local government + Private), so any county flagged by _any_ownership_cell_suppressed = 'Y' has a total that is a floor, not a complete count. _estabs and _emplvl_annual_avg are BLS's own 4-quarter averages, not a point-in-time snapshot; _total_annual_wages is a true annual sum; _avg_annual_pay_usd = total_annual_wages / emplvl_annual_avg and inherits that averaging.	US dollars 2024 annual averages	Measured	BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment
`naics62_health_socasst_avg_annual_pay_usd`	Average annual pay per employee, health care and social assistance establishments (NAICS 62), 2024, US dollars BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment, 2024 annual averages. Joins on county_fips against panel_county.csv at 3,121/3,143 (99.3%); 22 unmatched = Connecticut's 8 legacy counties plus 14 non-CT counties BLS itself suppresses or omits from this extract (HI 15005; MT 30103; NE 31005, 31007, 31009, 31075, 31085, 31103, 31117; NV 32009, 32011; ND 38087; TX 48173, 48301 -- confirmed by direct set-difference against panel_county.csv, not estimated). This BLS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column. BLS does not publish a county-level 'Total, all ownerships' row in this file; the _estabs and _total_annual_wages / _avg_annual_pay_usd columns here are a direct sum of the 4 published ownership-sector rows (Federal/State/Local government + Private), so any county flagged by _any_ownership_cell_suppressed = 'Y' has a total that is a floor, not a complete count. _estabs and _emplvl_annual_avg are BLS's own 4-quarter averages, not a point-in-time snapshot; _total_annual_wages is a true annual sum; _avg_annual_pay_usd = total_annual_wages / emplvl_annual_avg and inherits that averaging.	US dollars 2024 annual averages	Measured	BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment
`naics62_any_ownership_cell_suppressed`	'Y' if at least one of the 4 ownership-sector cells (Federal/State/Local government, Private) summed into the NAICS 62 county total was BLS-suppressed for confidentiality; blank otherwise BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment, 2024 annual averages. Joins on county_fips against panel_county.csv at 3,121/3,143 (99.3%); 22 unmatched = Connecticut's 8 legacy counties plus 14 non-CT counties BLS itself suppresses or omits from this extract (HI 15005; MT 30103; NE 31005, 31007, 31009, 31075, 31085, 31103, 31117; NV 32009, 32011; ND 38087; TX 48173, 48301 -- confirmed by direct set-difference against panel_county.csv, not estimated). This BLS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column. BLS does not publish a county-level 'Total, all ownerships' row in this file; the _estabs and _total_annual_wages / _avg_annual_pay_usd columns here are a direct sum of the 4 published ownership-sector rows (Federal/State/Local government + Private), so any county flagged by _any_ownership_cell_suppressed = 'Y' has a total that is a floor, not a complete count. _estabs and _emplvl_annual_avg are BLS's own 4-quarter averages, not a point-in-time snapshot; _total_annual_wages is a true annual sum; _avg_annual_pay_usd = total_annual_wages / emplvl_annual_avg and inherits that averaging.	flag ('Y' or blank) 2024 annual averages	Measured	BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment
`naics623_nursing_residential_care_estabs`	Number of nursing and residential care facility establishments (NAICS 623, a subsector of 62), 2024 annual average BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment, 2024 annual averages. Joins on county_fips against panel_county.csv at 3,121/3,143 (99.3%); 22 unmatched = Connecticut's 8 legacy counties plus 14 non-CT counties BLS itself suppresses or omits from this extract (HI 15005; MT 30103; NE 31005, 31007, 31009, 31075, 31085, 31103, 31117; NV 32009, 32011; ND 38087; TX 48173, 48301 -- confirmed by direct set-difference against panel_county.csv, not estimated). This BLS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column. BLS does not publish a county-level 'Total, all ownerships' row in this file; the _estabs and _total_annual_wages / _avg_annual_pay_usd columns here are a direct sum of the 4 published ownership-sector rows (Federal/State/Local government + Private), so any county flagged by _any_ownership_cell_suppressed = 'Y' has a total that is a floor, not a complete count. _estabs and _emplvl_annual_avg are BLS's own 4-quarter averages, not a point-in-time snapshot; _total_annual_wages is a true annual sum; _avg_annual_pay_usd = total_annual_wages / emplvl_annual_avg and inherits that averaging.	count 2024 annual averages	Measured	BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment
`naics623_nursing_residential_care_emplvl_annual_avg`	Average annual employment level at nursing and residential care facilities (NAICS 623), 2024 BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment, 2024 annual averages. Joins on county_fips against panel_county.csv at 3,121/3,143 (99.3%); 22 unmatched = Connecticut's 8 legacy counties plus 14 non-CT counties BLS itself suppresses or omits from this extract (HI 15005; MT 30103; NE 31005, 31007, 31009, 31075, 31085, 31103, 31117; NV 32009, 32011; ND 38087; TX 48173, 48301 -- confirmed by direct set-difference against panel_county.csv, not estimated). This BLS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column. BLS does not publish a county-level 'Total, all ownerships' row in this file; the _estabs and _total_annual_wages / _avg_annual_pay_usd columns here are a direct sum of the 4 published ownership-sector rows (Federal/State/Local government + Private), so any county flagged by _any_ownership_cell_suppressed = 'Y' has a total that is a floor, not a complete count. _estabs and _emplvl_annual_avg are BLS's own 4-quarter averages, not a point-in-time snapshot; _total_annual_wages is a true annual sum; _avg_annual_pay_usd = total_annual_wages / emplvl_annual_avg and inherits that averaging.	count 2024 annual averages	Measured	BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment
`naics623_total_annual_wages`	Total annual wages paid by nursing and residential care facilities (NAICS 623), 2024, US dollars BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment, 2024 annual averages. Joins on county_fips against panel_county.csv at 3,121/3,143 (99.3%); 22 unmatched = Connecticut's 8 legacy counties plus 14 non-CT counties BLS itself suppresses or omits from this extract (HI 15005; MT 30103; NE 31005, 31007, 31009, 31075, 31085, 31103, 31117; NV 32009, 32011; ND 38087; TX 48173, 48301 -- confirmed by direct set-difference against panel_county.csv, not estimated). This BLS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column. BLS does not publish a county-level 'Total, all ownerships' row in this file; the _estabs and _total_annual_wages / _avg_annual_pay_usd columns here are a direct sum of the 4 published ownership-sector rows (Federal/State/Local government + Private), so any county flagged by _any_ownership_cell_suppressed = 'Y' has a total that is a floor, not a complete count. _estabs and _emplvl_annual_avg are BLS's own 4-quarter averages, not a point-in-time snapshot; _total_annual_wages is a true annual sum; _avg_annual_pay_usd = total_annual_wages / emplvl_annual_avg and inherits that averaging.	US dollars 2024 annual averages	Measured	BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment
`naics623_avg_annual_pay_usd`	Average annual pay per employee, nursing and residential care facilities (NAICS 623), 2024, US dollars BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment, 2024 annual averages. Joins on county_fips against panel_county.csv at 3,121/3,143 (99.3%); 22 unmatched = Connecticut's 8 legacy counties plus 14 non-CT counties BLS itself suppresses or omits from this extract (HI 15005; MT 30103; NE 31005, 31007, 31009, 31075, 31085, 31103, 31117; NV 32009, 32011; ND 38087; TX 48173, 48301 -- confirmed by direct set-difference against panel_county.csv, not estimated). This BLS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column. BLS does not publish a county-level 'Total, all ownerships' row in this file; the _estabs and _total_annual_wages / _avg_annual_pay_usd columns here are a direct sum of the 4 published ownership-sector rows (Federal/State/Local government + Private), so any county flagged by _any_ownership_cell_suppressed = 'Y' has a total that is a floor, not a complete count. _estabs and _emplvl_annual_avg are BLS's own 4-quarter averages, not a point-in-time snapshot; _total_annual_wages is a true annual sum; _avg_annual_pay_usd = total_annual_wages / emplvl_annual_avg and inherits that averaging.	US dollars 2024 annual averages	Measured	BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment
`naics623_any_ownership_cell_suppressed`	'Y' if at least one of the 4 ownership-sector cells summed into the NAICS 623 county total was BLS-suppressed for confidentiality; blank otherwise BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment, 2024 annual averages. Joins on county_fips against panel_county.csv at 3,121/3,143 (99.3%); 22 unmatched = Connecticut's 8 legacy counties plus 14 non-CT counties BLS itself suppresses or omits from this extract (HI 15005; MT 30103; NE 31005, 31007, 31009, 31075, 31085, 31103, 31117; NV 32009, 32011; ND 38087; TX 48173, 48301 -- confirmed by direct set-difference against panel_county.csv, not estimated). This BLS file carries Connecticut's 9 post-2022 planning regions (FIPS 09110, 09120, 09130, 09140, 09150, 09160, 09170, 09180, 09190, 09999) instead of the 8 legacy county FIPS panel_county.csv uses (09001-09015); zero FIPS keys overlap between the two schemes, so all 8 CT counties are blank here at the COUNTY level. No crosswalk was built: Connecticut's 2022 transition split 7 of the 8 legacy counties across 2-3 planning regions each, so a correct county-level correspondence does not exist -- see data/dataset_expansion.json 'KNOWN LIMITATION -- Connecticut county-FIPS mismatch' entry. Real Connecticut figures ARE recovered at the STATE level in panel_state.csv (computed directly from this source file, grouped by the state FIPS prefix '09', which is unambiguous under either scheme) -- use panel_state.csv for a Connecticut figure, not this county column. BLS does not publish a county-level 'Total, all ownerships' row in this file; the _estabs and _total_annual_wages / _avg_annual_pay_usd columns here are a direct sum of the 4 published ownership-sector rows (Federal/State/Local government + Private), so any county flagged by _any_ownership_cell_suppressed = 'Y' has a total that is a floor, not a complete count. _estabs and _emplvl_annual_avg are BLS's own 4-quarter averages, not a point-in-time snapshot; _total_annual_wages is a true annual sum; _avg_annual_pay_usd = total_annual_wages / emplvl_annual_avg and inherits that averaging.	flag ('Y' or blank) 2024 annual averages	Measured	BLS Quarterly Census of Employment and Wages (QCEW), county health-sector employment
`cms_hospice_facilities_count`	Count of distinct Medicare-certified hospice facilities (by CCN) headquartered in the county Full national file (453,493 long-format measure-rows, one row per facility x quality-measure) was fetched and deduplicated to 6,669 unique CCNs, then matched to panel_county.csv by (state, normalized County/Parish name). 6,551/6,669 facilities matched (98.2%); the 118 unmatched are genuine CMS data-quality gaps in the source file itself (county field literally '-', 'UNITED STATES', or a state name instead of a county) plus Puerto Rico/Guam/US Virgin Islands facilities outside the continental panel -- left blank, not guessed. Counts skew heavily toward a small number of counties (Los Angeles County CA = 1,160, ~18% of the national total) which matches independently-reported 2023-2024 hospice-licensing surges in Los Angeles/Orange/Maricopa/Clark/Harris/Dallas counties, not a join artifact -- cross-checked against the raw per-county count before treating as plausible.	count 2026 (August 2026 Care Compare snapshot)	Measured	CMS Care Compare, Hospice - Provider Data
`cms_hospice_facilities_per_1000pop`	cms_hospice_facilities_count per 1,000 county population Denominator is this panel's own 'population' column, not a Care Compare field.	facilities per 1,000 population 2026 facility snapshot over panel population	Measured	Derived (cms_hospice_facilities_count / population * 1000)
`cms_home_health_agencies_count`	Count of Medicare-certified home health agencies located in the county This CMS file carries only a ZIP code per agency, no county field, so agencies were mapped ZIP -> county FIPS via the Census Bureau 2020 ZCTA-to-county relationship file (data/sources/census_zcta_county_crosswalk/), choosing the county with the largest land-area overlap for ZCTAs that straddle more than one county. 12,375/12,460 agencies (99.3%) matched to a panel county; the remainder fell in ZCTAs with no panel-county match (territories or ZCTA/county edge cases). This is an approximate ZIP-based assignment, not the agency's legal county of record.	count 2026 (July 2026 Care Compare snapshot)	Measured	CMS Care Compare, Home Health Care Agencies
`cms_home_health_agencies_per_1000pop`	cms_home_health_agencies_count per 1,000 county population Denominator is this panel's own 'population' column, not a Care Compare field.	agencies per 1,000 population 2026 agency snapshot over panel population	Measured	Derived (cms_home_health_agencies_count / population * 1000)
`nurse_practitioners_npi_2024`	Nurse practitioners, NPI-registered, headcount in the county AHRF field np_npi_24. This is a headcount of NPI-registered nurse practitioners, not an FTE or workload measure. Checked and confirmed absent from this AHRF release: any county-level RN, LPN, or nursing-aide/CNA total, and any geriatrician-specific physician specialty field -- AHRF2025hp.csv's 1,927 columns were searched for RN/nurse/LPN/CNA/aide/geriatric patterns and only NP, PA, and clinical-nurse-specialist NPI counts exist. That is a genuine, checked gap in this federal source, not an oversight here.	count 2024	Measured	HRSA Area Health Resources Files (AHRF) 2024-2025, health professions sub-file
`nurse_practitioners_per_1000pop_2024`	nurse_practitioners_npi_2024 per 1,000 county population Denominator reuses population_est_2023 already extracted in data/sources/hrsa_ahrf/county_physician_hospital_supply.csv (same AHRF release), not this panel's population column.	NPs per 1,000 population 2024 over 2023 population estimate	Measured	Derived, HRSA AHRF
`physician_assistants_npi_2024`	Physician assistants, NPI-registered, headcount in the county AHRF field pa_npi_24. Headcount, not FTE.	count 2024	Measured	HRSA Area Health Resources Files (AHRF) 2024-2025, health professions sub-file
`physician_assistants_per_1000pop_2024`	physician_assistants_npi_2024 per 1,000 county population Denominator reuses population_est_2023 already extracted in data/sources/hrsa_ahrf/county_physician_hospital_supply.csv (same AHRF release), not this panel's population column.	PAs per 1,000 population 2024 over 2023 population estimate	Measured	Derived, HRSA AHRF
`clinical_nurse_specialists_npi_2024`	Clinical nurse specialists, NPI-registered, headcount in the county AHRF field clin_nurse_spec_npi_24. Headcount, not FTE.	count 2024	Measured	HRSA Area Health Resources Files (AHRF) 2024-2025, health professions sub-file
`clinical_nurse_specialists_per_1000pop_2024`	clinical_nurse_specialists_npi_2024 per 1,000 county population Denominator reuses population_est_2023 already extracted in data/sources/hrsa_ahrf/county_physician_hospital_supply.csv (same AHRF release), not this panel's population column.	CNS per 1,000 population 2024 over 2023 population estimate	Measured	Derived, HRSA AHRF

This table is wider than a phone screen — swipe it sideways to see the rest.

Country panel

48 columns · 217 rows · 62 KB · 14 modelled · 5 not populated

Download panel_country.csv Column dictionary (JSON)

{'structure': "One entry per panel_country.csv column. 'year' gives the year of the value actually recorded in this file — for indicators where we took the latest available year per country, this is a range (countries differ in which year is most recent).", 'general_caveat': "All values are 'latest available' per country as of the data pull (2026-09-21), not a single fixed reference year, unless noted otherwise. Blank cells mean the source had no value for that country — never a true zero.", 'join_key': 'iso3 (ISO 3166-1 alpha-3), sourced from the World Bank country master list.'}

panel_country.csv — what each column holds and where it came from.
Column	What it is	Unit / year	Basis	Source
`iso3`	ISO 3166-1 alpha-3 country/territory code Join key. 217 economies as classified by the World Bank (excludes WB's regional/income aggregates like 'World' or 'OECD members'). Includes some small territories (e.g. ABW, BMU) that IHME/WHO burden sources do not separately estimate, so burden columns will be blank for those rows.	code n/a	Measured	World Bank Country API
`country_name`	Country or territory name World Bank's short name; may differ from WHO's preferred name (e.g. 'Korea, Rep.' vs 'Republic of Korea').	text n/a	Measured	World Bank Country API
`region`	World Bank region grouping World Bank regions, not WHO regions (task allowed either). WB regions are geographic/operational, e.g. 'Europe & Central Asia' spans EU and Central Asian states together.	category n/a	Measured	World Bank Country API
`income_group`	World Bank income classification (fiscal-year 2026 vintage) Reclassified annually by the World Bank; a country's group can shift year to year independent of the other economic columns' reference years.	category: high/upper-middle/lower-middle/low 2026 classification (based mostly on 2024 GNI per capita)	Measured	World Bank Country API
`dementia_prevalence_pct`	Estimated share of population living with dementia Left empty for all countries. The only country-level source is IHME's Global Burden of Disease (GBD); the GBD Results Tool requires a login and interactive query (no anonymous bulk API), and the one mirror that republishes GBD country data (Our World in Data) explicitly blocks CSV download of this series ('non-redistributable data'). Rather than transcribe numbers from chart tooltips/PDFs (risk of transcription error / hallucination), this column was left blank. See report for what was tried.	% of population n/a	Not populated	NOT POPULATED
`dementia_prevalence_count`	Estimated number of people living with dementia Same reason as dementia_prevalence_pct — GBD-only, not obtainable without a licensed/authenticated query.	count (persons) n/a	Not populated	NOT POPULATED
`dementia_deaths_total`	Estimated total deaths from Alzheimer's disease and other dementias, both sexes, all ages GHE cause-of-death estimates are themselves modelled (redistributing ill-defined/garbage-coded deaths across causes using WHO's cause-of-death ensemble models), not raw vital registration counts, especially for countries with weak civil registration systems. Values are WHO's own modelled estimates (GHE), distinct from and generally close to but not identical to IHME/GBD's dementia mortality estimates.	count (deaths/year) latest per country, 2000-2021 (most countries: 2021)	Modelled estimate	WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export
`dementia_deaths_per_100k_age_std`	Death rate from Alzheimer's disease and other dementias, both sexes IMPORTANT: despite the column name (kept for schema consistency with the task spec), this is WHO GHE's crude, all-ages death rate ('age_group_allages' in the source's own variable short-name), NOT age-standardized. It is therefore NOT safe to compare directly across countries with very different age structures without also controlling for pct_65plus/median_age in your regression — an older country will show a mechanically higher crude dementia death rate even with identical age-specific risk. True age-standardized GBD dementia mortality exists but sits behind the same IHME access wall described above.	deaths per 100,000 population latest per country, 2000-2021 (most countries: 2021)	Modelled estimate	WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export
`dementia_daly_per_100k`	Disability-adjusted life years (DALYs) from Alzheimer's disease and other dementias, both sexes, all ages Modelled estimate (see dementia_deaths_total caveat). Crude/all-ages, not age-standardized. DALYs = years of life lost to premature death (YLL) + years lived with disability (YLD); for dementia the YLD component dominates and rests on modelled disease-severity distributions, adding another layer of estimation uncertainty beyond the mortality data alone.	DALYs per 100,000 population latest per country, 2000-2021 (most countries: 2021)	Modelled estimate	WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export
`dementia_incidence_per_100k`	New dementia cases per 100,000 population per year GBD-only metric, same access restriction as prevalence. WHO GHE does not publish incidence (only mortality/DALY/YLL/YLD).	cases per 100,000/year n/a	Not populated	NOT POPULATED
`population_total`	Total population, all ages Midyear estimates; WB's most recent year is typically a projection carried from UN WPP revisions.	count (persons) latest per country, mostly 2025	Modelled estimate	World Bank (UN World Population Prospects-derived)
`population_65plus`	Population aged 65 and above, total Same vintage/estimation basis as population_total.	count (persons) latest per country, mostly 2025	Measured	World Bank
`pct_65plus`	Population aged 65+ as % of total population Directly derivable from population_65plus / population_total; included pre-computed as provided by WB.	% of total population latest per country, mostly 2025	Measured	World Bank
`pct_urban`	Urban population as % of total population National definitions of 'urban' vary by country and are not harmonized by WB.	% latest per country, mostly 2025	Measured	World Bank
`life_expectancy`	Life expectancy at birth, total population None material.	years latest per country, mostly 2024	Measured	World Bank

Show all 48 columns (15 shown so far)

This table is wider than a phone screen — swipe it sideways to see the rest.

Public datasets we drew on

18 sources

These are other people's datasets, not ours. Listed so you can go to the original rather than trusting our copy of it.

CDC BRFSS

open download

CDC / National Center for Chronic Disease Prevention and Health Promotion

Annual telephone survey of health-related risk behaviors, chronic conditions (including a cognitive decline/caregiver module), and preventive service use.

Geography: National, state, and (via SMART) select city/county
Years: 1984-present (cognitive decline module varies by state/year)

Why we used it: State-level cognitive-decline and caregiver-burden self-report data, and the underlying source for most of the pct_diabetes/obesity/smoking/etc columns in panel_state.csv.

Verified live 2026-09-21; annual microdata files plus GIS/analysis tools; also indexed on HealthData.gov.

cdc.gov

CDC WONDER

open download

CDC / NCHS

Ad-hoc mortality query system; underlying/multiple cause-of-death counts and rates by ICD-10 code (G30 = Alzheimer's), year, age, sex, race, state, county.

Geography: National, state, county
Years: 1968-present (detailed cause-of-death 1999-present)

Why we used it: The authoritative source for filling the mortality gaps left blank in timeseries_national.csv, and for cross-checking the Facts & Figures state mortality table.

Verified live 2026-09-21. Query interface is form-based/interactive; cannot be scraped via a simple GET request -- would need a scripted form submission to build a continuous year-by-year mortality series.

wonder.cdc.gov

NHANES

open download

CDC / NCHS

Combined interview and physical exam survey with labs, biospecimens, and (in later cycles) cognitive testing modules.

Geography: National only (not state-representative)
Years: 1960s (initial cycles), continuous since 1999; most recent public releases through 2021-2023

Why we used it: Biomarker/lab and cognitive-screen data for national-level covariates (e.g., cardiometabolic risk factors) against national AD trends.

Verified live 2026-09-21.

cdc.gov

NHATS (National Health and Aging Trends Study)

registration

Johns Hopkins Bloomberg School of Public Health, funded by NIA

Annual in-person interviews of Medicare beneficiaries 65+ on disability, cognitive impairment, dementia status, and late-life trajectories, linkable to Medicare claims.

Geography: Nationally representative sample (respondent-level, not state-representative)
Years: 2011-present (annual rounds)

Why we used it: Individual-level dementia/disability trajectories linkable to Medicare claims -- strong for cost/utilization correlation work.

Verified live 2026-09-21. Free registration + data-use terms required for public-use files; more sensitive/restricted files require a separate application.

nhats.org

ADNI (Alzheimer's Disease Neuroimaging Initiative)

restricted application

Multi-site consortium, coordinated via LONI/USC, funded by NIA

Longitudinal MRI/PET imaging, biomarkers (CSF, blood, genetics), and clinical/cognitive data from ~2,500+ participants across the AD spectrum.

Geography: Site-level, US and Canada (~60 sites), not population-representative
Years: 2004-present (ADNI 1/GO/2/3/4 phases)

Why we used it: Gold-standard imaging/biomarker cohort for validating disease-stage definitions used to interpret national prevalence estimates.

Verified live 2026-09-21. Requires creating an LONI/IDA account, agreeing to the ADNI Data Use Agreement, and (for biospecimens) a separate formal application/review -- genuinely gated, not open download.

adni.loni.usc.edu

NACC (National Alzheimer's Coordinating Center)

registration

University of Washington, funded by NIA, aggregating NIA-funded Alzheimer's Disease Research Centers (ADRCs)

Centralized clinical, neuropathology, imaging, and biomarker data pooled from 37+ active ADRCs (56,000+ participants, 217,000+ clinical visits).

Geography: Multi-site (25 states with active ADRCs), not population-representative
Years: ~1984-present (UDS harmonized data primarily 2005-present)

Why we used it: Clinically-adjudicated dementia diagnoses across research centers -- useful to sanity-check self-report/administrative prevalence estimates.

Verified live 2026-09-21 via https://www.naccdata.org/data-request-process: submit a brief project request, sign an electronic Data Use Agreement (DUA), then receive a download link within ~3 business days -- a real gate but lighter-weight than ADNI's, not simply 'open download.'

naccdata.org

CMS ResDAC (Medicare/Medicaid claims research data)

restricted application

CMS, technical assistance via University of Minnesota ResDAC

Access point/documentation for Medicare (Parts A/B/C/D) and Medicaid claims, enrollment, and assessment (e.g., MDS) research-identifiable files.

Geography: Beneficiary/claim level, aggregable to county/state/national
Years: Varies by file type; Medicare claims generally 1999-present

Why we used it: The actual source needed to split Medicare-only vs. Medicaid-only Alzheimer's spending, which the published Facts & Figures aggregates do not cleanly separate (see caveats on the *_spend_total_usd columns).

Verified live 2026-09-21. Requires submitting a data request form plus a CMS Data Use Agreement (DUA) to CMS; CMS notes high request volume and processing delays.

resdac.org

IHME GBD (Global Burden of Disease)

registration

Institute for Health Metrics and Evaluation, University of Washington

Modeled estimates of deaths, DALYs, prevalence, and risk-factor attribution for Alzheimer's/dementia by cause.

Geography: National and subnational (204 countries, 660 subnational locations)
Years: 1990-2023 (GBD 2023 round)

Why we used it: Enables international/cross-national benchmarking of the US Alzheimer's burden trend against comparable modeled global estimates.

Direct fetch returned 403 (bot-blocked); confirmed via search that free registration/sign-in is required to download CSV extracts (up to 100,000 rows/request) -- the interactive viewing tool itself is public.

vizhub.healthdata.org

US Census / ACS

open download

US Census Bureau

Annual demographic, economic, and housing survey (ACS) plus intercensal population estimates by age; source for the 65+ denominator and most of the demographic/socioeconomic panel columns.

Geography: National, state, county, tract (ACS 1-yr/5-yr)
Years: ACS ongoing since 2005; Population Estimates Program annual since 2000s

Why we used it: Provides the population-65+ denominator required to compute per-capita AD prevalence/mortality rates by state/year, plus income, education, race, and poverty covariates.

Verified live 2026-09-21.

census.gov

County Health Rankings

open download

University of Wisconsin Population Health Institute / Robert Wood Johnson Foundation

County- and state-level composite health outcome and health-factor measures (mortality, morbidity, social determinants).

Geography: County and state
Years: 2010-present annual releases

Why we used it: County-level social determinants (income, education, access to care) to correlate against local Alzheimer's mortality/prevalence at finer geography than state.

Verified live 2026-09-21; web explorer confirmed open, no login required.

countyhealthrankings.org

Area Health Resources File (AHRF)

open download

HRSA, Bureau of Health Workforce / National Center for Health Workforce Analysis

6,000+ variables on health care professions, facilities, population characteristics, and health economics.

Geography: County, state, national
Years: Multi-decade historical files; dashboard covers 2019-2024

Why we used it: County-level physician/geriatrician and long-term-care workforce supply data to correlate against the dementia_care_workforce_requirement policy variable and healthcare-access columns.

Verified live 2026-09-21; downloadable in SAS/CSV/ASCII with no registration barrier found.

data.hrsa.gov

Alzheimer's Association state-level data (Facts and Figures)

open download

Alzheimer's Association

Annual state-by-state Alzheimer's prevalence, projected growth, deaths, caregiver, geriatrician-supply, Medicare/Medicaid spending, and workforce estimates.

Geography: State (all 50 + DC)
Years: Annual editions, at least 2010-2026 confirmed

Why we used it: The primary published state-level dataset underlying most of panel_state.csv's Alzheimer's-specific columns and timeseries_national.csv.

Verified live 2026-09-21 (and the 2025 full-text edition was fetched directly as the primary source for panel_state.csv's alz_*, unpaid_caregiver_*, medicaid_dementia_spend_total_usd, geriatricians_count and medicare_percapita_ffs_payment_usd columns).

alz.org

NIH RePORTER / RCDC Categorical Spending

open download

NIH Office of Extramural Research

Searchable database of NIH-funded grants/contracts, publications, and patents, including by RCDC disease category (Alzheimer's Disease, and Alzheimer's Disease and Related Dementias combined).

Geography: Project/institution level, aggregable to state/national
Years: FY1985-present (ExPORTER bulk files); categorical spending back to FY2008, FY2019-2025 fetched directly for this project

Why we used it: Grant-level Alzheimer's research funding data; the direct source for the research_funding_nih_adrd_usd column in timeseries_national.csv.

Verified live 2026-09-21 -- fetched directly as text and confirmed a static HTML table with columns for FY2019-FY2025 (values used in research_funding_nih_adrd_usd). ExPORTER (https://report.nih.gov/reporter) additionally offers bulk downloadable raw project-level data.

report.nih.gov

ClinicalTrials.gov API

API

NIH National Library of Medicine

Registered/results data for all US and international clinical trials, including condition, phase, status, enrollment, sponsor, location, and outcomes.

Geography: Trial-site level, global (includes US state/city)
Years: 1999-present (registry inception); ongoing

Why we used it: Tracks the AD drug-development pipeline (trial counts/phases by year) as a leading indicator alongside the fda_approved_alz_drugs_cumulative column.

Verified via direct query 2026-09-21: GET https://clinicaltrials.gov/api/v2/studies?query.cond=Alzheimer&pageSize=1 returned valid structured JSON (NCT ID, status, design, outcomes, locations) -- fully open REST API v2, no key required.

clinicaltrials.gov

EPA Air Quality System (AirData)

open download

US Environmental Protection Agency

Ambient air quality monitoring data including PM2.5 annual averages by monitor, county, and state (CBSA).

Geography: Monitor, county, state, national
Years: 1980-present (PM2.5 monitoring network established ~1999)

Why we used it: Raw monitor-level and county-level PM2.5 data would allow a finer-grained air-pollution correlate than the single state-average column included in panel_state.csv.

Confirmed the AirData download portal is publicly accessible; the pm25_avg_ugm3 column in panel_state.csv was sourced via America's Health Rankings' aggregation of this EPA data rather than pulled directly from AirData's raw file interface in this session.

epa.gov

KFF State Health Facts

open download

KFF (Kaiser Family Foundation)

Aggregated state-by-state health policy indicators (insurance coverage, chronic disease prevalence, provider supply, hospital/nursing-home capacity), sourced from CDC BRFSS, ACS, AHA Annual Survey, and CMS.

Geography: State and national
Years: Varies by indicator, generally 2019-2024

Why we used it: Single consistent access point for many of the risk-factor and healthcare-capacity columns already included in the panel; useful for re-pulling fresher vintages later.

Verified live 2026-09-21; used as the primary aggregator for most of panel_state.csv's risk-factor and healthcare-access columns because it re-publishes CDC BRFSS/AHA data in clean, directly fetchable state tables (CDC's own BRFSS interactive tool is not simply scrapeable).

kff.org

CMS Care Compare, Hospice - Provider Data

open download

CMS (Centers for Medicare & Medicaid Services)

Medicare-certified hospice facility directory with CAHPS Hospice Survey and clinical quality measures, one row per facility x measure (453,493 rows nationally as fetched).

Geography: Facility-level (CCN), aggregated here to county via facility address
Years: 2026 (August 2026 snapshot)

Why we used it: Direct source for panel_county.csv's cms_hospice_facilities_count and cms_hospice_facilities_per_1000pop columns.

Fetched/re-verified 2026-09-27 (see data/sources/cms_hospice/README.md); metastore endpoint re-confirmed live (HTTP 200) 2026-09-30. Deduplicated to 6,669 unique CCNs and matched to panel_county.csv by (state, normalized county name) -- 6,551/6,669 (98.2%) matched; the remainder are genuine source-file gaps (missing/placeholder county field) or territories outside the continental panel.

data.cms.gov

CMS Care Compare, Home Health Care Agencies

open download

CMS (Centers for Medicare & Medicaid Services)

Medicare-certified home health agency directory (~12,460 agencies), including services offered, star ratings, and outcome measures.

Geography: Agency-level (ZIP code), mapped here to county via a Census ZCTA-to-county crosswalk
Years: 2026 (July 2026 snapshot)

Why we used it: Direct source for panel_county.csv's cms_home_health_agencies_count and cms_home_health_agencies_per_1000pop columns.

Fetched/re-verified 2026-09-27 (see data/sources/cms_home_health/README.md); metastore endpoint re-confirmed live (HTTP 200) 2026-09-30. This CMS file carries only a ZIP code, no county field, so agencies were mapped ZIP -> county FIPS via the 2020 Census ZCTA-to-county relationship file, using the largest land-area overlap for ZCTAs spanning multiple counties -- 12,375/12,460 (99.3%) matched to a panel county; an approximate ZIP-based assignment, not the agency's legal county of record.

data.cms.gov

Terms, in plain words

You may use these files. Most of the underlying data is public — from CMS, CDC, the NIH, ClinicalTrials.gov, Congress.gov and published papers. Our contribution is the collecting, cleaning and labelling. Please cite the original source rather than us where one exists; each record names it.

We make no warranty. These files are assembled by an automated research pipeline and checked against the source pages it opened. Errors are possible. Do not use them to make a clinical decision.

How to read the labels: Measured counted, surveyed or written in law. Modelled estimate produced by a statistical model — an estimate, not a count. Basis not stated the source did not say which.

Found a wrong number? Tell us and we will fix it or remove it.