Rural & Persistent Poverty
Counties Atlas

Definitions

Persistent-poverty county

A county whose poverty rate was 20% or higher in all four reference periods: 1990 Decennial Census, 2000 Decennial Census, 2007–2011 ACS 5-year, and 2017–2021 ACS 5-year. The flag is carried in USDA ERS's 2025 County Typology Codes.

Nonmetro county

A county with RUCC code 4–9 — outside a Metropolitan Statistical Area.

Completely rural county

A county with RUCC 8 or 9 — no urban cluster of 2,500+ people. Split into RUCC 8 (adjacent to a metro) and RUCC 9 (not adjacent).

Poverty rate

Share of residents with income in the past 12 months below the federal poverty threshold, as reported in ACS Subject Table S1701. Stored as a 0–1 fraction throughout the pipeline.

Rural hunger risk score

A 0–1 composite combining the ACS poverty rate, the ERS persistent-poverty flag, and a RUCC-based rural gradient. Labeled throughout the atlas as a visualization and ranking aid, not a direct measurement.

Persistent poverty — the 30-year test

The persistent-poverty designation comes from USDA ERS and has been part of congressional appropriations language since the 10-20-30 provision was first applied in 2009. The 2025 typology update shifted the reference vintages to use ACS 5-year estimates alongside the Decennial Censuses, so the current definition is:

poverty_rate_1990 >= 0.20
 AND poverty_rate_2000 >= 0.20
 AND poverty_rate_2007_2011_ACS >= 0.20
 AND poverty_rate_2017_2021_ACS >= 0.20

A county only carries the flag if every reference period crossed the 20% threshold. Spikes or single-period exceedances do not qualify. The effect is that a persistent-poverty county is one where 20%-or-higher poverty has been a stable structural feature for roughly 30 years. In the 2025 typology, 318 counties meet the test.

A related persistent child poverty flag in the typology uses the same four-period test applied to the under-18 poverty rate. We carry this flag through the pipeline but surface it only at the state-rollup level.

Rural-Urban Continuum Codes, 2023 Edition

RUCC classifies every county on a 1–9 scale. The full definitions, taken from USDA ERS:

We derive is_metro from RUCC 1–3 as the canonical rule. For counties missing a RUCC value we fall back to the ERS Metro2023 boolean attribute carried in the same file.

Additional typology flags

Beyond persistent poverty, the 2025 County Typology Codes carry several demographic-stress indicators. We preserve each as a boolean on every county:

• Low employment — share of prime-age adults who are employed falls materially below the national average across a multi-year window.
• Low education (post-secondary) — below-average share of adults with any post-secondary degree.
• Population loss — county population declined in both 2000→2010 and 2010→2020 decennial measurements.
• Housing stress — above-average share of households cost-burdened or inadequately housed.
• Retirement destination — net in-migration of residents age 60+ above the national average.

On each state page, any-stress-flag county count aggregates counties that carry one or more of these indicators. This is a surface-level screen, not a causal claim — a single county can carry several flags and many flagged counties also carry the persistent-poverty designation.

Data sources

All federal sources are public and keyless. Citations with access dates are in the Sources section below.

The rural hunger risk score

To make the county choropleth legible in a single view, we compute a composite 0–1 score per county. The formula is:

rural_gradient = (rucc_code - 1) / 8         // 0 for metro 1M+, 1 for completely-rural-nonadjacent

rural_hunger_risk_score =
    0.40 * poverty_rate
  + 0.30 * (is_persistent_poverty ? 1 : 0)
  + 0.30 * rural_gradient

The three weights reflect an editorial choice: current poverty carries the most weight because it is a direct measurement; the persistent-poverty flag captures structural duration; the RUCC gradient captures geographic remoteness, which is a proxy for how thin food retail and transport infrastructure tend to be. Counties missing any of the three inputs have rural_hunger_risk_score = null and are rendered in gray on the map.

This score is a visualization and ranking aid, not a direct measurement. Do not cite it as a poverty or food-insecurity rate. For program design or funding decisions, always cite the underlying ACS poverty rate and ERS typology flags directly.

Pantry density

Every state page shows a pantries per 100K residents figure. The numerator is the number of rows in PantryPath's pantries table that (a) have is_pantry_confirmed = true and (b) map to the state through pantries_states. The denominator is the state's ACS 2019–2023 5-year total population.

pantries_per_100k = pantries_count / total_population * 100000

Because pantries self-report inclusion in PantryPath's directory, this metric is a minimum-density indicator rather than a saturated census. It is useful for comparing states on the coverage we have confirmed, not for ruling out the existence of pantries in any area.

County-level pantry density is not yet published because the vast majority of rows lack county-level geocoding. Improving county resolution is on the roadmap; until then county-by-county pantry counts should not be inferred from this atlas.

State and national rollups

• State aggregates are computed by summing the county-level rows within each state. persistent_poverty_county_count is a simple count of counties where the flag is true; nonmetro_county_count counts RUCC 4–9.
• State poverty rate comes directly from ACS S1701 at the state geography — not re-summed from county rates. This ensures the state headline matches the Census-published figure.
• Population living in PPCs sums the ACS total population of each flagged county in the state.
• Mean rural hunger risk score is an unweighted average across the state's counties — every county counts equally regardless of population. The state table column uses this unweighted mean because it surfaces small rural counties that get lost in population-weighted averages.
• National aggregates sum state values. National poverty and child-poverty rates come directly from ACS S1701 at the US geography.
• Rankings surface six data-driven stories: most persistent-poverty counties (raw), highest persistent-poverty share, most nonmetro counties, highest state poverty, highest child poverty, and highest mean risk score.

Limitations

• Typology code vintage. The 2025 typology edition was released by USDA ERS in 2025 using reference-period data that includes ACS 2007–2011 and 2017–2021. Persistent-poverty status is by construction slow-moving, but a handful of counties cross thresholds between editions — always cite the edition year.
• RUCC reclassifications. RUCC definitions are tied to Metropolitan Statistical Area boundaries published by OMB. When MSAs are redrawn (roughly every decade) some counties shift categories without any change in underlying population. The 2023 edition is the current benchmark.
• ACS margin of error. County-level poverty rates in small-population counties carry non-trivial sampling error in the ACS 5-year release. We do not display MOE in the tables but absolute poverty-rate comparisons between sparsely populated counties should be interpreted conservatively.
• County equivalents. Louisiana parishes, Alaska boroughs, and Virginia independent cities are treated as counties throughout. Connecticut's planning regions replaced its counties as Census geographies effective 2023; the ACS S1701 values here use the post-2023 geography.
• Risk-score weights are editorial. The 40/30/30 weighting of poverty, persistent-poverty flag, and rural gradient reflects editorial judgment. We use the score only for visualization and ranking, never as a measurement. Readers may recompute with their own weights from the raw fields in each county's JSON.
• Pantry density is state-only. As noted under Pantry density, we do not currently publish county-level pantry counts. Rural pantry access is certainly not uniform within a state.
• Territories not included. USDA ERS typology + RUCC cover the 50 states plus DC. Puerto Rico, Guam, the U.S. Virgin Islands, American Samoa, and the Northern Mariana Islands are outside the scope of this atlas.

Reproducibility

The entire atlas is driven by one pipeline script: scripts/rural-poverty/build-counties.mjs. It (1) fetches the USDA ERS RUCC 2023 CSV and the County Typology 2025 CSV, (2) queries ACS S1701 and B01003 for every state via the Census API, (3) queries PantryPath's Supabase for verified pantry counts by state, (4) computes the county-level composite risk score, and (5) writes per-state + per-county JSON plus a national summary + rankings file. Emitted files live in src/data/rural-poverty/.

Pipeline metadata — dataset vintages, build timestamp, plan-year verification date, composite-score formula — is embedded in every emitted JSON file and surfaced on the atlas's index page. Refresh cadence: re-run the script when USDA ERS publishes a new typology edition or RUCC edition, or when a fresh ACS 5-year vintage lands (annual, usually December).