FIPI · FOOD INTERVENTION PRIORITY INDEX
30-day independent research project · PostgreSQL · Tableau Public

Food deserts don't fail people one dimension at a time. They compound.

FIPI is a SQL-derived composite index that ranks U.S. counties by cumulative food access vulnerability — merging four federal datasets on a single FIPS key so the counties failing on every front at once are named, not buried in four separate spreadsheets.

"If you could open only 10 new food pantries in America tomorrow, which 10 counties would save the most lives — and why?"
— the question this project was built to answer
72,531
census tracts scored, 51 states + D.C.
9,293
tracts flagged as food deserts (12.8%)
1,054
counties scored for SNAP effectiveness
80
counties rated "Critical — SNAP Failing"
The Gap

The federal government mapped where hunger exists. Nobody ranked it.

Approximately 19.5 million Americans live in low-income, low-access census tracts — the nearest grocery store is over a mile away in cities, or ten miles in rural counties. Diabetes prevalence in high-food-desert counties runs 25–30% above counties with full food access. Rural food desert counties have median incomes more than $15,000 below the national median.

The data to see this already exists — USDA, Census, County Health Rankings, and SNAP administrative records each describe part of the picture. What doesn't exist is a single, reproducible ranking that says which counties are failing on every dimension at once, and by how much.

FIPI treats that as a data-integration and scoring problem, not a mapping problem: four datasets, one FIPS key, one weighted composite score, ranked.

The SNAP gap is invisible

A county with high poverty but low SNAP uptake has a safety net that exists on paper but fails in practice — 45% of the 1,054 counties scored here fall into "Severe" or "Critical" SNAP-strand territory.

Description without prioritization

Existing tools identify that a tract is underserved. None weight the compounding effect of low access + low SNAP reach + high health burden happening in the same county simultaneously.

Fragmentation across four agencies

USDA ERS, USDA FNS, RWJF/County Health Rankings, and the Census Bureau each publish independently. No unified county-level view connects all four for resource-allocation decisions.

Data Architecture

Four federal datasets. One join key.

Every dataset below is joined on a 5-digit county FIPS code — eliminating name-matching failures like "St. Louis" vs. "Saint Louis" — except SNAP administrative data, published at the state level.

01 / FOOD ACCESS

USDA Food Access Research Atlas

USDA Economic Research Service

Core food desert flag (1mi urban / 10mi rural), low-access population, poverty rate, median family income, vehicle access — at the census tract level.

GRAIN: census tract · 72,531 rows
02 / SNAP PARTICIPATION

SNAP Households, Persons & Benefits

USDA Food & Nutrition Service

Enrollment counts, average monthly benefit. Combined with ACS eligibility estimates to engineer the SNAP participation gap.

GRAIN: state, monthly · 51 states + D.C.
03 / HEALTH OUTCOMES

County Health Rankings

Robert Wood Johnson Foundation

Diabetes prevalence, adult obesity rate, food insecurity rate — the health-consequence dimension that turns geography into a public health argument.

GRAIN: county · 3,142 rows
04 / POVERTY & DEMOGRAPHICS

ACS 5-Year Estimates (S1701)

U.S. Census Bureau

County poverty rate, median household income, SNAP receipt %, no-vehicle % — the eligibility side of the participation gap.

GRAIN: county · 3,142 rows
Join key: county_fips — 5-digit standardized identifier present in all four sources
Methodology

Compounding vulnerability is categorically different from single-dimension poverty.

A food desert is concerning. A food desert with low SNAP uptake, above-average diabetes rates, and no nearby pantry is a compounding systemic failure — and needs to be named as one.

Food Access Score30%
% of a county's tracts flagged low-income/low-access (LILATracts_1And10), normalized 0–100.
Poverty Score30%
Average tract-level poverty rate for the county, normalized 0–100.
SNAP Gap Score25%
Share of the county's SNAP recipients (TractSNAP) who live in a flagged food-desert tract.
Vehicle Access Score15%
Low-access population with no vehicle (lahunv1), normalized 0–100 — the transportation barrier.
10_final_FIPI_scores.sql — CTEs + window functions
final_fipi_score = (access_score  × 0.30) +
                   (poverty_score × 0.30) +
                   (snap_score    × 0.25) +
                   (vehicle_score × 0.15)

-- each dimension min-max normalized first:
ROUND((value - MIN(value) OVER()) /
  NULLIF(MAX(value) OVER() - MIN(value) OVER(), 0)
  * 100, 1)

-- counties with fewer than 10 tracts are dropped
-- (too small to score reliably):
HAVING COUNT(*) >= 10 AND SUM(lilatracts_1and10) > 0
Priority Ledger

The 25 highest-priority counties in America, ranked.

Every row below is a real scored county from final_fipi_scores.csv. Click any column header to re-sort.

↕ sortable · showing all 25 scored counties
Rank County State % Desert Tracts Poverty Rate SNAP Strand Rate People Affected FIPI Score
State Vulnerability

Scale vs. concentration aren't the same map.

SNAP Participation Gap

1,054 counties, scored by how well SNAP actually reaches people.

Statistical Validation

Does food access actually predict health outcomes — or is the link assumed?

County-level food desert concentration was correlated against County Health Rankings outcomes across 3,142 counties using Pearson's r (scipy.stats.pearsonr).

Diabetes Prevalence
r = 0.095
Pearson correlation vs. % desert tracts
p = 1.04 × 10⁻⁷  ✓ significant at p < 0.001
Adult Obesity Rate
r = 0.146
Pearson correlation vs. % desert tracts
p = 1.69 × 10⁻¹⁶  ✓ significant at p < 0.001

Both relationships are statistically significant at p < 0.001 across n = 3,142 counties, confirming the model differentiates high-need from low-need counties on a real downstream outcome — not just on the variables that define the index itself. Coefficients are modest in size, which is expected: food access is one of several drivers of chronic disease, which is exactly why FIPI weights health burden as one of four dimensions rather than the sole criterion.

Repository

Every query, every output, fully reproducible.

Full 12-file SQL progression (including both exploratory passes at days 06 and 07) lives in /sql. Data dictionary and executive insights are in /docs and /outputs.

Interactive Tableau Public dashboard

State and county-level maps, filterable by FIPI score, SNAP strand rate, and health burden.

View Live Tableau Dashboard →