School Hunger &
Free Lunch Atlas

Definitions

National School Lunch Program (NSLP)

Federal child-nutrition program authorizing reimbursed meals in public schools, administered by USDA FNS. Schools opt in at the local level. The CCD flag lunch_program ∈ {1, 2} indicates participation.

Community Eligibility Provision (CEP)

A NSLP option created by the Healthy, Hunger-Free Kids Act of 2010 that lets high-poverty schools and districts serve universal free meals without collecting individual applications, as long as a minimum share of students is "identified" through SNAP/TANF/Medicaid/foster care. In the 2023–24 CCD, CEP schools carry lunch_program == 2 (which also includes Provision 2 schools on the universal-free track).

Free/reduced eligibility rate

The share of enrolled students certified at ≤185% of the federal poverty level — directly reported in the CCD as free_lunch + reduced_lunch divided by enrollment. Stored as a 0–1 fraction throughout the pipeline.

Direct certification

The administrative process that enrolls children in free meals automatically because their family already receives SNAP, TANF, or Medicaid, without a separate application. Reported per school in the CCD.

School food access score

A 0–1 composite combining eligibility rate, CEP share, and NSLP share. Labeled throughout the atlas as a visualization and ranking aid, not a direct measurement.

The CCD school universe

For every state we page through the Urban Institute's /api/v1/schools/ccd/directory/2023/?fips=XX endpoint — the 2023–24 school year, school-level directory file. We filter to regular open-status schools:

school_status == 1  // open
school_type   == 1  // regular

This excludes special-education standalone schools, vocational schools, alternative schools, juvenile justice schools, and closed/merged schools from the denominators. Nationally the filter retains about 89,800 of roughly 98,000 CCD records.

The Urban Institute API uses negative sentinel values (e.g. -1, -9) to mean "missing / not applicable" on integer counts. We treat any negative value as zero at ingestion for counts (free_lunch, enrollment, etc.) and as null for categorical flags. This avoids overstating totals when a single school is suppressed, and it is the same convention used by the Urban Institute's own published dashboards.

The CEP / Provision 2 proxy

The CCD directory does not carry a dedicated boolean that says "this school participates in CEP." What it carries is a categorical lunch_program field with these values:

We treat lunch_program == 2 as the CEP proxy. This bundles CEP schools with Provision 2 schools (a smaller, older universal-free pathway) but excludes schools using standard per-student certification. Validation against FRAC's 2023–24 CEP adoption report shows the proxy tracks published state-level adoption closely: Vermont at 100%, Kentucky at 91%, Mississippi at 99%, New Mexico at 82%, Louisiana at 83% — all match FRAC-reported values. Low-adoption states (Utah 7%, Wyoming 4%, North Dakota 8%, New Hampshire 1%) also match.

Caveat. A handful of Provision 2 schools predate CEP and continue on the older pathway. For county-level reporting the distinction rarely matters — both are universal-free — but do not cite our "CEP share" figure as pure Community-Eligibility participation when engaging with FRAC's database directly.

The free/reduced eligibility rate

For every school in the filtered universe we read free_lunch, reduced_price_lunch, direct_certification, and enrollment. The free+reduced count is the sum of the first two fields. The county eligibility rate is:

eligibility_rate = sum(free_lunch + reduced_lunch) / sum(enrollment)

summed across every regular open public school whose county-FIPS matches the county. Schools with zero enrollment, or with suppressed certification fields, contribute zero to the numerator and nothing to the denominator if enrollment is also zero — i.e., they drop out of both.

Why some states read 0% free/reduced

Seven states show a 0% free/reduced eligibility rate in the 2023–24 CCD: Arizona, Delaware, the District of Columbia, Massachusetts, North Carolina, Tennessee, and West Virginia. This is not a data error on our side — it is a CCD reporting artifact.

When a state operates universal-free meal policies statewide or relies on CEP for the vast majority of its schools, individual students are not certified under standard per-student NSLP rules. The certification counts per school become administratively meaningless (the school feeds everyone without checking income) and states increasingly report them as zero or leave the fields blank. We carry the zero forward rather than imputing because imputation would hide the underlying policy reality.

On state pages for the affected states, a banner explains this clearly. The atlas's other metrics — NSLP school count, CEP share, enrollment, SAIPE school-age poverty — remain reliable in those states, and the composite access score weights them so the score does not degrade to zero.

School → county aggregation

Each CCD school record carries a county_code field — a 5-digit county FIPS assigned by NCES based on the school's physical address. We group by that field directly, so no district-to-county crosswalk is needed. This is the biggest methodological win of using the CCD school universe instead of district-level state DOE files: a school district that spans three counties gets decomposed naturally into three county bucketings.

Connecticut's 2023 shift from counties to planning regions as Census geographies is carried through by the CCD — we use the planning-region FIPS that the file already assigns.

SAIPE school-age poverty (5–17)

The Census Bureau's Small Area Income and Poverty Estimates program publishes an annual county-level estimate of the number of children aged 5–17 in families in poverty. We use the 2023 vintage, pulled from the Census timeseries API:

https://api.census.gov/data/timeseries/poverty/saipe
  ?get=SAEPOV5_17R_PT,SAEPOVRT5_17R_PT
  &for=county:*&in=state:XX&time=2023

SAEPOV5_17R_PT is the count of related children 5–17 in families in poverty; SAEPOVRT5_17R_PT is the rate (reported as a percent by Census, which we convert to a 0–1 fraction at ingest).

SAIPE is not the NSLP eligibility population (NSLP uses ≤185% FPL; SAIPE uses ≤100% FPL for school-age children specifically). We carry it as context, not as a benchmark. Counties with high SAIPE poverty and low NSLP eligibility are candidates for directed outreach — but the two rates should never be compared one-to-one.

The school food access composite

Three county-level inputs compress into a single 0–1 visualization aid:

school_food_access_score =
    0.50 * eligibility_rate
  + 0.30 * cep_participation_share
  + 0.20 * nslp_participation_share

The weights reflect an editorial choice: certification eligibility is the closest-to-ground-truth "who qualifies" measure so it carries the most weight; CEP share captures how much of a county's NSLP footprint is on universal free meals (a direct access amplifier); and NSLP share itself captures whether children even have meals served in their schools. Counties with suppressed per-student data still produce a meaningful score because the CEP and NSLP inputs stay accurate.

This score is a visualization and ranking aid, not a direct measurement. Do not cite it as a program-participation rate. For program design or funding decisions, cite the underlying certification and CEP figures directly.

State and national rollups

• State aggregates sum county-level counts (schools, enrollment, free, reduced, direct-cert). Shares are recomputed at the state level — not averaged across counties.
• SAIPE school-age poverty at the state geography is summed from county SAIPE counts; the state rate is saipe_school_age_poor / saipe_school_age_5_17_population.
• National aggregates sum the 51 state rollups. National shares are recomputed from national counts.
• Rankings on the national hub surface six data-driven stories: highest eligibility, highest CEP share, highest enrollment, highest direct certification, highest access score, and the alphabetical full-state list for directory-style navigation.

Limitations

• CCD suppression. Free/reduced certification counts are state-reported and subject to state policy on whether per-student counts are published. Seven states report zero for SY 2023–24 (see the Why some states read 0% section above). State pages for those states carry a banner explaining this.
• CEP vs. Provision 2 bundling. Our CEP proxy includes a small number of Provision 2 schools. For precise CEP reporting, cite FRAC's annual database directly.
• Private and BIE schools excluded. The CCD is a public-school universe. Private schools (even those participating in NSLP) and Bureau of Indian Education schools are not included. This undercounts NSLP participation in states with material private-school NSLP participation (e.g. Louisiana, Pennsylvania).
• SAIPE vs. NSLP. SAIPE's school-age-poverty rate is a ≤100% FPL measure; NSLP's free/reduced eligibility is ≤185% FPL. The two metrics answer different questions. We surface both as complementary views of child need, not as interchangeable measures.
• School-year alignment. CCD is SY 2023–24. SAIPE is calendar 2023. FRAC CEP adoption covers 2023–24. The atlas as published in April 2026 is therefore one complete school year behind the latest lived reality — the inherent lag of federal education microdata.
• Territories not included. Puerto Rico, Guam, the U.S. Virgin Islands, American Samoa, and the Northern Mariana Islands are outside the scope of this atlas. NSLP operates in territories under separate FNS agreements.

Reproducibility

The entire atlas is driven by one pipeline script: scripts/school-hunger/build-counties.mjs. It (1) pages through the Urban Institute Education Data Portal for all 51 state FIPS to fetch the 2023–24 CCD school universe, (2) filters to regular open public schools, (3) groups by county FIPS and sums counts, (4) fetches Census SAIPE 2023 school-age poverty per county, (5) fetches ACS B01003 for county names, (6) computes the school-food-access composite, and (7) writes per-state + per-county JSON plus a national summary + rankings file. Outputs live under src/data/school-hunger/.

Pipeline metadata — dataset vintages, build timestamp, composite weights — is emitted into every JSON file and surfaced via the sources section below. Refresh cadence: re-run the script annually after NCES publishes the next CCD vintage (typically November–January) and after Census publishes the next SAIPE release (typically December).