Low-Income Neighborhoods
The criteria for identifying a census tract as low income in the 2025 dataset are from the U.S. Department of Treasury’s New Markets Tax Credit (NMTC) program (Community Development Financial Institutions (CDFI) Fund, 2000)
This program defines a low-income census tract as any tract where:
- The tract’s poverty rate is 20 percent or greater;
- The tract’s median family income is less than or equal to 80 percent of the State-wide median family income; or
- The tract is in a metropolitan area and has a median family income less than or equal to 80 percent of the metropolitan area's median family income.
Urban-Rural Classification
Census tracts are classified as urban or rural based on the location of their population-weighted center points relative to 2020 Census Bureau Urban Area boundaries, where center points within an urban area boundary are urban tracts, and those outside the boundaries are considered rural. Center points (centroids, in technical terminology) were obtained from the Census Bureau’s tract-level Centers of Population dataset.
These shared definitions apply across both Supplemental Nutrition Assistance Program (SNAP)-authorized Retailer Access Map (SRAM)and Large Retailer Access Map (LRAM) to ensure consistency in classification and interpretation.
Data and Methodology – SRAM
SRAM integrates Supplemental Nutrition Assistance Program (SNAP)-authorized food retailer locations, gridded high-resolution population data (i.e., LandScan), and census tract-level socioeconomic characteristics to identify populations with limited geographic proximity to SNAP-authorized food retailers using both Euclidean-based (straight-line) and network-based (driving paths along roads) distance measures.
Data Sources and Geographic Area
SNAP-authorized Food Retailer Locations
The Food and Nutrition Administration (FNA; formerly Food and Nutrition Service) provided the complete list of food retailers authorized to accept SNAP benefits as of June 2025 from the Store Tracking and Redemption System (STARS). ERS excluded farmers markets and delivery routes because they do not necessarily operate year round or with regular hours (farmers markets) or have fixed locations (delivery routes).
The STARS data include retailer addresses and geographic coordinates. For SRAM, retailer addresses were geocoded using Environmental Systems Research Institute, Inc. (Esri)’s ArcGIS StreetMap Premium. When StreetMap Premium did not return a matching store location, the original STARS coordinates were retained.
Store coordinates were also validated by comparing the State and county listed in the STARS data to the Census boundaries in which the geocoded coordinates were located. Any geocoded point outside the expected State boundary was flagged for review. This process helped:
- Identify geocoding errors or mis‑assigned coordinates,
- Catch incorrect or incomplete address inputs,
- Improve the reliability of subsequent spatial analysis and reporting, and
- Maintain consistency across datasets that rely on accurate location information.
In addition to automated checks, manual review methods were also used to verify food retailer locations flagged during the validation process.
Edits and improvements to the original foodstore locations were made based on the previously described geocoding processes. Location issues in the original data set ranged from legitimate geocoded results differing from the latitude/longitude within the provided data as well as potential human error during initial data entry. While corrections to errors were made where possible, some could not be resolved.
Population Distribution Data
Population is not evenly distributed across census tracts, some of which cover large geographic areas. To account for within‑tract variation, population distribution was estimated using gridded population data obtained from the LandScan USA 2020 nighttime dataset, which is produced by Oak Ridge National Laboratory at 3-arcsecond (~90-meter, but varies with latitude) resolution. LandScan models ambient population using a variety of data sets, such as Census enumeration data, land cover classifications, and infrastructure information. Using gridded population data allows SRAM to more accurately represent where people are located within a tract, which can then be combined with distance calculations to figure out how many people live within certain distances of SNAP‑authorized food retailers.
Demographic and Socioeconomic Data
Tract-level demographic and socioeconomic variables were acquired from two Census Bureau sources:
- 2020 Decennial Census: Total population and housing unit counts
- 2020–24 American Community Survey (ACS) 5-Year Estimates: Detailed demographic, socioeconomic, SNAP participation, and vehicle availability characteristics
Tract-level geography was selected to mitigate the effects of differential privacy in the 2020 Decennial Census and to reduce margins of error present in ACS estimates at smaller geographies. The Census Bureau's differential privacy implementation a statistical noise for disclosure avoidance, with noise injection proportionally greater at smaller geographic units where individual identification risk is higher in the Decennial Census. For the ACS, estimates at smaller geographies carry larger margins of error due to reduced sample sizes. Tract-level data balances statistical reliability with geographic detail.
Geographic Boundary Data
Geographic boundary files were obtained from the Census Bureau's Topologically Integrated Geographic Encoding and Referencing (TIGER)/Line and cartographic boundary products, including 2020 and 2024 Census tracts, 2020 Urban Areas, 2020 Core-based Statistical Areas, and U.S. States.
Calculation of Distance to Stores for Each Tract
To determine whether a tract was within a specified distance to a SNAP-authorized food store, zones with access to foodstores were generated using two distance‑based approaches: (1) Euclidean (straight‑line) distance and (2) network‑based (driving) distance. Both methods produced foodstore access polygons (areas) at four distance thresholds: 0.5, 1, 10, and 20 miles, consistent with prior versions of the ERS Food Access Research Atlas. People living outside these polygons are considered to have low access to foodstores. Overlapping routed areas within each threshold were merged to create unified Euclidean (straight-line) and network‑based (driving) access areas.
Network distances represent the driving path from where people live to the nearest SNAP‑authorized food retailer.
Euclidean Distance Methods
Circular buffer polygons were generated around each SNAP‑authorized retailer at each of the four distance thresholds. Straight-line distances were calculated using geodesic calculations, which account for the curvature of the Earth and provide more accurate measurements than planar (flat surface) calculations, especially across large geographic areas.
Network-Based Driving Distance Methods
Driving distances were calculated using Esri ArcGIS Network Analyst with StreetMap Premium Custom Roads North America 2026 (Release 1) as the network dataset. All network-based distances measure travel distance to the nearest SNAP‑authorized retailer along roadways. ArcGIS Pro’s Make Service Area Analysis Layer, Add Locations, and Solve tools were used to calculate areas reachable within each distance threshold along the road network to each retailer location.
Since the road network does not include home driveways or store parking lots, buffers were applied to the roadways to capture populations reasonably close to service areas. Specifically, a 50-meter buffer was added for the 0.5- and 1-mile thresholds and a 250-meter buffer for 10- and 20-mile thresholds. Larger buffers were applied to the longer (rural) thresholds to accommodate lower road density and larger parcel sizes typical of rural areas.
The SNAP-authorized foodstores file included 101 stores (97 in Alaska and 4 in the conterminous United States) that could not be incorporated into the drive‑distance calculations. The primary reasons were that the associated road segments were not accessible year‑round, vehicle traffic was not permitted, or the stores were located more than 1 kilometer from the road network. As a result, they are excluded from drive‑distance calculations
Population Mapping (Dasymetric Allocation)
To represent where people live within each census tract at a finer geographic scale, tract‑level population counts were distributed onto a high‑resolution grid using a dasymetric mapping approach based on LandScan USA 2020 population density patterns. This approach maps population density based on additional details about local geography, such as land cover. This process creates a detailed, gridded population surface that can later be overlaid with low food access areas to determine how many people fall within each distance threshold.
The dasymetric allocation process consisted of three steps:
- Tract-level LandScan summation: Zonal statistics were calculated to sum LandScan population values within each census tract boundary, producing tract-level population totals derived from the gridded data.
- Cell-level proportion calculation: For each raster (a grid of small square cells) cell, a population proportion was calculated representing the ratio of the cell's LandScan population value to the total LandScan population within its parent tract.
- Raster output: Tract-level census population and housing variables were then distributed to raster grids to produce gridded census variable datasets.
Tract-level grid data were aligned to LandScan rasters so they matched in coverage, cell size, and grid structure.
This dasymetric approach relies on two key assumptions:
- Population distribution accuracy: LandScan accurately represents within-tract population distribution patterns.
- Demographic uniformity: The spatial distribution of demographic subgroups within each tract mirrors the overall population distribution represented in LandScan. This assumption may not hold in tracts with spatially clustered demographic patterns.
Low Food Access Area Identification
After the food access zones were created (using both straight line and road network driving distances), these zones were combined with the gridded population data to identify where people do and do not have nearby geographic access to SNAP authorized food retailers.
Each LandScan grid cell contains an estimated number of people or homes and has a known distance to the nearest retailer. Grid cells were compared with the food access zones:
- Cells inside an access zone (within 0.5, 1, 10, or 20 miles of a store) were marked as “having access.”
- Cells outside all access zones were marked as “low access.”
This process was repeated for each of the three regions (the conterminous United States (CONUS), Alaska, and Hawaii), each distance method (straight-line and driving), and each distance threshold. A fourth region, Alaska-East, was not processed because it had no SNAP-authorized retailers, meaning the entire region is classified as low‑access.
Population Summary and Tract-Level Aggregation
Finally, low-access population and housing counts from the grid cells were aggregated for each census tract. A tract’s low‑access values reflect the number of people or homes within grid cells that fall outside the access zones.
When LandScan showed no population in a tract but Census population or housing data showed at least one person, for example, on military bases or university campuses, low-access was calculated as the share of the tract’s total area that was classified as low‑access. This process was repeated for each unique combination of census variable, distance method (either Euclidean or network), distance threshold, and region.
Tract-level outputs from this processing step were used in subsequent data processing to apply tract-level access classifications.