FP dataset validation: precision and recall assessment
We created a national dataset of 34,475 FPs. The dataset was evaluated for precision and recall by two independent evaluators who were blinded to the study. Using the two-stage precision validation, we evaluated all entries for correctness and operational status. During stage one, automated information retrieval identified 49,637 entities associated with the 34,475 candidate FP locations, followed by manual verification (stage two). Our results demonstrated a precision of 91.6%, where 31,593 out of 34,475 were valid. The remaining 2,882 entries were determined to be invalid, either because the organization no longer existed, had closed or discontinued food assistance services, or was misclassified as a FP. To further ensure accuracy, we assessed inter-rater agreement in a random subsample, which demonstrated 93% agreement betweenevaluators, the remaining 7% of discrepancies reflected differences in organizational naming at the same verified FP address rather than disagreement about FP status.
For recall, each evaluator created a gold standard dataset of 150 FPs. For the first evaluator, 121 matches were identified within our FP dataset, yielding a recall of 0.807. For the second evaluator, 114 matches were identified within our FP dataset, yielding a recall of 0.760. After combining both evaluators’ datasets and eliminating duplicates, the recall was 0.758, corresponding to 179 matches out of 236 unique entities.
We conducted an error analysis of the unmatched entities, as detailed in Supplementary Fig. 1. From the first evaluator’s dataset, 23 were food banks, which typically distribute to FPs rather than to individuals, two provided drive-thru services, one was a mobile pantry, and the remaining three were typical FPs that were missed by our dataset. From the second evaluator’s dataset, 16 were food banks, seven had religious affiliations, one catered primarily to college students, one initiated operations after we built the FP dataset, one had recently changed its location, and the remaining 10 were typical FPs that were missed by our dataset. Excluding food banks, the recall improved to 0.952 and 0.851 for the two gold standards.
Accessibility analysis: urban and rural perspectives
Table 1a and b detail the FP accessibility statistics and characteristics for the 198,767 BGs in urban areas and 41,013 BGs in rural areas, respectively. For each BG, we categorized FP accessibility as high, medium, or low access based on walking/public transit time to the nearest FP in urban areas and driving distance to the nearest FP in rural areas. Detailed accessibility criteria can be found in the Methods section.
Table 1 Summary statistics of FP accessibility and characteristics of BGs in (a) urban and (b) rural areas in the U.S.
Table 1a shows that in urban areas 97,639 (49.1%), 48,414 (24.4%), and 52,714 (26.5%) of BGs have high, medium, and low access, respectively, to their nearest FPs. Table 1b shows that in rural areas 29,482 (71.9%), 8,029 (19.6%), and 3,502 (8.5%) of BGs have high, medium, and low access, respectively, to their nearest FPs. On average, BGs in rural areas are more socioeconomically disadvantaged than BGs in urban areas (ADI 71.3 vs. 45.8, respectively).
Figure 1a shows the distribution of BGs by ADI in urban areas, organized by FP accessibility category. For BGs with medium and low access to FPs, the distribution is skewed toward lower ADIs. For BGs with high access to FPs, there is an approximately U-shaped distribution. Figure 1b shows the corresponding distributions for BGs in rural areas. The distribution of BGs is heavily skewed toward higher ADIs across all three FP access levels, indicating significant socioeconomic vulnerability. In rural areas, over 25% of BGs with low access to FPs have an ADI between 90 and 100 whereas only 4% of BGs with low FP access fall within this ADI range in urban areas.
Fig. 1
The distribution of BGs by ADI, organized by FP accessibility category in (a) urban and (b) rural areas. Note that the BG percentages within the same FP accessibility category (i.e., same-colored bars) sum to 100%.
Figure 2 shows the distribution of BGs by FP accessibility category, organized by ADI in urban and rural areas. In both urban and rural areas, high-access BGs constitute the largest proportion within each ADI range, but they have different trends. In urban areas, the proportion of BGs with high access to FPs is U-shaped. The opposite pattern is seen for BGs with low access to FPs. This indicates that the highest occurrence of BGs with low access to FPs is in areas of lower socioeconomic vulnerability. The proportion of BGs with medium access to FPs gradually decreases at the higher ADI ranges. In rural areas, the proportion of BGs with high access to FPs remains relatively stable across all ADI ranges, with a subtle decreasing trend from lower to higher ADI. Low-access BGs make up a small proportion overall but show a slight increase in both the mid-range (31–70) and the highest ADI range(91–100), suggesting that even in rural settings, areas with higher deprivation may face greater challenges in accessing FPs.
Fig. 2
The distribution of BGs by FP accessibility category, organized by ADI in urban and rural areas.
Table 2a, b present state-level FP accessibility in urban and rural areas, respectively. In urban areas (Table 2a), the five states with the highest FP access were New York (69.1% of BGs with high access), Hawaii, Rhode Island, Oregon, and Illinois. States with the lowest access were concentrated in the South, including West Virginia (52.3% of BGs with low access), Mississippi, South Carolina, Alabama, and Louisiana. Across states with low FP access, mean ADI values for low-access BGs consistently exceeded those in high-access states, indicating that urban areas with limited FP access tend to have greater socioeconomic vulnerability.
Table 2 Summary statistics of top-five states with high and low access to FPs in (a) urban and (b) rural areas.
In rural areas (Table 2b), four of the five highest-access states were in the Northeast (Connecticut, Massachusetts, New Hampshire, and Vermont), with Connecticut leading at 98.3% of BGs with high access. States with the lowest access were concentrated in the West (Alaska, Arizona, Nevada, and California). Alaska had the highest proportion of low-access BGs (46.1%), with a mean ADI (66.1) substantially higher than the state mean (49.1). Mean ADI values for low-access BGs exceeded state averages across all low-access states, indicating an association between limited FP access and higher socioeconomic vulnerability. For detailed state-level data, see Fig. 3 and Supplementary Figs. 2–7.
Fig. 3
Nationwide accessibility to FPs at the BG level. (Map generated by Python 3.12).
Table 3a, b summarize demographic characteristics by FP accessibility in urban and rural areas, respectively. In urban areas (Table 3a), high-access BGs had younger populations (22.8% aged 18– < 30 vs. 16.5% in low-access areas), greater racial/ethnic diversity, including lower proportions of non-Hispanic White residents (48.6% vs. 72.9%), and higher proportions of non-Hispanic Black (17.9% vs. 7.6%) and Hispanic/Latino (22.6% vs. 11.8%) residents. This pattern suggests FP placement may align with communities experiencing greater food insecurity risk, given that Hispanic and Black populations are disproportionately affected in the U.S. Rural areas showed different patterns (Table 3b). While high-access BGs were also younger (18.8% aged 18– < 30 vs. 16.3% in low-access areas), low-access BGs in rural areas were more racially diverse, with higher proportions of Hispanic/Latino (11.3% vs. 8.3% overall) and non-Hispanic Native American residents (7.2% vs. 1.8% overall). These findings indicate that in rural areas, limited FP access is more commonly experienced in older and more racially diverse communities.
Table 3 Summary statistics of FP accessibility and demographic characteristics of BGs in (a) urban and (b) rural areas in the U.S.
Tables 4a, b and 5a, b present education, employment, household composition, income, poverty status, and public assistance data stratified by FP accessibility. In urban areas (Tables 4a and 5a), high-access BGs demonstrated indicators of greater socioeconomic vulnerability: lower educational attainment (13.8% without high school diploma vs. 8.6% in low-access areas), higher unemployment (4.2% vs. 2.8% in low-access areas), more single-parent households (13.3% single female householders with children vs. 6.6% in low-access areas), lower incomes (41.9% earning < $50,000 vs. 29.6% in low-access areas), higher poverty rates (16.0% vs. 8.6%), and greater public assistance participation (18.6% vs. 8.6%). This pattern indicates that FPs in urban areas are more concentrated in socioeconomically vulnerable communities.
Table 4 Summary statistics of education attainment, labor force status, and household composition of BGs stratified by FP accessibility in (a) urban and (b) rural areas in the U.S.Table 5 Summary statistics of household income, poverty status, and public assistance stratified by food pantry (FP) accessibility in (a) urban and (b) rural areas in the U.S.
Rural areas (Tables 4b and 5b) showed more complex patterns. While high-access BGs in rural areas had higher educational attainment (21.6% with bachelor’s degree or higher vs. 18.3% in low-access areas) and employment rates (54.6% vs. 49.5%), differences in household income, poverty rates, and public assistance participation were modest across access levels. For instance, poverty rates ranged from 14.7% in medium-access areas to 17.3% in low-access areas, compared to 15.7% overall. These findings suggest that in rural areas, FP placement is not as tightly linked to socioeconomic indicators as in urban areas, and limited access may reflect geographic barriers rather than strategic placement based on community need.
Statistical approaches
We used generalized linear regression models to examine associations between BG characteristics and access to FPs, measured by log-transformed travel time (urban) or driving distance (rural). Full regression results are provided in Supplementary Tables 1 to 10.
Across both urban and rural areas, several consistent patterns were observed. FPs were more accessible in BGs with greater socioeconomic vulnerability, as indicated by higher ADI, younger age composition (18– < 40 years), higher proportions of non-Hispanic Black and Hispanic/Latino residents, lower household incomes, higher poverty rates, and greater public assistance participation. These associations persisted after adjustment for land area, suggesting that FP placement is broadly aligned with community need.
Important differences were observed between urban and rural areas in educational attainment, employment, and minority populations. In urban areas, FPs were more accessible in BGs with lower educational attainment, higherunemployment, and higher proportions of single-parent households, especially those headed by single females. These patterns are consistent with targeted placement in high-need communities. In rural areas, these relationships were weaker or in the opposite direction. Higher educational attainment and greater employment were associated with shorter travel distances, indicating that rural FP placement may be influenced more by geographic and infrastructure constraints than by need-based targeting. In both urban and rural areas, BGs with higher proportions of individuals not in the labor force had poorer FP access. Associations with Native American populations were small, with slightly longer travel times in urban areas and no significant differences in rural areas.
Gender composition also showed consistent patterns. BGs with higher proportions of female residents had better FP access, whereas higher male composition was associated with longer travel times or distances. Although effect sizes were generally small, likely reflecting the large national sample, the overall findings suggest stronger alignment between FP placement and socioeconomic vulnerability in urban areas than in rural areas.
