Household Poverty Accounts for Most Neighborhood Food-Access Variation Compared With Store Distance
We compare local poverty and SNAP rates with store distance across city, suburb, and rural census tracts.
Local health agencies map census tracts to decide where to send mobile food vans and grocery outreach. Some neighborhood maps focus on distance to the nearest supermarket, while others weight household poverty and CalFresh enrollment. Comparing poverty rates with store distance across suburban and rural tracts guides where each tool fits.
Across 408 Santa Clara County tracts, household poverty accounts for 62.2% of the 5-part food index (58.1 percentage points of unique explained spread, vs. 0.003 points from distance, p = 0.827) because poverty and SNAP hold 55% of index weight and 93.4% of tracts sit within 1 mile.[1][2] Distance maps alone flag suburban foothills (4.1% of spread) where 26 of 27 tracts past 1 mile have under 20% poverty.[2][3] Pairing poverty with trip miles separates urban outreach from rural and valley gaps in Fresno (30 of 225 tracts, 13.3%).[1][3]
Table 1. Store Trips, Local Poverty & Car Access Across Four California Areas
We compare store distance, local poverty, car access, and composite index spread across four California areas.
| Area & Index Measure |
Group 1
Campus Renter Tract
|
Group 2
High-Pay Car Foothills
|
Group 3
Dense Suburb County
|
Group 4
Rural & Valley Counties
|
|---|---|---|---|---|
| Trip to Nearest SNAP Store | 1.218 miles>1.0 mi city line | 2.970 mileslow-density hills | 0.573 mi mean0.409 mi median | Nevada 2.92 miFresno 1.45 mi |
| Tract Poverty & Pay Ratio | 22.1% poverty≥20% USDA line | 3.7% poverty2,640 residents | 532% FPL county mean pay | Fresno 19.1% povNevada 10.5% |
| SNAP Use & Car Access | 5.3% SNAP | 4.8% no car52 HH | Low SNAPHigh car ownership | 5.5% SNAP | 93.4% ≤1 mi381/408 | 30 of 225 Fresno tracts low-access |
| USDA Low-Access Pattern | Only 1 of 408 tracts0.2% pop | 26 of 27 >1 mi tracts <20% pov | Distance alone R² = 4.1%minus | Max 12.43 mi (Nev) / 15.53 mi (Fre) |
| Share of Food Index Explained | +58.1 pp from poverty60.6% partial | +0.003 pp from distance0.009% partial | Poverty slope 132x larger in index | 13.3% Fresno USDA food deserts |
| Standardized Index Slope | +0.787 slope from poverty | -0.006 slope from distance | 62.2% full-model index R² | Nevada median trip 1.98 miles |
Table 1 Notes & * Modeled Estimates
Expand notes
Table 1 Notes (Mode B Observed Tract & County Data): We use observed U.S. Census ACS 5-Year tract estimates and USDA SNAP Retailer Locator coordinates across Santa Clara County (N = 408 tracts, 514 SNAP stores, 755 DEH permits), Fresno County (N = 225 tracts), and Nevada County (N = 26 tracts; county_rankings_ca_2023.csv). In Santa Clara County (regression_model_statistics.csv and regression_results_poverty_distance.csv, White HC3), the outcome is the 5-component Extended Vulnerability Index (0-1 scale: 30% poverty rate, 25% SNAP rate, 20% normalized store distance, 15% no-vehicle share, 10% inverse density). Full-model R² = 62.17% (poverty alone R² = 62.17%, distance alone R² = 4.11%). Unique R² gain (semi-partial ΔR², labeled Partial R² in the saved CSV) is +58.06 pp for poverty and +0.0033 pp for distance; true partial R² is 60.55% for poverty and 0.0088% for distance. Poverty β_std = +0.787 (t = 12.51, p < 0.0001) vs. distance β_std = -0.006 (t = -0.22, p = 0.827), VIF = 1.59.
Column Scope. Campus Renter Tract. Mountain View Tract 5116.08 (1 of 408 county tracts, 3,252 residents).. High-Pay Car Foothills. Saratoga / Los Altos Hills Tract 5073.02 (26 of 27 tracts >1 mile from store).. Dense Suburb County. Santa Clara County baseline (N = 408 tracts, 514 SNAP stores / 755 permits).. Rural & Valley Counties. Fresno County (N = 225 tracts) & Nevada County (N = 26 tracts)..
Two-Ruler Benchmark Summary. Mode B Composite Vulnerability Index Split (30% Poverty, 25% SNAP, 20% Distance, 15% No-Car, 10% Density; Santa Clara N = 408 Tracts). Ruler 1: Tract Poverty Rate (30% Index Weight): 62.2% of Index Spread (+58.1 pp Unique R² Gain). Tract poverty accounts for 62.2% of the 5-part index spread (a +58.1 pp unique R² gain above distance, 60.6% partial R², and a +0.787 standardized slope) because poverty and SNAP vary widely across city tracts. Ruler 2: Store Trip Distance (Checked With Poverty): +0.003 pp Unique R² Gain (Flat -0.006 Slope, p = 0.827). Store distance alone accounts for 4.1% of index spread with a minus sign because 26 of 27 tracts past 1 mile are lower-poverty foothills. Accounting for poverty leaves a +0.003 pp unique R² gain (p = 0.827).
Method & Data Audit (TooEarlyToSay.com): Food Desert Comparison Claim Record → · Statewide Food Access Comparison Record → · Statewide Mobility Desert Estimate Record →
Table 2. Pairing Income and Trip Measures Across Suburban, Valley, and Rural Tracts
We match the screening rule to what poverty and store distance measure in each California area.
| Choice Factor | Campus Renter Tract | High-Pay Car Foothills | Dense Suburb County | Rural & Valley Counties |
|---|---|---|---|---|
| Local Area & Renter Context | High-rent (84.8%) Mountain View tract near Shoreline Park with 95.2% car access. | Saratoga and Los Altos Hills tracts with large lots and no retail zoning. | Dense suburb county where 381 of 408 tracts (93.4%) sit within 1 mile of a SNAP store. | Fresno County (30 of 225 low-access tracts) and rural Nevada County (2.92 mi mean trip). |
| What the Tract Measures Show | Only 1 county tract meets both USDA cutoffs (1.218 mi and 22.1% poverty), with 5.3% SNAP use. | 26 of 27 Santa Clara tracts past 1 mile from a SNAP store have under 20% poverty. | Poverty accounts for 62.2% of the 5-part index spread while distance adds +0.003 pp unique spread. | In Fresno (max 15.53 mi) and Nevada County (max 12.43 mi), long trips and poverty overlap. |
| Risk of Distance-Only Screening | Relying on the 1-mile line alone flags 1 tract while skipping high-SNAP urban renter tracts. | Sending food vans to tracts past 1 mile without a poverty check puts 26 of 27 routes in rich hills. | Treating a composite index slope as a store-opening trial misreads an observational accounting split. | Applying dense suburban distance patterns to rural Sierra or valley tracts leaves out 12-to-15-mile trips. |
| Best Targeting & Screening Action | Check SNAP & Cars Before Siting. Check SNAP use (5.3%) and no-car homes (52 HH) alongside the 1.2-mile trip. | Pair Distance Maps With Poverty. Filter >1-mile tract lists by poverty (≥20%) and no-car shares before routing vans. | Target CalFresh in High-Poverty Tracts. Use poverty and SNAP rates to target benefit outreach in dense urban tracts. | Route Vans & EBT to Rural Gaps. Target mobile pantries and online EBT delivery where >1-10 mi trips meet high poverty. |
Table 2 Notes: Composite Index Definition, Semi-Partial vs. Partial R² & Provenance
Expand notes
• Composite Index Definition & Regression Decomposition (Mode B, N = 408 Santa Clara Tracts): In regression_poverty_vs_distance.py (regression_results_poverty_distance.csv and regression_model_statistics.csv), the dependent variable is the 5-component Extended Vulnerability Index (0-1 scale, mean 0.1960, range 0.0988 to 0.4818; extended_vulnerability_index_SNAP_VERIFIED_20251029_metadata.json), weighted 30% poverty rate (ACS S1701), 25% SNAP participation, 20% normalized distance to 514 SNAP stores, 15% no-vehicle households (ACS B25044), and 10% inverse population density. Because poverty and distance are components of this index, the OLS model (Index_0_1 = 0.1294 + 0.9920(Poverty_Rate, SE = 0.0793, t = 12.51, p < 0.0001, β_std = +0.787) - 0.0026(Distance_mi, SE = 0.0117, t = -0.22, p = 0.827, β_std = -0.006), White HC3, R² = 0.6217, F = 80.24, VIF = 1.59) is a single-year variance decomposition of the index, not a causal estimate of store openings. Poverty alone gives R² = 0.6217 (62.2%) and distance alone gives R² = 0.0411 (4.1%). The semi-partial ΔR² (R²_full - R²_reduced, labeled Partial R² in regression_model_statistics.csv) is 0.5806 (58.1 pp) for poverty and 0.000033 (0.003 pp) for distance; the true partial R² (ΔR² / (1 - R²_reduced)) is 0.6055 (60.6%) for poverty given distance and 0.000088 (0.009%) for distance given poverty. On the 0-100 index scale in PHASE6_FINAL_DELIVERABLE.md, the slopes are +0.7768 and -0.0173 (β_std ratio 0.7870 / 0.00594 = 132.5x).
• Data Provenance: Exact results come from FoodSecurity/analysis/outputs/regression_results_poverty_distance.csv, FoodSecurity/analysis/outputs/regression_model_statistics.csv, FoodSecurity/data/processed/extended_vulnerability_index_SNAP_VERIFIED_20251029_metadata.json, FoodSecurity/data/processed/statewide/county_rankings_ca_2023.csv, FoodSecurity/FOOD_DESERT_LOCATION_DETAILS.md (L1-115), and FoodSecurity/PHASE6_FINAL_DELIVERABLE.md (L1-110).
What Single-Year Tract Maps Can and Cannot Tell a County Team
A census-tract map summarizes where low-income households and grocery stores sit, but a single distance threshold can misclassify dense suburbs. In Santa Clara County (N = 408 tracts, 514 verified SNAP stores), 93.4% of tracts (381 of 408) sit within 1.0 mile of a SNAP store (mean 0.573 miles, median 0.409 miles). Of the 27 tracts beyond 1.0 mile, 26 are lower-poverty foothill areas (such as Tract 5073.02 at 2.97 miles and 3.7% poverty), while only 1 tract (Mountain View Tract 5116.08, 3,252 residents, 1.218 miles, 22.1% poverty) meets both USDA food-desert cutoffs.
Decomposing the county's 5-part vulnerability index (weighted 30% poverty, 25% SNAP, 20% store distance, 15% no-car share, and 10% density) shows that poverty accounts for 62.2% of index variance (+58.1 pp unique R² gain, 60.6% partial R²), whereas store distance adds +0.003 pp unique R² gain (p = 0.827) once poverty is included. Because this is an observational decomposition of a composite index in a store-dense county, it does not test the causal effect of building a store. Instead, it shows that suburban teams should pair poverty and SNAP rates with car access instead of screening by distance alone, while rural and valley counties with wider store gaps (such as Fresno County with 30 of 225 tracts at 13.3% USDA low-access up to 15.53 miles, and Nevada County averaging 2.92 miles up to 12.43 miles) require joint income-and-distance targeting.
Data Sources, Model Inputs, Notes & Endnotes
(Numbered endnotes [1]–[3], state data sources & math rules)
Expand details
Case Study Focus: Study Credibility & Measurement · Spatial Proxies vs. Trip Distance (Mode B 408 Santa Clara & 251 Rural/Valley Tracts)
Numbered Endnotes & Main Sources
- [1]U.S. Census Bureau: American Community Survey (ACS) 5-Year Census Tract Poverty, SNAP & Vehicle Access Estimates (N = 408 Santa Clara Tracts, N = 225 Fresno Tracts & N = 26 Nevada County Tracts).
- [2]USDA Food and Nutrition Service (FNS): SNAP Retailer Locator Geo-Verified Store Coordinates (514 Santa Clara SNAP Stores).
- [3]USDA Economic Research Service (ERS): Food Access Research Atlas Low-Income and Low-Access Tract Definitions (1-Mile Urban and 10-Mile Rural Thresholds).
Mode B Composite-Index Decomposition & County Comparisons: All regression figures (N = 408, R² = 62.17%, poverty-only R² = 62.17%, distance-only R² = 4.11%, semi-partial ΔR² = 58.06 pp vs. 0.0033 pp, true partial R² = 60.55% vs. 0.0088%, β_std = +0.7870 vs. -0.0059, p = 0.827, F = 80.24, VIF = 1.59) come from FoodSecurity/analysis/outputs/regression_model_statistics.csv and regression_results_poverty_distance.csv on the 5-component Extended Vulnerability Index (extended_vulnerability_index_SNAP_VERIFIED_20251029_metadata.json). Rural and valley comparisons (Fresno County: 30 of 225 tracts = 13.3% food deserts, mean 1.45 mi, max 15.53 mi; Nevada County: 26 tracts, mean 2.92 mi, median 1.98 mi, max 12.43 mi) come from FoodSecurity/data/processed/statewide/county_rankings_ca_2023.csv. Companion TooEarlyToSay pages (including the article-reported 12% or 1,086 of 9,039 tracts statewide mobility-desert series flagged in FoodSecurity/analysis/MOBILITY_DESERTS_STATUS.md for incomplete transit agency coverage and coordinate-projection reconciliation) are excluded from empirical claims and policy recommendations on this page.
Main Sources: We use U.S. Census Bureau American Community Survey (ACS 5-Year Tract Estimates) and USDA FNS SNAP Retailer Locator (Geo-Verified Store Coordinates). We also use USDA ERS Food Access Research Atlas.
Companion Method & Code Guides (TooEarlyToSay.com)
- The Food Desert Myth: Evidence Beyond Proximity. USDA low-income, low-access tract measures; the published zero-food-desert finding and 4.7x SNAP participation comparison are under reconciliation.
- Food Access Comparison Under Reconciliation. Statewide comparison across 9,033 tracts (1.03 versus 1.78 miles to stores by tract composition); its populations and direction conflict with a related page and are under reconciliation.
- Mobility Deserts: Transit Proximity Under Review. Article-reported 12% statewide mobility-desert rate across 9,039 tracts; the transit-stop inventory (64,060 raw versus about 24,000 unique records) is under reconciliation.
Common Questions & Method Notes
- Why did store distance show a negative single-factor R² of 4.1% if its unique R² gain after accounting for poverty is +0.003 pp (p = 0.827)?
- That pattern reflects spatial sorting across Santa Clara County's 408 tracts. Exactly 26 of the 27 tracts located more than 1.0 mile from a SNAP store are lower-poverty foothill areas (under 20% poverty). In a single-factor regression on the composite index, distance acts as a proxy for foothill affluence. Once poverty rate is included, distance adds +0.003 pp of unique R² gain (0.009% partial R², p = 0.827).
- What is the difference between semi-partial ΔR² (58.1 pp) and partial R² (60.6%), and what does the composite index measure?
- The outcome is a 0-1 composite index weighted 30% poverty, 25% SNAP rate, 20% normalized store distance, 15% no-vehicle households, and 10% inverse density. Semi-partial ΔR² (58.06 pp, labeled Partial R² in the saved summary CSV) subtracts distance-only R² (4.11%) from full-model R² (62.17%). True partial R² divides that 58.06 pp gain by the 95.89% variance left unexplained by distance alone, giving 60.55% (60.6%).
- Why was only 1 of 408 Santa Clara County tracts flagged as a USDA food desert compared with 30 of 225 tracts (13.3%) in Fresno County?
- The USDA urban food-desert rule requires BOTH ≥20% tract poverty AND >1.0 mile to a supermarket (>10 miles in rural tracts). Santa Clara County has 514 verified SNAP stores (mean distance 0.573 miles, 93.4% of tracts within 1.0 mile), so only Mountain View Tract 5116.08 (1.218 miles, 22.1% poverty, 3,252 residents) crosses both lines. Fresno County spans wider rural and agricultural tracts (mean 1.45 miles, max 15.53 miles), where 30 of 225 tracts (13.3%) cross both thresholds.