Defending a Food-Access Map at a Public Hearing
A county health department maps food access with 2019–2023 American Community Survey five-year estimates for 46 census tracts. Staff label a tract a 'food desert' if its population-weighted centroid is more than 1.6 km from a full-service grocery, measured as straight-line distance. Result: 12 tracts and 22,400 residents qualify. One qualifying tract's new store is 1.7 km from the centroid but 0.6 km along a lit pedestrian path from its densest apartment cluster. Another tract's only store closed in 2023, after the ACS period. A third qualifies only because a large, sparsely populated industrial fringe pulls the centroid away from its residential core. The county has $600,000 in healthy-food grants and must defend the map at a public hearing.
1. Identify the methodological problems in the current analysis and propose a defensible alternative.
Read the worked answer
Problems:
- Centroid fallacy. A population-weighted centroid is not where people live. The industrial-fringe tract qualifies only because its centroid is pulled away from its residential core.
- Modifiable Areal Unit Problem (MAUP). Tract boundaries are administrative; a tract can contain both served and unserved blocks, and results change if block groups are used.
- Straight-line distance. Euclidean buffers ignore rivers, rail lines, and pedestrian paths. The tract 0.6 km from a store by path is wrongly classified.
- Edge effects. Residents near a tract boundary may shop in an adjacent tract or county, and stores just outside the study area are missed.
- Data vintage. Store openings and closures after the ACS period are invisible.
Defensible alternative: use block-group or parcel-level population; compute network distances on the pedestrian and street network; set thresholds by urban form (for example 0.8 km walking, 1.6 km in low-density areas); run sensitivity checks with alternative thresholds and aggregation levels; and verify the store list against business licenses and SNAP retailer data.
2. How should the county treat the three disputed tracts, and what general rule does this illustrate?
Read the worked answer
Tract 1 - reclassify as served. The 0.6 km path distance from the densest housing is the relevant measure; the 1.7 km centroid distance is an artifact. Still check path quality, lighting, and crossing safety, because a lit path is not automatically a safe one.
Tract 2 - reclassify as qualifying. The only store closed in 2023, after the ACS period; a stale store list is the error. Confirm with a business license check and a site visit, then update the layer.
Tract 3 - deprioritize. Residents are near a store, and the industrial fringe is pulling the centroid. Recompute using block-group centroids, or better, the share of residents beyond the threshold.
The general rule: classify places by conditions where people actually live, not by a single geometric point. Publish the revised classifications with reasons so the hearing record shows the method was corrected rather than manipulated.
3. Recommend how to allocate the $600,000 and how to communicate uncertainty at the hearing.
Read the worked answer
Allocation. Do not split $600,000 evenly across 12 tracts. Build a composite need index with four components: share of residents beyond the network threshold, share below 200% of the poverty line, share without a vehicle, and population density. After correction, perhaps two or three tracts are genuinely high-need.
Uses: fund a mobile market or subsidized delivery in the highest-need tract, a small-store refrigeration upgrade in the second, and produce prescriptions through a clinic in the third. Reserve 15–20% for evaluation and a repeat network analysis in 12 months.
Communication. Present a sensitivity map showing which tracts are stable across methods and which flip, with ACS margins of error and the store-list date. State plainly that the first map overcounted and explain why. At a hearing, acknowledging the centroid problem and showing the corrected network-based map is more credible than defending a flawed method - and the grant decisions are then tied to a documented, reproducible procedure rather than to a single buffer.