Population-level measures of perceived food access reveal barriers beyond geographic proximity
Teresa Groton, Benjamin rachunok
cs.CL
Quick verdict
Uses 25k Google reviews to score 5 perceived food-access dimensions in Raleigh via BERTopic+zero-shot NLP; promising, though validation is topic-level only.
Summary
The paper converts 25,125 Google Maps reviews of 49 grocery stores in Raleigh, NC (2010–2025) into store-level scores for five perceived food-access dimensions (availability, accessibility, affordability, accommodation, acceptability). BERTopic topic modeling retains 41 topics, mapped to dimensions by zero-shot embedding matching with 85.4% agreement against a single researcher's manual labels; star ratings are collapsed to -1/0/+1 signals, aggregated with Bayesian partial pooling, and standardized within the region. The five scores correlate weakly (|r| ≤ 0.39), and 66.8% of reviews map to at least one dimension. The authors show that tightening perceived-access thresholds shrinks residents' opportunity sets (an illustrated resident's nearest acceptable store lies 7.5 miles away), and that one identical corporate policy — Harris Teeter's 2020 end of 24-hour operation — was discussed in ~80% of accommodation reviews at one store and never at a nearby sister store. Tract-level weighted regressions show acceptability rising with income (R² = 0.415) while perceived affordability is higher in tracts with larger Black populations, partially offsetting geographic disparities. Limitations include reviewer self-selection (partially probed in supplements), scores that are only region-relative, cross-sectional confounding with retailer siting, a single-city sample, and validation confined to topic level by one coder; no code or data release is stated.
Review complete
Assessment at a glance
5
risk flags identified
● high 0
● medium 3
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Online grocery reviews can be transformed into scalable store-level measures of five perceived food-access dimensions, with automated topic-to-dimension assignment agreeing with manual labels on 85.4% of topics.
- The five perceived dimensions are largely distinct constructs (all |r| ≤ 0.39), and perceived thresholds sharply contract residents' opportunity sets beyond what geographic proximity implies (illustrated nearest acceptable store: 7.5 miles).
- Identical store policies can be perceived very differently across locations, evidenced by divergent review responses to Harris Teeter's March 4, 2020 chain-wide end of 24-hour operation (~80% vs 0% of accommodation reviews mentioning hours at two nearby stores).
- Perceived food access follows socioeconomic patterns that broadly parallel but do not replicate geographic access: acceptability rises with income (R² = 0.415) while perceived affordability is higher in tracts with larger Black populations (β = 1.640, p < 0.001), partially offsetting geographic disadvantage.
Strengths
Novel, fully specified measurement pipeline
BERTopic (UMAP 5-d/30 neighbors, HDBSCAN min cluster 120) plus zero-shot embedding matching against the established Caspi/Penchansky-Thomas five-dimension framework; eight candidate topic models compared (coherence 0.528, diversity 0.709) and all hyperparameters (τ = 0.117, δ = 0.17) reported in Methods.
Dimensional distinctiveness empirically demonstrated
All pairwise store-level correlations |r| ≤ 0.39 (accessibility–accommodation 0.39, accessibility–acceptability 0.35); only 2 of 49 stores at/above the regional mean on all five dimensions, and none above +0.20 SD everywhere — the multidimensionality claim is tested, not assumed.
Clever naturalistic contrast for the resident–environment fit thesis
Proper handling of unequal review counts plus honest bias probing
Candid limitations section
Weaknesses
Validation is topic-level, single-coder, and tuned in-sample
No sentiment modeling: dimension scores are star-rating means over topic subsets
Sparse dimensions and thin cells
Statistical reporting gaps in Table 2
Illustrative results asserted as general; novelty overstated
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper claims population-scale measurement of **perceived food access**, derived from 25,125 Google Maps reviews (2010–2025) of 49 chain grocery stores in Raleigh, NC, scored on the five Caspi/Penchansky-Thomas dimensions. Four claims: (1) the measures are valid and capture distinct constructs; (2) geographic proximity overestimates access because perceived thresholds contract residents' opportunity sets (Fig. 2: a resident with three stores within ~1 mile must travel 7.5 miles to the nearest store in the top 7% of acceptability); (3) identical corporate policies are perceived differently across locations (Fig. 3: Harris Teeter's March 4, 2020 end of 24-hour operation appears in ~80% of accommodation reviews at one store and never at a sister store <1 mile away); (4) socioeconomic gradients in perceived access parallel but do not replicate geographic ones (Table 2).
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What does this paper do?
It turns about 25,000 Google Maps reviews of 49 grocery stores in Raleigh, North Carolina into store-level scores for five perceived dimensions of food access — availability, accessibility, affordability, accommodation, and acceptability — using topic modeling (BERTopic) plus zero-shot classification, and shows these scores capture barriers that simple distance-to-store measures miss.
How accurate is the review-to-dimension classification?
The automated topic-to-dimension assignments agreed with one researcher's manual labels on 85.4% of the 41 retained topics (macro AUC 0.915, prevalence-weighted AUC 0.861). Validation was done at the topic level by a single coder, not at the level of individual reviews.
What are the main findings?
The five perceived dimensions correlate only weakly with each other; tightening perceived-access thresholds can dramatically shrink residents' store options (one illustrated resident's nearest acceptable store is 7.5 miles away); an identical chain policy — Harris Teeter's 2020 end of 24-hour operation — was mentioned in about 80% of accommodation reviews at one store and never at a nearby sister store; and perceived affordability is higher in neighborhoods with more Black residents while perceived acceptability rises with income (R² = 0.415), partially offsetting geographic disparities.
What are the limitations?
Google reviewers are self-selected; scores are relative to the local retail market only; the cross-sectional design cannot separate resident perception from retailer siting decisions; validation was topic-level by a single coder; only one city is studied; and no code or data release is stated.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figure 4 (pipeline diagram) labels are garbled by PDF extraction of the diagram, but the Methods prose fully specifies the pipeline.
- No truncation detected: the extraction ends at page 14/14 and includes all three supplement tables (Tables 4–6).
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.