The study assembled a statewide, establishment-level panel to examine how the distribution of unhealthy retail—including liquor stores, bars, convenience stores, and fast-food outlets—relates to neighborhood income and how that relationship is sustained over time in Rhode Island from 2016 to 2025.
Across nine annual cross-sections the authors followed the same outlets year to year to measure counts by census tract and to decompose how the standing income gradient arises.
The authors obtained business establishment records under license from Data Axle USA and constructed a panel spanning 2016–2025. The resulting dataset comprised 480,923 geocoded establishment-years across nine annual cross-sections. Geographic units were census tracts, and tract-level covariates were drawn from the American Community Survey and TIGER/Line boundaries; a Historic Redlining Indicator was also used.
The Data Axle business records are not publicly redistributable under the data license; the authors reported publicly available sources for census inputs and the redlining indicator.
Outlets were classified into categories that the authors describe as unhealthy retail, specifically liquor stores, bars, convenience stores, and fast-food establishments. Counts of these outlets were aggregated to the census-tract level and normalized per resident for analytic comparisons across tracts with different populations.
To estimate the tract income gradient in unhealthy outlets the authors used three complementary methods: negative binomial regression to model counts, a concentration index to summarize inequality, and a Bayesian spatial model to account for spatial dependence. These methods provided robustness checks on the presence and magnitude of the income gradient.
The authors also tested the gradient’s stability over time and decomposed the observed inequality into two dynamic components: openings (entry of new establishments) and closures (exit or survival of existing establishments).
All three analytic strategies produced consistent results: a robust income gradient in unhealthy outlets. The authors report about 30 percent fewer unhealthy outlets per resident for each standard deviation increase in tract income. This gradient was stable across the observation period and robust to alternative model specifications.
When tract income was included in models, the authors report that racial composition and poverty measures were no longer associated with unhealthy outlet counts in the adjusted analyses they describe.
Decomposing the gradient into openings and closures, the authors found that the income gradient was reproduced through openings rather than differential survival. Closures occurred relatively evenly across income levels; they did not selectively reduce unhealthy outlets in lower-income tracts.
In contrast, new unhealthy outlets opened disproportionately in lower-income tracts: the rate of new unhealthy openings per resident in the lowest-income tracts was about 2.2 times that in the highest-income tracts. Thus entry dynamics—where new establishments choose to locate—drive the persistent income gradient.
The excess burden in the standing stock of unhealthy retail was dominated by convenience stores and off-premise alcohol retailers. However, the tendency for new entrants to favor lower-income tracts was not specific to unhealthy outlets: new healthy food retail also tilted toward lower-income tracts at least as strongly, and the share of openings that were unhealthy did not increase as income fell according to the authors’ decomposition.
This indicates that lower-income tracts experienced higher rates of retail turnover and net additions across multiple outlet types, not only an inflow of unhealthy retail.
The findings suggest that policies and interventions aiming to reshape the retail health environment will likely be more effective if they focus on controlling or influencing new openings rather than attempting to change the survival of established outlets. Instruments targeted by outlet type (for example, licensing, zoning, permitting, or incentives) may provide leverage to prevent or redirect the disproportionate entry of unhealthy outlets into lower-income neighborhoods.
Because closures were even-handed across income levels, measures that address only existing establishments would have limited impact on the observed income gradient unless accompanied by mechanisms that affect location decisions of new entrants.
The authors declared no competing interests. They reported following relevant ethical guidelines and obtaining necessary approvals, and they confirmed appropriate consent and reporting standards. Business establishment records were licensed from Data Axle USA and cannot be redistributed by the authors; the American Community Survey, TIGER/Line boundaries, and the Historic Redlining Indicator are publicly available. The study reported Robert Wood Johnson Foundation funding.
Limitations explicitly stated in the source text are limited; the article notes data licensing restrictions for the underlying business records and relies on classification decisions for what constitutes unhealthy versus healthy retail. The source did not provide further methodological limitations or sensitivity analyses beyond the three analytic approaches summarized above.