This natural experiment involved 4 annual time points of data collection (T1-T4) from 2014 to 2017 in stores in Minneapolis, MN (where the Staple Foods Ordinance was being implemented) and St. Paul, MN (the comparison community).
The study sample and store recruitment process has been described previously [5, 6, 26, 28,29,30,31,32,33]. Stores exempted from the ordinance (including those that would not reasonably be expected to stock a minimal amount of foods and stores located the core downtown commercial district) were excluded from the evaluation in Minneapolis and St. Paul. The evaluation study targeted retailers that were not already expected to meet the new minimum stocking requirements; thus supermarkets, mass-merchandizers, and stores participating in the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) were excluded from the sample. Finally, stores with invalid licensing addresses were excluded.
Of 255 eligible stores, 180 were randomly selected to participate. After visiting these stores to collect data at baseline (T1), 23 were identified as ineligible (e.g., due to new participation in WIC), and an additional 17 refused to participate. At the three subsequent data collection time points, study staff re-visited stores that refused to participate at T1 to attempt data collection. The final analytic sample had 155 unique stores (n = 140 at T1, 139 at T2, 137 at T3, and 127 at T4). Among the eligible sample, all stores were categorized as corner stores, gas-marts, dollar stores, pharmacies, or general retailers at T1. The study was approved by the Institutional Review Board at University of Minnesota.
Store environment was assessed using a tool modified from the Rudd Center for Food Policy and Obesity that was developed to evaluate changes in WIC policy revisions in small food stores in 2009 . The tool, described previously [5, 6, 26, 28], is similar in format to Nutrition Environment Measure Survey in Stores (NEMS-S), but modified to align with the 10 Minneapolis Staple Food Ordinance requirements.
From the data generated using the store assessment tool, three indicators of ordinance compliance were generated:
80% compliance with ordinance requirements: the percent of stores that met at least 8/10 of the product category requirements of the ordinance (yes/no).
10-point scale: the total number of ordinance requirements met by stores (range 0–10), presented as an average across stores.
Carrying any food in each of the 10 categories: whether stores had any food in each of the categories required by the ordinance, even if the food was not in the appropriate package size, form, or quantity required by the ordinance (e.g., eggs sold in dozen containers were required by the ordinance, but eggs packed in half-dozen containers met the criteria for any eggs). The indicator is presented as the percent of stores that had any in all 10 categories (yes/no).
Store size (small/larger) was measured during the assessment as the number of cash registers in the store. Small stores had 1–2 cash registers and larger stores had at least 3 cash registers.
Store ownership status (corporate/independent) was determined during an interviewer-administered survey with store managers in which they were asked whether the store was independently-owned, corporately-owned, or part of a franchise. Franchise and corporately-owned stores were combined into a single category. In stores where ownership status was not available from the manager survey, two study team members (CEC and MNL) determined status based on publicly available information about the store (e.g., name, number of locations). Stores that were part of well-known chains were assigned corporate status; stores that had only one location were deemed independent.
Neighborhood data were obtained from 5-year American Community Survey estimates (ACS, 2009–2015)  and attributed to stores based on census tract location. Store census tracts were classified into lower-socioeconomic status (SES) or higher-SES. Lower-SES census tracts had > 50% of residents at or below 185% of the federal poverty income guidelines .
Using the USDA Food Access Research Atlas , stores were classified as low-income/low-access if the census tract they were located in was both low-income and low-access. Low-income tracts met any of the following criteria:  median family income ≤80% of the state-wide of the metropolitan area’s median family income ; poverty rate > 20%. Low-access tracts had ≥100 households located > 1/2 mile from the nearest supermarket and had no access to a vehicle.
Descriptive statistics were computed for store and neighborhood characteristics at baseline for Minneapolis and St. Paul separately, expressed as number and percentage of stores/neighborhoods. We also computed chi-square tests to compare the store and neighborhood characteristics across the two cities. Using data from T1-T4 from both cities, a mixed regression model for each of the three compliance outcome measures was conducted to examine the overall movement towards compliance in Minneapolis compared with the control condition in St. Paul. For each model, we tested an overall time-by-city interaction, adjusted for neighborhood race/ethnicity (the only significant covariate in bivariate city comparisons).
Subsequent analyses were limited to stores in Minneapolis only, to compare magnitude of changes in compliance within different stores and neighborhoods of the policy area. A mixed regression model was computed for each compliance outcome with store (size, ownership status) and neighborhood characteristics (SES, low-income/low-access status) as independent variables. Models tested the interaction between time and store/neighborhood characteristic (3 degrees of freedom) for each outcome. For interactions that reached statistical significance, we tested changes between baseline and each time point (T1 to T2, T1 to T3, and T1 to T4). All models were adjusted for repeated measures over time. All analyses were conducted in SAS (SAS/STAT Version 9.4).