Study sample
We conducted an ecological study across all public and private schools of the second largest city in Brazil. Rio de Janeiro has 6,775.561 inhabitants and a Human Development Index (HDI) of 0.799. While 31.4% of its citizens have a per capita income of half a minimum wage, the average per capita income is around 4.2 minimum wages.
Secondary database from 2019 was extracted from the State Education Secretariat (Secretaria Estadual de Educação – SEE). Seventy-nine schools were excluded from the final sample. 38 offered only professional education, 7 that attended only special education, and 34 with missing segregation data. The final sample comprised 3,159 schools, which are categorized into public and private.
Neighborhood inequality measures
We collected three measures of socioeconomic spatial inequality, and aggregated at the neighborhood level—which is composed of an aggregate of census tracts: (i) per capita income, (ii) segregation index, and (iii) deprivation index. Data to build the indicators was collected from the last 2010 country census. Besides, we converted all current values from Brazilian Real to US Dollars, using 2010 as the conversion date. Importantly, census tracts are defined in Brazil as the smallest territorial unit, formed by a continuous area, fully contained in an urban or rural area and determined according to the number of households [20].
Per capita income was calculated as a ratio between total income of the neighborhood and its population. We categorized per capita income into terciles: lowest tercile ranging from U$249.21 and U$593.43; Middle tercile from U$593.44 and U$1,022.44; and highest tercile from U$1,022.55 and U$6,510.32.
As per the segregation index, we used the Getis-Ord Local Gi*statistic (or Gi*statistic). It encompasses a spatially weighted Z-score that represents how much a neighborhood’s income composition deviates from the larger metropolitan areas in its surroundings. Therefore, this index is unique because it (a) takes into account the spatial clustering of segregation within cities, and (b) considers racial composition and social inequalities within and between neighborhoods.
The segregation index is calculated as the standard deviations (SD) between the economic composition of the neighborhoods—assessed by the proportion of heads of household in neighborhoods that earn a monthly income within 0 to 3 SD of the minimum wage—in relation to the surrounding neighborhoods. Thus, we can detect segregation at neighborhood-level and thereby examine segregation within metropolitan areas. Data from the 2010 Brazilian Census were used to determine the proportion of heads of households in a neighborhoods earning a monthly income within 0–3 minimum wage (approximately US$ 0·00–US$ 900·00 in 2010) [21].
The census tracts were weighted using a first-order rook spatial weight matrix and three categories of segregation were created: (1) High: Gi * statistic ≥ 1·96; (2) Medium: Gi * statistic between 0 and 1·96 and (3) Low: Gi * statistic < 0 according to the distribution of the Z-score. Higher, positive scores represented census tracts that are more segregated—meaning that the proportion of households receiving 0–3 minimum wages are overrepresented in the neighborhood, while lower, negative scores suggest the opposite. Values close to 0 represent neighborhoods in which spatial segregation is low.
Finally, the deprivation index was retrieved from a recently published technical report [22]. It is calculated as a combination of three main indicators: (i) percentage of households receiving less than ½ minimum wage, (ii) percentage of illiterate inhabitants age 7 or older, and (iii) average of individuals with inadequate access to sanitation. Because this index is available for census tracts and our unit of analysis is the neighborhood, we average those sectors and aggregate their value to fit the neighborhood unit of analysis. We then use as thresholds the average of deprivation +-½ SD. Low deprivation areas are those below the lower end of the average vulnerability subtracted by ½ SD, while high deprivation is above the average + ½ SD. Medium deprivation is the interval between the SD from the average. Then, we also categorized this index into three, based on the Health Vulnerability Index [23]: high, medium, and low deprivation neighborhoods.
Community school food environment
Secondary database of food establishments from 2019 was extracted from the Rio de Janeiro State Treasure Secretariat (Secretaria Estadual da Fazenda). We classified establishments as food stores based on the National Classification of Economic Activities (Classificação Nacional de Atividades Econômicas, CNAEFootnote 1), which informs the main economic activity of each registered establishment.
Food establishments were classified in accordance with the Technical Study on Mapping Food Deserts in Brazil—CAISAN), in which categorizes establishments in three: (i) in natura establishments, which included butcheries, seafood shops, and fruits and vegetables establishments; (ii) ultraprocessed establishments, which included bars, snack bars, convenience stores, candy shops, and street vendors; and (iii) mixed establishments, which included markets, hypermarkets, mini markets, bakeries, restaurants and food stores (general food stores as natural and dietetic products, frozen foods, ice-cream stores, cake factories, warehouses, and commercial food stores with predominant retail of processed foods) [24].
Furthermore, the food environment around the schools was analyzed considering food deserts and swamps. Food deserts are defined as neighborhoods with limited access to healthy food, while food swamps are defined as neighborhoods with high availability of unhealthy foods [25]. To determine food deserts, we adopted the methodology proposed by CAISAN, which calculates the density of healthy establishments per 10,000 inhabitants. Healthy establishments correspond to the sum of in natura acquisition and mixed establishments. Given the above, food deserts are the neighborhoods that are below the percentile for the distribution of the density of healthy establishments [24]. In the present study, the 25th percentile corresponds to 27.35. To determine food swamps, on the other hand, we calculated the sum of convenience stores, snack bars, grocery stores and candy stores. When the sum of these establishments in the neighborhood was greater than four establishments, the neighborhood was classified as a food swamp [5, 6]. We thus used the adapted metric of Hager and colleagues (2017), first utilized by Peres and colleagues (2017). In this adapted metric, we account for the Brazilian context by adding Cafeterias (‘lanchonetes’)—food establishments largely characterized by the abundance of ultraprocessed foods. Additional results using mRFEI can be found in the Appendix.
Statistical analysis
Descriptive statistics are presented in absolute and relative frequencies. We calculated measures of central tendency as the median and the interquartile range (p25-p75). We compared relative frequencies using chi-square tests. The analyses were performed in SPSS 19.0, and maps were elaborated in QGis 2.14.9.