Intervention as Environmental Context
Although the current analyses are not designed to test the effectiveness of an intervention, it is important to note that these data were collected in the context of a multi-year, county-wide campaign to reduce sugary drink consumption, especially among school-aged children. In December 2012, a health-focused non-profit foundation joined community partners to launch the campaign. Based on the social-ecological model [26], comprehensive strategies were designed to promote behavior change at interpersonal, organizational, community, and policy levels. The aim was to reach the entire county population, which was approximately 300,000 people in 2012 [25]. A series of successful policy change campaigns: (a) removed student accessible vending machines from all middle schools; (b) set strong nutrition standards for student accessible vending machines in high schools; (c) significantly improved the comprehensiveness and quality of the school district wellness policy; (d) required all childcare facilities to eliminate sugary drinks and serve only healthier beverages; and (e) promoted healthier beverage options in all government-owned vending machines, and in recreation and parks youth programming. The campaign also employed digital marketing ads; cable television commercials; direct mail to households; and sponsored social media posts. An online tool, the Better Beverage Finder, was created to help residents search for healthier beverage options. “Street teams” conducted outreach to market the Better Beverage Finder at pools, parades, sporting activities, and other community events. Healthcare providers were encouraged to counsel their patients on sugary drink consumption and better prevent, diagnose, and treat childhood obesity. Additional details about the components of the campaign are presented in the primary outcome paper on the change in retail sales of beverages during the campaign [17].
Sample and Survey Administration
The study was approved by the Institutional Review Board at the University of Connecticut. The county where the study took place is a single school district. Surveys were administered to all sixth-grade classes across the middle schools during the school day using an online survey platform. Students were asked about their consumption of sugary drinks as part of a larger set of questions about diet, physical activity, media use, and sleep. The baseline survey was administered in November of the 2012–2013 school year and follow-up surveys were administered for the next four academic years between April and June. The survey took approximately 20 min to complete and classroom teachers were responsible for administration. The completed survey data were accessed and deidentified by the district Research and Program Evaluation office. In school year 2012–2013, students self-reported race/ethnicity, and in 2013–2014 and 2014–2015, students self-reported gender and race/ethnicity. In the 2015–2016 and 2016–2017 surveys, the district linked the survey responses to student records before deidentifying the data and provided demographic variables (e.g., gender and race/ethnicity) in order to reduce participant burden. The school district also provided race/ethnicity data for the full 6th grade class in each school for each year of the survey to allow comparisons between the sample and the population of interest.
Student Survey Questions
The questions about sugary drink consumption were identical across all years of the survey and were based on items in the California Healthy Eating Active Living Youth Nutrition and Physical Activity Survey and the Boston Youth Survey [27, 28]. Students were asked to report consumption of five types of sugary drinks: regular soda, fruit drinks, sports drinks, energy drinks, and flavored water and teas. Questions included example brands for each drink type and specified not to include “diet” drinks. Students reported consumption frequency as Never; I drink it but not every day; 1 time per day; 2–4 times per day; or 5 or more times per day. Students then selected the container size they usually consume for each drink type (e.g., glass, can, bottle, pouch, juice box). Daily calories from each target product were calculated by multiplying (a) student consumption frequency; (b) average size in ounces for the selected container type; and (c) average calories per ounce. The nutrition information for each type of drink was obtained from a comprehensive list of 644 sugary drinks commonly marketed to youth at the time of the baseline survey [29]. Appendix Table A1 in the supplementary material lists the calories per ounce by drink and container type. Total daily sugary drink calories were the sum of calories for each drink type per day.
Assessment of Neighborhood Food Environments
Previous research has defined “food deserts” as residential areas with limited access to affordable, healthy food (often operationalized as distance from a supermarket), and “food swamps” as areas where the availability of fast food and junk food supersedes healthy food options (often operationalized as the ratio of fast food and convenience stores to supermarkets) [30]. However, neither of these metrics are appropriate for the location of this study because this county has excellent access to healthy food options and 100% of the population lives within half a mile of a supermarket [31]. Because our focus is on young adolescents, we quantified the neighborhood food environment based on the prevalence of establishments where youth can independently obtain sugary drinks (i.e., fast-food restaurants, convenience stores, and gas stations with food and beverages) in the neighborhoods where they live and attend school.
We purchased address-level food store data from the National Establishment Time-Series (NETS) Database for 2014 (the midpoint year of the study period). NETS data reflect archival establishment information from Dun and Bradstreet [32]. The micro-level dataset contains each business’ name; address and contact information; years active; and primary industry classification. We obtained data with the North American Classification system codes of 445110 (supermarkets and other grocery stores), 445110 (fast-food restaurants), 445120 (convenience stores), and 447110 (gas stations with food and beverages).
To measure students’ food environments, we utilized the middle school “attendance zones,” which are the neighborhoods surrounding each middle school building where all of the students who attend that school live. We obtained the school attendance zone shape files from the school district and merged them with the food establishment data using ArcGIS Desktop 10.2.2. We defined high exposure to unhealthy food retail zones as school attendance zones where the number of unhealthy food retailers (i.e., fast food, convenience stores, and gas stations) was higher than the in-sample average number of establishments. For our analyses, we constructed a binary variable equal to 1 if the school attendance zone met this criterion and equal to 0 if the number of unhealthy food establishments was below average.
When the study began in 2012, there were 19 middle schools; however, in school year 2015–2016, a new middle school was opened and students were drawn from the three surrounding middle schools. To ensure that we were comparing the same neighborhoods over time, we combined the new school’s attendance zone with the zones of the three surrounding schools to create one combined attendance zone. This did not change the coding of the food environment for any of the schools; each of the three original middle schools were coded as “high exposure” on their own before the new school was built. This designation remained the same for the combined zone of four schools.
Outcome Measures
The primary sugary drink consumption outcome measures were: share of students reporting daily sugary drink consumption (i.e., ≥ 1 time per day; total and by drink type) and estimated daily calories consumed from sugary drinks (total and by drink type). Students who reported not consuming a beverage were coded as consuming 0 calories from that beverage.
Data Analysis
Data analyses were completed in Stata/SE 15.0 [33]. First, we assessed the survey response rate. Second, we compared the racial/ethnic distribution of our sample with the racial/ethnic distribution of the full 6th grade population in each school using paired t-tests. Because the method of assessing the race/ethnicity variable changed over time (i.e., self-report during the earlier waves of the survey and drawn from administrative data for the last two waves of the survey), we conducted one set of t-tests with self-report data and a second set with administrative data to examine any shifts in the representativeness of our sample.
Next, we assessed how our independent variable “exposure to unhealthy food retail” was associated with population density (which might explain the higher number of food outlets) and grocery store availability (which would suggest that the number of grocery stores should be included in the assessment of the food environment). We used a t-test to compare the middle school enrollment sizes for high versus low exposure zone schools for 2013–2014, which was the year corresponding to the retail outlet data. A t-test was also used to assess whether the number of grocery stores differed between the high versus low unhealthy retail exposure zones. The mean number of fast-food restaurants and convenience stores/gas stations by exposure zone was also calculated. The proportion of students (overall and by race/ethnicity) living in high exposure zones was calculated and we used chi-square analyses to test whether there were significant differences in the likelihood of specific racial/ethnic groups living in these neighborhoods.
As noted above, we selected two outcome variables: the percent of students reporting daily sugary drink consumption and estimated daily calories consumed from sugary drinks. Logistic regression models were used when the percent of students reporting daily sugary drink consumption was the outcome and linear regression models were used when the number of calories consumed was the outcome. All models included random effects parameters for school-level nesting and robust standard errors. To assess changes over time, the outcomes for each year were compared with 2012–2013 (baseline) levels using unpaired two-sided t-tests with a Bonferroni correction to account for a potential inflation of type 1 error rate from multiple comparisons. Only results significant at α = 0.01 are discussed in the text of the “Results” section.
For the first outcome, logistic regression models were used to estimate the percent of students reporting daily sugary drink consumption, comparing baseline levels (2012–2013) to each subsequent year of the survey. This was done for “any sugary drink,” and each of the five types of drinks separately. The analyses that included the full sample were adjusted for student race/ethicity. We also completed separate models for each of the five racial/ethnic groups.
In the next set of analyses, the students were divided into two groups: those exposed to high versus low levels of unhealthy food retail. We used a t-test to assess whether daily sugary drink consumption was different between high and low exposure zones. Then, logistic regression models were used to estimate the percent of students reporting daily sugary drink consumption, comparing baseline levels (2012–2013) to each subsequent year of the survey. The analyses that included the full sample were adjusted for student race/ethnicity. We also completed separate models for each of the five racial/ethnic groups.
For the second outcome, linear regression models were used to estimate daily calories consumed from sugary drinks, comparing baseline levels (2012–2013) to each subsequent year of the survey. Following the same strategy, this was done for “any sugary drink,” and each of the five types of drinks separately. Again, after the analyses of the full sample (adjusted for student race/ethnicity), separate models were run for each of the five racial/ethnic groups. In the final set of multiple linear regression analyses, the sample was divided into those who live in high versus low exposure zones. We examined estimated daily calories consumed from all sugary drinks and compared baseline levels (2012–2013) to each subsequent year of the survey. This was assessed first for the full sample (adjusted for student race/ethnicity), followed by parallel models for each of the five racial/ethnic groups.