This study used anonymized monthly household purchase data from January 2016 through December 2017 from Kantar WorldPanel Peru. A panel of 3800 households was recruited through stratified random sampling. Households that left the panel were replaced by randomly selected households with similar demographic characteristics, after completing a three-month run-in period for quality assurance. With replacement, the analytic sample comprised 5145 unique households, with an average of 18 months of follow-up (median: 23), providing 90,654 household-month observations. Households were recruited from 14 major cities in six geographic regions of the country (Lima, central coast, northern coast, southern coast, the Andean highlands, and the Amazon). Kantar WorldPanel Peru provides monthly household sampling weights to ensure the panel’s representativeness of 67% of the urban population of Peru. An area is considered urban if it has at least 100 dwellings in a contiguous group or is a district capital, and it has at least 2000 inhabitants [18]. Kantar WorldPanel Peru excludes households with a potential conflict of interest related to products studied and households who do not meet minimum purchasing standards (e.g., not purchasing any items from a 15-category “basket” of basic goods in a month).
Trained data collectors visited participating households weekly to scan barcodes of all packaged food and beverage items purchased for at-home consumption, using standardized codebooks for bulk products and items without barcodes. Panelists were instructed to save all receipts and empty containers to be scanned. The dataset contains item-level information including barcode, product name, brand, description, volume, price per unit, and date of purchase. Product-level data were used to link beverage purchases to nutrition facts panel (NFP) data based on barcode, brand, and product description, as in previous studies [19].
NFP data were obtained from product photographs collected in grocery stores by a team of Peruvian research assistants in 2018, which has been described elsewhere [20] and managed using REDCap electronic data capture tools hosted at UNC [21]. If no collected NFP data were available for a purchased product, it was linked to nutrition facts panel data from Mintel Latin America. After the linkage, all beverages in the dataset were categorized by trained nutritionists according to beverage type and tax status under the 2018 regulation [15] (untaxed, lower-sugar taxed, high-sugar taxed) (Supplemental Table 1). High-sugar taxed beverages, such as soda, were defined as drinks containing ≥6 g of sugar per 100 mL, while lower-sugar taxed beverages like diet soda were defined as drinks containing < 6 g of sugar per 100 mL. Beverages containing no added sweeteners, such as bottled water, 100% fruit juice, plain milk, as well as drinkable yogurt, infant formula, and powdered fruit-flavored drink mixes, were exempt from the regulation.
Beverage types included water, milk, regular soda, diet soda, fruit juice drinks (juices or nectars containing < 100% fruit juice), refrescos (fruit-flavored drinks), dairy drinks (flavored or sweetened evaporated or condensed milk drinks), coffee, tea, 100% fruit juice, and sports and energy drinks. Beverage purchases were reported per month by beverage type. Volume (mL) purchases for each beverage type were divided by household size to calculate per-capita volume purchases. Each observation therefore represents one household’s per-capita beverage purchases in a month.
Demographic data was collected upon enrollment in the study and updated annually. Key demographic characteristics reported included socioeconomic status (SES), head of household educational attainment, region, number of children under 13 years of age and household size. Head of household was defined the person who lives in the home and generates the most income for the household and/or makes the financial choices of the family. Head of household education was categorized into three bins for analysis (less than high school, completed high school, and more than high school) from ten original categories.
Household SES was calculated from an assets index and key sociodemographic characteristics. This measure was developed by the Peruvian Association of Market Research Firms (APEIM) and has been applied to data from the Peruvian National Household Survey (ENAHO) to provide population-level SES estimates [22] and has been used in prior studies [23, 24]. Specifically, household SES was categorized from A (high) to E (low), based on ownership of items such as washing machines and cars, as well as living conditions like floor material and bathroom type (indoor/outdoor), and sociodemographic characteristics like education and insurance status. Because the proportion of households in category A was small (< 5% of households), these households were combined with category B for analysis.
Statistical analysis
To assess trends in beverage volume purchases, separate weighted OLS regressions of total volume, untaxed volume, lower-sugar taxed volume, and high-sugar taxed volume purchases per capita per month on a linear time trend (i.e., treating month as a continuous variable with range 1–24) were run. Month dummies [1,2,3,4,5,6,7,8,9,10,11,12] were included to account for seasonality and standard errors were clustered at the household-level. Plots generated from weighted OLS regressions of total volume purchases and volume purchases by tax status in L per capita per month treating month as a factor variable were visually inspected to understand seasonal variation.
The percentage of households purchasing each beverage type in a month and the mean unadjusted volume of purchases by beverage tax status, overall and by key demographic characteristics (region, education, and SES) were calculated, using survey weights. Weighted, mean unadjusted volume purchases by beverage type for high-sugar taxed, lower-sugar taxed and untaxed beverages, overall and by key sociodemographic characteristics, were estimated to explore the beverage types contributing the largest volume to beverage purchases by taxation status. Volume purchases among households that purchased any beverages with a particular tax status (i.e., excluding those who made no purchases in a tax category in a month) were also assessed. Volume purchases per capita of each beverage type, independent of tax status (i.e., combining both taxed and untaxed refrescos into a single category), were also assessed. Because the pattern of results was similar across years, we report results from 2017 only (2016 results are available in Supplemental Tables 4, 5, and 6). As a sensitivity analysis, we also conducted the analyses without survey weights (available in Supplemental Table 3). Furthermore, because young children may consume a lower volume than older children or adults, we performed two sensitivity analyses excluding a) children under 2 and b) children under 5 from the total number of household members when calculating the volume per capita (results available in Supplemental Tables 7 and 8).
All analyses used cluster-robust standard errors at the household level to account for repeated measures. Analyses were conducted using Stata 16 (College Station, TX, USA).