Carbonated Beverages

Prior to the survey, data were collected on participants’ key demographic characteristics, usual consumption of common foods and beverages, and frequency of consumption of a range of beverages including carbonated drinks, energy drinks, milk, and juices. These also served as screening questions to ensure respondents were in the target age range and had consumed SSBs in the previous two months.
Following randomisation, participants viewed the allocated image, and used a seven-point rating scale to answer questions regarding attitudes and predicted product preferences. Specific semantic differential attitude statements assessed whether participants believed the displayed SSB was: expensive/cheap, unattractive/attractive, low quality/high quality, uncool/cool, unhealthy/healthy, and tasted bad or good. As adolescents and young adults often consume SSBs in public settings, and brands are used to communicate aspects of consumers’ identities [24 (link)], we also asked questions regarding their perceptions of a peer if they were drinking from the can displayed. Perceptions were measured using four semantic differential questions anchored by: boring/interesting, unpopular/popular, unfashionable/fashionable, and old/young.
The 11-point Juster Scale was used to measure the “in the moment” probability of purchasing the displayed drink if it were one of the options available at a convenience store, where 0 represented “no chance or almost no chance”, 5 represented “fairly good possibility”, and 10 represented “certain or practically certain” [9 (link), 25 ]. Five-point Likert scales were used to measure participants’ attitudes towards proposed implementation of warning labels and taxes on SSBs. Two versions of the survey were used, one for the young adult group (17–24 years), and another, with simpler language, targeted to the adolescent group (13–16 years).

Since the tax was implemented nationally, it was not possible to have a true experimental design to study the effect of the SSB tax. Therefore, we applied a pre-post quasi-experimental approach. We used fixed effects models to control for all variables that do not change over time. We also included variables that change with time and can explain changes in the demand and prices of beverages such as annual projected population[30 ] and annual gross domestic product[31 ].
We applied the model to all SSBs that are subject to the tax and we also stratified the models by two beverage categories given the differences described in terms of their prices and price elasticities: 1) carbonated sweetened beverages (CSB) that includes non-diet soft drinks and 2) non-carbonated sweetened beverages (NCSB) including flavored water, juices and nectars. The fixed effects model is laid out as follows:
$P_{itj}\mathrm{ }=\mathrm{ }{\delta Y+\beta }_{y}Y+{\beta }_{m}M+{\beta }_{m2}M2+{\beta }_{s}S+{\beta }_{p}Do+{\beta }_{g}G+{\alpha }_i+u_{itj}$
The dependent variable P is the real price per liter of a specific beverage i at month t, year j, Y is a vector of dummies for each calendar year (leaving 2013 as the reference), M is a count variable for the entire month/year period 2011–2014, M2 is the variable month squared to test for non-linear associations, S denotes the season that takes the value 1 during the period of higher prices (April to September), 0 otherwise, D is total annual population projected (in millions) in the country and G is the annual gross domestic product in the previous year in millions of real pesos, α are time invariant unobservable factors associated with P and u is the error term.
To see how the taxes passed over time, we included a model where we adjust for each month of 2014 and compares prices with 2013.
We explored the heterogeneity of the effect of the tax on SSB prices by region and the two beverage categories (CSB and NCSB) and we also ran the fixed effects models by package size to see if there were differences in changes in price after the tax was implemented.
To test the robustness of the results, we applied two other model specifications as sensitivity analyses. First, we used an Arellano-Bond Dynamic Panel Estimator that includes lag of prices as instrument variables for endogenous regressors and that addresses the potential autocorrelation[32 ]. We also applied interrupted time-series analyses (ITSA) that have been used to estimate in non-experimental designs the impact of tobacco taxes[33 (link)]. As in time series, for the ITSA price data are collapse by month to run the estimations. All models adjust for the same variables as in the fixed effects regression for comparison. Results from all models are presented showing unweighted and weighted estimations using Nielsen purchase data as described above. For fixed effects, we used the command–areg- in Stata (linear regression with a large dummy-variable set) that allows using time variant weights. All empirical analyses were run in Stata 13[34 ].

Information on the frequency of coffee, tea, and carbonated beverage intake over the past year was obtained using the questionnaire. It consisted of the following 10 categories in 2007–2011: almost no intake, 6–11 times per year ((8.5/52.1)/week), 1 time per month ((1/4.3)/week), 2–3 times per month ((2.5/4.3)/week), 1 time per week (1/week), 2–3 times per week (2.5/week), 4–6 times per week (5/week), 1 time per day (1/day), 2 times per day (2/day), and 3 times per day (3/day). It consisted of the following nine categories in 2012–2016: almost no drinking, 1 time per month ((1/4.3)/week), 2–3 times per month ((2.5/4.3)/week), 1 time per week (1/week), 2–4 times per week (3/week), 5–6 times per week (5.5/week), 1 time per day (1/day), 2 times per day (2/day), and 3 times per day (3/day). It consisted of the following seven categories in 2019–2020: <1 time per week (0.5/week), 1 time per week (1/week), 2–4 times per week (3/week), 5–6 times per week (5.5/week), 1 time per day (1/day), 2 times per day (2/day), and 3 times per day (3/day). We categorized the frequency of intake as follows: group 1 (<1/week), group 2 (1/week to <1/day), and group 3 (≥1/day).

We used STATA version 14.0 (StataCorp., College Station, TX, USA) for statistical analysis, and p < 0.05 was set as the statistical significance level. The KNHANES was conducted using a two-stage stratified cluster sampling method. Therefore, we assigned weights to the stratified data in our analysis.
Linear regression analyses and χ² tests were used to analyze and compare the participants’ various characteristics according to sex and group. Logistic regression analyses by sex were used to assess CV risk factors (HTN, DM, DL, or MetS) according to coffee, tea, and carbonated beverage intake frequency. Adjusted odds ratios (ORs) were derived after controlling for potential confounding variables, such as age, daily nutritional intake (total and fat), average monthly household income, education, smoking, alcohol drinking, walking, BMI status, the frequency of intake of coffee, tea, and carbonated beverages, and menopausal status (only in women). In addition, the proportion of age groups according to the frequency of intake of coffee, tea, and carbonated beverages was derived. As a sensitivity analysis, logistic regression analyses were also conducted for participants 20 to less than 60 years.