Data

The main information source for the structural model for the Mexican non-dairy and non-alcoholic beverage market is the Nielsen Mexico Consumer Panel Service (Nielsen CPS) from 2012 to 2015 [10]. These data contain information on household purchases for pre-packaged foods and beverages [10]. Nielsen CPS is representative of cities with >50,000 inhabitants, representing 63% of the Mexican population and 75% of the national food and beverage expenditure in 2014 [10,11].

For the epidemiological model, we used pre-tax data (i.e., before 2014). Information on Mexicans’ weight and height and SSB consumption by age group and sex in urban areas (>2,500 inhabitants) came from the 2012 National Health and Nutrition Survey (ENSANUT in Spanish) [12]. We derive information on prevalence, incidence, and case fatality at the disease level in Mexico in 2013 based on the Global Burden of Disease (GBD) study and overall mortality in 2013 from the Mexican National Population Council (CONAPO in Spanish) [13,14]. Specifically for disease-specific information using GBD data, we calculated prevalence as the ratio between cases of the relevant disease and the overall relevant population, incidence as the ratio between new cases of the relevant disease and susceptible population (i.e., subtracting existing cases from the overall relevant population), and case fatality as the ratio between deaths and cases of the relevant disease. We calculated all these measures by age group and sex in 2013. The GBD study lacked information on the incidence of hypertensive heart disease, and we estimated it using DISMOD II [15]. Population size by sex and age at the beginning of 2014 came from CONAPO [14]. We used a population size equivalent to 63% of the national population aged ≥ 20 years, in line with the representativeness of Nielsen CPS and the epidemiological model explained below.

We retrieved average information on the annual healthcare cost per patient from the Mexican Ministry of Health for most diseases [16]. Healthcare cost information for diabetes by age group came from Sanchez-Romero et al. [3]. We assume that the average cost for diabetes among people aged 20–54 years in our CBA was the average for age groups of 35–44 and 45–54 years in the study by Sanchez-Romero et al. [3] because this study had no cost information for people younger than 35 years. Due to the lack of information on kidney cancer costs, we derived it from a disease cost ratio. We informed this ratio based on cost information from Ecuador and Chile based on an analysis by Fernandez et al. [17]. These countries are likely to share common characteristics with Mexico due to all being in Latin America. Specifically, we assumed that the ratio of the annual cost per patient of kidney cancer to colon cancer in Mexico was the average of this ratio in Ecuador and Chile [17].

In our study, the economic consequences linked to the tax effect on mortality are composed of two major components. First, the government will have to fund public retirement pensions and healthcare utilization for people who would have died in the absence of the tax. We retrieved information on the average public retirement pensions for people aged ≥ 65 years and the probability of getting this pension from the National Expenditure and Income Survey [18]. Moreover, we calculated the costs of the additional healthcare utilization using the information on the public health spending not allocated to treat obesity-related diseases and how the average per capita healthcare cost varies as people age. Second, to account for the economic valuation by consumers linked to mortality reductions after the tax implementation, we used the value of a statistical life (VSL). VSL corresponds to people’s willingness to pay to reduce their mortality risk and is supposed to capture the monetary and non-monetary aspects that people attach to this reduction in mortality risk [19,20]. We used the VSL estimate from the Mexico City metropolitan area (i.e., USD 280,743; 2014 value) [21].

We complemented the economic data using the information on wages for people with the right to paid sick days and disability pensions from the Mexican Ministry of Health and information on the percentage of people working in Mexico from the Mexican National Occupation and Employment Survey [16,22]. We calculated the average caregiving time by patients’ relatives by disease and sex group using information for Mexico from the Economic Commission for Latin America and the Caribbean [17]. We translate this time into monetary value based on the average hourly minimum wage in Mexico in 2014 [23], from which we assumed an average working day of eight hours. S1.1 Table in S1 File lists the reference of all the above-mentioned inputs. In S1.2 Table in S1 File, we report the summary statistics of the main model input information. In S1 Metadata in S1 File, we present the disaggregated economic information in S1.2 Table in S1 File by sex-age group (when applicable). Moreover, this metadata includes epidemiological information on population size and overall mortality, prevalence, incidence, case fatality by disease of interest, body mass index (BMI) and variability in SSB consumption.

Structural model

The structural model of demand and supply for the Mexican non-dairy and non-alcoholic beverage market allows us to assess the effect of different SSB tax policies in urban Mexico [2]. This market is composed of untaxed (i.e., water and diet SSB) and taxed beverages (i.e., SSB). The demand model is built upon the assumption that consumers purchase the taxed or untaxed beverage that provides the highest level of satisfaction (known as utility in economics) according to their preferences over products’ attributes with no health-related concerns. Below, we relax this assumption by allowing some degree of health awareness. The demand model corresponds to a variety of a logit model (i.e., random-coefficients logit model) whose functional form entails that the demand does not hold a linear relationship with prices. For the empirical application of the model for the beverage market in Mexico, the authors modeled households’ preference over sugar content and prices while accounting for brand-fixed effects in the utility function [2]. For the supply model, the authors of the structural model assumed that producers compete in a Bertrand-Nash supply model and set prices that guarantee the highest profits across their product portfolio, including taxed and untaxed beverages. In practice, the supply-related part of the structural model is recovered by using the first-order condition under which producers are theoretically supposed to maximize profit when setting prices.

When an SSB tax comes into effect, the structural model accounts for the simultaneous response by consumers (seeking to maximize utility) and producers (seeking to maximize profits) that results in new equilibrium information on prices, purchases, tax revenue, consumer surplus, and profit. Derived changes from the model in consumer surplus and profit capture offsetting gains arising from the substitution between taxed and untaxed beverages in response to price increases after the tax implementation. We used the structural model to estimate all these outcomes for the three tax levels of interest (i.e., $1 MP, $2 MP, and $3 MP). Appendix A2 in S1 File shows equations to calculate the tax impact on relevant outcomes. S1.3 Table in S1 File presents the summary statistics of the data used in the structural model. This table shows price and purchase information from Nielsen CPS [10] before (i.e., 2012–2013) and after (i.e., 2014–2015) the tax implementation.

Epidemiological model

We estimated the SSB tax effect on health outcomes using a published epidemiological life-table model by Veerman et al. [9]. This model simulates health-related trajectories for the reference population and the intervention population. For the former population, trajectories are based on life tables that depend on sex-age groups’ information on overall mortality, the no-intervention-affected BMI, and linked mortality and morbidity for a set of diseases. In a similar fashion, the model estimates the health-related trajectories for the intervention population but accounting for the calorie reduction due to the intervention of interest. This calorie reduction will lead to a body weight loss following Hall et al., who posited that a permanent daily change of 10 calories would generate a steady-state change in adults’ weight of one pound [24]. The body weight loss will reduce the population’s BMI and thus the mortality and morbidity for a set of diseases according to the potential impact fraction. This fraction captures reductions in disease-specific risks of incidence in response to reductions in a risk factor [25], which corresponds to BMI in the model. The model allows the population’s members to develop multiple diseases. These diseases correspond to diabetes, ischemic heart disease, stroke, hypertensive heart disease, osteoarthritis, and a set of cancers (i.e., breast, colon, and kidney).

The intervention effect arises from the difference in outcomes between the reference population and the intervention population. The epidemiological model provides yearly estimates by sex for adults aged ≥20 years from the intervention implementation until people turn 95 years old or die (split into 15 five-year age groups). Hence, the model represents a closed cohort that gets smaller over time. Below, we present the results for different time horizons.

For the empirical model, we assumed a normal distribution for SSB purchases and a lognormal distribution for BMI and relative risks of incidence at the disease level. Moreover, we assumed that as people age, their average BMI would be the average for the next age group, which entails an increasing trend for several age groups (see S1.4 Table in S1 File). Due to the structural model provides a point estimate of SSB purchases, we allowed these purchases to resemble the variation in the SSB consumption by age group and sex in urban Mexico as in the 2012 ENSANUT [12]. We present the SSB consumption distribution by sex-age group in S1.5 Table in S1 File.

We set 2014 as the year of the SSB tax implementation. We run the model using Excel and the add-in by Epigear software [26]. This software allows us to estimate the potential impact fraction and to account for the uncertainty from the tax effect on BMI and disease-specific relative risks. We conducted a Monte Carlo simulation for each tax policy based on 2,000 iterations to account for the model’s uncertainty. We report their average and 95% uncertainty ranges based on these iterations.

Cost-benefit analysis

The SSB-tax economic benefits for the government include healthcare savings, fewer disability pensions and paid sick days, and tax revenue. However, the government will face costs by funding public retirement pensions and healthcare services unrelated to overweight or obesity due to the survival increase after the tax implementation. See Appendix A3 in S1 File for the procedure to estimate the cost of the latter healthcare services.

We decompose the tax effect into three components for society’s members other than the government. The first component is the costs arising from reductions in producers’ profit and consumer surplus due to SSB purchase decreases attributable to the tax. Second, due to fewer disease cases after the tax implementation, consumers’ relatives will benefit by allocating less time to serve as caregivers, which we translate into monetary value according to the average hourly minimum wage. Third, we transform the mortality reduction into economic benefits for consumers based on VSL.

Under a conservative scenario, there would be no need to account for VSL benefits by assuming consumers are fully informed and rational and thus fully internalize any adverse health effect when purchasing SSB. Hence, consumer surplus would reflect this internalization. Conversely, under an optimistic scenario, consumers would experience the full VSL benefit when completely lacking information on SSB-related adverse health effects. Due to the infeasibility of these extreme scenarios, we set the economic benefit to be equivalent to 80% of the VSL per avoided death under the assumption that consumers internalize 20% of the negative SSB effect on health, which entails it is already part of the consumer surplus. This 20% is close to the expected proportion of around 22% of Mexican adults who use information from front-of-package labels (for the label design preceding the warning labels implemented in October 2020) and distinguish an unhealthy ultra-processed product (i.e., high in nutrients linked to poor health outcomes such as sugar and sodium) according to these labels. Specifically, 36% of Mexican adults reported sometimes or usually using front-of-package labels, and 61.5% of Mexican adults classified an unhealthy ultra-processed product as unhealthy or partially unhealthy based on these labels (36% x 61.5% ≈22%) [27].

The SSB tax net benefit arises from the difference between costs and benefits described above. In sensitivity analyses, we assume no VSL benefits, which implies an internalization rate of 100% when purchasing SSB in line with the conservative scenario described above. Conversely, the optimistic scenario corresponds to full benefits from VSL equivalent to an internalization rate of 0%, which we also include as part of the sensitivity analyses.

In light of the relevant stakeholders described above, our CBA is from the perspective of the government, producers, and consumers. We transformed all economic information for our CBA into 2014 values using the Mexican consumer price index [28]. We report all our economic results in USD based on the exchange rate in 2014 and a discount rate of 4%. This discount rate corresponds to the recommended rate for an upper-middle-income country like Mexico [29]. In Appendix A4 in S1 File, we explain how we calculate each CBA component.