Decomposing consumer and producer effects on sugar from beverage purchases after a sugar-based tax on beverages in South Africa

Abstract

Growing global concern about obesity and diet-related non-communicable diseases has raised interest in fiscal policy as a tool to reduce this disease burden and its social costs, especially excise taxes on sugar-sweetened beverages (SSBs). Of particular interest have been nutrient-based taxes to improve diet quality. These can incentivize producers to reformulate existing products and introduce healthier alternatives into their ranges. In 2018, South Africa adopted a sugar-based tax on SSBs, the Health Promotion Levy (HPL). Early findings suggest that purchases of higher-sugar taxable beverages fell and purchases of no- and lower-sugar beverages increased, alongside significant reductions in the sugar content of overall beverage purchases. However, underlying these changes are consumption shifts as well as product reformulation and changes in producers’ product portfolios. Drawing on a household scanner dataset, this study employed a descriptive approach to decompose changes in the sugar content of households’ non-alcoholic beverage purchases into producer factors (reformulation and product entry and exit) and consumer factors (product switching and volume changes as a result of price changes, changing preferences, or other factors). We look at these factors as the tax was announced and implemented across a sample of over 3000 South African households, and then by Living Standard Measures (LSM) groups (middle vs. high). The sugar content of beverage purchases fell by 4.9 g/capita/day overall, a 32% decrease. Taken in isolation, consumer switching and volume changes together led to a reduction equivalent to 71% of the total change, while reformulation accounted for a decrease equal to 34% of that change. Middle-LSM households experienced larger reductions than high-LSM households due to larger changes on the consumer side. For both LSM groups, reformulation-led reductions mostly occurred after implementation, and most changes came from taxable beverage purchases. As sugary drink tax designs evolve with broader implementation globally, understanding both supply- and demand-side factors will help to better assess the population and equity potential of these policies.

1. Introduction

Globally, excess sugar consumption is a major risk factor for obesity and related non-communicable diseases (NCDs) like type 2 diabetes, hypertension, liver and kidney damage, heart disease, and some cancers (World Cancer Research Fund International, 2015; World Health Organization and United Nations Development Programme, 2017; Malik and Hu, 2019; Malik et al., 2019; World Health Organization, 2021). In South Africa, these diseases are among the top ten causes of death (Statistics South Africa, 2021), obesity being one of the top five risk factors for early death (Ritchie and Roser, 2017). Indeed, obesity rates in South Africa are the highest in Sub-Saharan Africa and continue to increase rapidly (Ritchie and Roser, 2017). Nearly 40% of women and 11% of men have obesity in South Africa, and 69% of women and 39% of men have overweight or obesity (NCD Risk Factor Collaboration, 2016; Shisana O et al., 2013). Moreover, obesity among children aged 5–19 years increased in every region of the world from 1975 to 2016, with the largest growth in South Africa (about 400%) (Abarca-Gémez et al., 2017). The 2012 South African National Health and Nutrition Examination Survey (SANHANES) found combined prevalence of overweight and obesity of 14.2% in 6–14-year-olds and an increase in overweight among 2–5-year-olds from 10.6% to 18.2% compared to 2005 (Shisana O et al., 2013). If these trends continue, children with obesity are likely to face overweight and obesity as teenagers and adults (Singh et al., 2008; Freedman et al., 2005; Wang and Lobstein, 2006; Serdula et al., 1993; Power et al., 1997), develop NCDs at a younger age and face shorter life expectancies (Power et al., 1997; Harvard T.H. Chan School Of Public Health, 2021; World Health Organization, 2021, 2016; Sun et al., 2008; Must and Strauss, 1999; Reilly and Kelly, 2011; Olshansky et al., 2005; Daniels, 2009).

The World Health Organization, the World Cancer Research Federation and other international groups recommend that total sugar be limited to 10% or less of energy intake (World Cancer Research Fund International, 2015; World Health Organization and United Nations Development Programme, 2017). However, sugar consumption from sugary drinks is increasing globally (Singh et al., 2015; Popkin and Hawkes, 2016). In South Africa in particular, sugary drink sales grew by 6% annually between 2006 and 2020 (authors’ own calculation using Euromonitor data, and aligned with past studies (Popkin and Hawkes, 2016)). Sugary drinks are a significant source of sugar and are particularly harmful to the body because its liquid form allows the sugar to be absorbed more quickly by the liver and processed in a way that increases fat and glycogen deposits (Malik and Hu, 2015; Sundborn et al., 2019; Stanhope et al., 2018, 2011; Stanhope, 2012), which can lead to fatty liver disease and increase risks for diabetes and other NCDs (Stanhope et al., 2018; Jensen et al., 2018). Additionally, consumption of caloric beverages is typically not compensated for in subsequent or simultaneous food intakes, leading to higher total energy intakes, in particular when SSBs are consumed given their high energy contents (Appelhans et al., 2013; Maersk et al., 2012; Flood et al., 2006; DellaValle et al., 2005; Almiron-Roig and Drewnowski, 2003; Van Wymelbeke et al., 2004). Sugary drinks are also particularly bad for dental health (Chi and Scott, 2019; Valenzuela et al., 2020; Bridge et al., 2019), and can contribute to undernutrition when consumed in place of foods or drinks with greater micronutrient density, worsening undernutrition and stunting (Marriott et al., 2007; Zehner, 2016; Jaacks et al., 2017; Audain et al., 2019; Pries et al., 2019; Pries et al., 2019; Nordhagen et al., 2019).

In this context, the South African government announced in February 2016 its intention to adopt a tax on sugary beverages. After consultations, the Health Promotion Levy (HPL) became law in December 2017 and came into effect in April 2018. The HPL is payable by producers and importers of sugary beverages (Supplemental Table S3) at a rate of 2.1c per gram of total sugar in excess of 4 g/100 ml, which corresponds to an effective tax rate of approximately 10% of the average retail price of the most popular soft drink brand (Stacey et al., 2019). Its rate has not changed since its introduction.

The decision to tax sugar content was explicitly motivated by the intention to incentivize producers to reformulate beverages or introduce new ones with less or no sugar, in addition to nudging consumers to reduce their consumption of sugary beverages (National Treasury, 2016). However, it is worth noting that there is no requirement to inform consumers of the tax at point of purchase, which has been shown to be a salient behavioral factor (Doble et al., 2020). Despite this, the HPL attracted substantial media attention before and after its implementation (Essman et al., 2021), helping to raise awareness about the health impacts of sugary beverages (Murukutla et al., 2020). In its first two years of existence, it raised R5.8 billion of tax revenue, accounting for about 0.2% of the consolidated budget revenue each year (National Treasury, 2020, 2021).

Evidence strongly suggests that the HPL has led to price increases for taxable carbonates relative to nontaxable beverages and to declines in sugar-sweetened beverage (SSB) purchases and in the sugar content of those purchases, leading to reduced intakes (Hofman et al., 2021). Specifically, constant (December 2016) prices of taxable carbonates (the largest category of taxable beverages) were estimated to have increased on average by R1 per liter as a result of the tax (roughly 6% of the average price over the period from January 2013 to March 2019), whereas price changes did not appear to differ from trends for bottled water, 100% juice, and noncarbonated taxable beverages (Stacey et al., 2019). Interestingly, price changes were similar for low- and high-sugar carbonates, and reformulated beverages sold at a premium (average increases of R1.08 and R1.16 per liter for all reformulated beverages and beverages reformulated below the 4 g/100 ml threshold, respectively). Reductions in SSB purchases and consumption and associated declines in sugar have been documented in multiple studies. Relative to pre-announcement trends, the volume of taxable beverages purchased was estimated to have fallen by 29% in urban households nationally, with sugar from these purchases falling by 51% (Stacey et al., 2021). These decreases were slightly larger at lower socioeconomic levels: 32% and 57%. Diet recall data from convenient samples in Langa, Cape Town, showed a 37% reduction in the volume of taxable beverages consumed and a 31% reduction in sugar intake from these beverages between February-March 2018 and the same period a year later (Essman et al., 2021). Similarly, a longitudinal survey of adolescents and adults in Soweto, Johannesburg, showed that SSB intake frequency fell by two times per week in medium-intake consumers and by seven times per week in high-intake consumers 12 and 24 months after the HPL came into effect (Wrottesley et al., 2021).

While these studies have found evidence of reformulation, the full extent of producer and consumer responses (reformulation and product entry and exit on the producer side, switching between products and shifts in beverage demand on the consumer side) as drivers of change in sugar intake from beverages is still unknown. This gap is an important one for at least three reasons. First, as mentioned above, the HPL was designed to reduce sugar intake from beverages by raising SSB prices and incentivizing reformulation and the introduction of new beverages with less or no sugar. To what extent consumer and producer factors contributed to reducing the sugar content of beverage purchases remains to be seen. Second, we believe it is important to monitor reformulation in the beverage market because of the lack of conclusive evidence on the health effects of non-sugar sweeteners in humans. While we did not look at alternative sweeteners directly, sugar-based reformulations may increase exposure to such sweeteners, as producers seek to minimize taste alterations to their products. Third, a better understanding of these drivers of change in sugar intake from beverages can shed light on differences in outcomes for various subpopulations in the context of the HPL and similar taxes. In particular, it is of interest to probe how these factors contributed to differences in the magnitude of changes by socioeconomic level.

This paper aims to address this gap by decomposing the changes in the sugar content of beverage purchases after the announcement and implementation of the HPL into their producer components (reformulation and net product entry) and consumer components (product switching and changes in volumes purchased), overall and by socioeconomic level. Although each of these components depends on structural factors such as prices (which influence demand) and consumer preferences (which influence supply), our goal was to decompose these changes in closed form. While we did not attempt to establish causality, findings are discussed in light of previous trends and contextual factors.

This paper also adds to the broader literature on SSB taxes. To-date, studies on SSB taxes have examined their impacts on prices, purchases or consumption of SSBs and of sugar from SSBs, and long-term health effects in multiple countries and jurisdictions (Backholer et al., 2017, 2018; Krieger et al., 2021; Redondo et al., 2018; Teng et al., 2019). Demand elasticities, externalities, internalities, regressivity, and other topics have also attracted attention (Allcott et al., 2019). By contrast, research on SSB taxes and reformulation is limited (Vandevijvere and Vanderlee, 2019) but interest in this topic is growing. This may be particularly relevant in cases where there is a lag between the announcement of a tax or similar policy and its implementation. A recent study examined the impact of the UK’s Soft Drink Industry Levy (SDIL), which had a two-year gap between its announcement and implementation, on the composition of beverage selections at leading supermarkets. It found that the proportion of high-sugar beverages fell gradually after the SDIL was announced, more rapidly closer to implementation, and more slowly afterwards, while the proportion of exempt fruit juices and milk-based drinks barely changed (Scarborough et al., 2020). Reported evaluation results on Portugal’s SSB tax show shifts in the share of low- and high-sugar beverages consumed nationally, but do not disentangle reformulation, net product entry, and switching effects (Goiana-da-Silva et al., 2018). (Reformulation is estimated to have led to an 11% reduction in total energy intake from SSBs but details are absent.) Lastly, structural models for ex-ante analyses of the impact of taxes on nutrients of concern are starting to incorporate reformulation effects (Allais et al., 2020).

Additionally, this paper makes a methodological contribution to food policy studies that have used the decomposition applied here (Griffith et al., 2017; Spiteri and Soler, 2018). Decompositions applied to-date have focused on changes in the concentration of a population’s food or beverage purchases in some ingredient of concern, say grams of sugar per 100 g or 100 ml. However, this ignores changes in quantities purchased, which also contribute to how much of that ingredient people purchase. Building an extension of this decomposition, we were able to analyze changes in the sugar content of beverage purchases (i.e. the amount of sugar from beverage purchases).

2. Data and methods

2.1. Data

We used household purchase data from Europanel for the period January 2014 to March 2019, collected from approximately 3000 households in South Africa each year. Panel members recorded all items purchased and taken home by barcode using scanners and a barcode booklet provided to them for products without barcodes. The sample includes urban and rural households but excludes households with a Living Standards Measure (LSM) under 4 (5–10% of the South African population; see Supplemental Table S1) due to the potential lack of electricity required for recording purchases. LSM is a widely used market research approach to stratifying South Africa’s population and is derived from a set of questions about household ownership of goods, access to water, electricity, media, financial services, area and type of residence, education and income.

Households were recruited by telephone, text, or online, with poor reporters (less than either five categories of items purchased or one shopping trip per week) dropped on a rolling basis followed by targeted replenishment based on sociodemographic attributes. As a result, 38% of households in the sample were present in six months or less, 54% in 12 or less, and 72% in at most 24. The weighted sample was designed to represent 13.7 million households, or 42–45 million people from all nine provinces, covering approximately 90–95% of the country’s population. Supplementary Table S2 compares the weighted survey demographic measures with Statistics South Africa’s General Household Survey (GHS) for 2014, 2016, and 2018. The average household head is slightly younger in the weighted sample, which also has a slightly higher share of households in the 6–9 LSM range and thus a slightly lower share of households in the 4–5 range. Otherwise, the weighted sample is broadly in line with the GHS.

To determine the nutritional content of these purchases over time, we merged that data with nutrition facts data from several sources. These include: the Mintel Global New Product Database (2015–2019); data collected by Discovery Vitality and The George Institute (2015–2017); and data collected by our own research team in 2018 and 2019 following a standardized protocol for data collection, entry and review to exclude implausible or inconsistent values. The purchase and nutrition facts data were matched on barcode and date to maximize accuracy in how beverages were categorized as taxable or non-taxable based on the South Africa Revenue Service (SARS) regulations on tariff codes. Since the tariff codes are broad for some categories (e.g., carbonates), there are beverages under taxable tariff codes that have tax rates of R0 (e.g., diet carbonates with sugar concentrations not exceeding 4 g/100 ml); we consider them as taxable in this analysis. Supplemental Table S3 describes the beverage categories included in this study, alongside their taxable status and corresponding tariff codes.

2.2. Methods

To disentangle consumer and producer pathways of dietary change after the announcement and implementation of the HPL, we decomposed the changes in the sugar content of beverage purchases between the pre-announcement and interim periods and between the interim and post-implementation periods. From these, we show the cumulative changes and contributions (we also decomposed changes between the pre-announcement and post-implementation periods, but the results were qualitatively similar to cumulative changes). These decompositions proceeded in two steps. In the first step, we decomposed changes in the sugar concentration of beverage purchases (in grams of sugar per 100 ml) into consumer and producer components, using an accounting decomposition from the economics literature (Foster et al., 2001) that has found recent applications in nutrition research (Griffith et al., 2017; Spiteri and Soler, 2018). We then nested these components into second-stage decompositions of changes in sugar content (in grams of sugar per capita per day), accounting for changes in the volume of beverage purchases.

2.2.1. Decomposing changes in the sugar concentration of beverage purchases

In this first step, the sugar concentration of all beverage purchases in period t can be expressed as:

$$C_t=\sum _i{w}_{i,t}{c}_{i,t}$$ (1)

where i indexes beverages, and w_i,_t and c_i,_t are beverage i’s share of total volume purchased and sugar concentration in t, respectively. This quantity can change through multiple ways: product reformulation can result in higher or lower sugar concentrations, consumers can switch between products, and producers can discontinue products and introduce new ones with different sugar concentrations. To identify these pathways, we start by writing the change in C between t = 0 and t = 1 as:

$$C_1-C_0=\sum _i{w}_{i,1}{c}_{i,1}-\sum _i{w}_{i,0}{c}_{i,0}=\sum _{i\in \mathcal{C}}{w}_{i,1}{c}_{i,1}+\sum _{i\in \mathcal{N}}{w}_{i,1}{c}_{i,1}-\sum _{i\in \mathcal{C}}{w}_{i,0}{c}_{i,0}-\sum _{i\in \mathcal{X}}{w}_{i,0}{c}_{i,0}$$ (2)

where $\mathcal{C}$, $\mathcal{N}$ and $\mathcal{X}$ are the sets of continuing, new and discontinued products. Adding and subtracting terms (see (Griffith et al., 2017) for details) yields:

$$C_1-C_0=\underset{\text{Reformulation}}{\underbrace{\sum _{i\in \mathcal{C}}{w}_{i,0}\left(c_{i,1}-c_{i,0}\right)}}+\underset{\text{Consumer}\text{switching}}{\underbrace{\sum _{i\in \mathcal{C}}\left(c_{i,0}-C_0\right)\left(w_{i,1}-w_{i,0}\right)}}+\underset{\text{Cross}\text{term}}{\underbrace{\sum _{i\in \mathcal{C}}\left(c_{i,1}-c_{i,0}\right)\left(w_{i,1}-w_{i,0}\right)}}+\underset{\text{Net}\text{product}\text{entry}}{\underbrace{\sum _{i\in \mathcal{N}}{w}_{i,1}\left(c_{i,1}-C_0\right)-\sum _{i\in \mathcal{X}}{w}_{i,0}\left(c_{i,0}-C_0\right)}}$$ (3)

The cross term is a remainder term that results from holding volume shares and sugar concentrations constant to obtain the reformulation and consumer switching components, respectively. A positive (negative) cross term indicates that volume shares and sugar concentrations tended to move in the same (opposite) direction(s).

2.2.2. Decomposing changes in the sugar content of beverage purchases

To go from sugar concentration to sugar content, we must further account for changes in the volume of beverage purchases. The sugar content of all beverage purchases in t, i.e. the total amount of sugar contained in all beverages purchased, is simply the total volume purchased in t, V_t, multiplied by the sugar concentration of these purchases, C_t. Dividing by population size and the number of days in t yields beverage purchases’ sugar content on a per-capita-per-day basis. For exposition purposes and without loss of generality, we assume that these two divisors are constant across periods and normalize them to one. Adding and subtracting terms, the change in V_tC_t between t = 0 and t = 1 can be decomposed as:

$$V_1{C}_1-V_0{C}_0=\left(V_1-V_0\right)C_0+V_0\left(C_1-C_0\right)+\left(V_1-V_0\right)\left(C_1-C_0\right)$$ (4)

with C₁ −C₀ decomposed into its underlying components (Eq. 3). The first term represents the impact of volume changes had sugar concentration not changed. The second term is the part of the change attributable to sugar concentration changes (reformulation, consumer switching, and net product entry) had volumes remained constant. Like in the decomposition of sugar concentration changes, a residual term arises because both volume and sugar concentration are allowed to vary. However, as changes in the volume of all beverage purchases may be small and mask shifts across taxable categories, we further decomposed each of these components into their taxable and nontaxable subcomponents, as explained in the methodological supplement.

2.2.3. Empirical application

We could not identify individual products over time due to inconsistencies in how product characteristics (e.g., product description) were recorded both within and between survey years and across nutrition label sources. Therefore, we carried out the decompositions on beverage items, i.e., sets of beverages in the same beverage category (e. g., 100% juice), brand, sub-brand, and taxable category. An example of a beverage item is regular Coca-Cola, which includes different flavors (original, vanilla, etc.) and package sizes (e.g., 1.5-liter bottle, multipacks of 12 cans). We sorted by taxable category because 2% of beverage category, brand, and sub-brand combinations had both taxable and non-taxable beverages due to further variety of products within them. Importantly, taxable category does not refer to whether a beverage is actually taxed but to whether the HPL applies to that beverage category, so it is not affected by reformulation (see Supplemental Table S3 for details on taxable and non-taxable categories). Beverage items were classified as continuing between two periods if they were purchased in both periods, new if they were only purchased in the end period, and discontinued if they were only purchased in the base period.

An imperfect proxy for the number of distinct products (where we do not differentiate by package size) in each beverage item is the number of distinct product identifiers per item. These product identifiers are generated by Kantar Europanel based on label information (e.g., barcodes). For short, we will refer to these identifiers as barcodes, although not all of them are (e.g., stock keeping units). Each barcode was assigned to a single product (as confirmed by manual inspection of the product descriptions of all duplicate barcodes). However, the reverse is not true: some product descriptions as well as different package sizes of the same products mapped to multiple barcodes. Therefore, the number of distinct barcodes per beverage item provides an upper bound for the number of distinct products per item. Supplemental Table S4 provides summary statistics for the number of distinct barcodes per beverage item by period. In total, there were on average 20 distinct barcodes per beverage items. To further validate our unit of observation, we calculated the standard deviation of sugar concentration across barcodes for each beverage item in 2014, 2016, and 2018 (rather than by period, as reformulation at the barcode level between years contributes to variation within periods, each of which spans more than one year). Supplemental Table S5 presents summary statistics for this measure of sugar concentration heterogeneity within beverage items. In each year, sugar concentration was virtually constant across barcodes within 50% of beverage items, and had a standard deviation of close to or less than 2 g/100 ml for 90% of them, attesting to a high degree of homogeneity within beverage items overall.

Nevertheless, this choice of unit of observation implies that all decomposition components refer to beverage items rather than individual products. For example, the replacement of an old product by a new product within the same beverage item would be captured by the reformulation component rather than by the net product entry component. Likewise, if a household substituted a beverage for another within the same beverage item, this change would be captured by the reformulation component.

The sugar concentration of a beverage item in a given period was obtained by summing its sugar content across all purchases of that item (accounting for multiple purchases) during that period and dividing by the total volume of these purchases. These sums were survey-weighted and reconstituted volume was used for concentrates. Total volume and volume shares were obtained analogously. Due to attrition and replenishment of the sample, population size in a given period was obtained as the average of the weighted sums of household size in each month during that period. Finally, to ensure that each quarter was equally represented within and across periods and thereby avoid seasonality imbalances, we did not use data from the first quarter of 2016 (out of 21 quarters of data in total).

2.2.4. Income level analyses

To see if the pathways of change in the sugar content of beverage purchases varied by socio-economic level, we conducted additional decompositions for two LSM subpopulations: 4–6 (middle LSM) and 7–10 (higher LSM). This creates a discrepancy in the decompositions because not all continuing beverage items were purchased in each period by both LSM groups, and similarly not all new (discontinued) items were purchased by both groups in the periods they were introduced (discontinued). We refer to this discrepancy as the stratification component.

2.2.5. Sensitivity analyses

We conducted two sets of sensitivity analyses. First, we decomposed changes between years to see if results were qualitatively similar overall and around announcement and implementation. For these decompositions, we used all available data until the end of 2018 (we did not use 2019 data because only the first quarter was available). Second, as sugar content was imputed for 5% of all beverage purchases and 2% remained missing (see the methodological supplement for details), we repeated the analysis for energy, which was available for virtually all observations.

3. Results

3.1. Descriptive statistics

Tables 1 and 2 briefly describe our beverage item dataset. Table 1 shows the frequencies of continuing, new, and discontinued beverage items by taxable category and period, as well as the total number of beverage items by taxable category in all periods combined.

Table 1.

Continuing, new, and discontinued beverage items by taxable category.

	Pre-announcement	Interim	Post-implementation	Total
Taxable
Continuing	−	497	486	−
New	−	213	56	−
Discontinued	−	132	224	−
Total taxable	629	710	542	889
Nontaxable
Continuing	−	478	410	−
New	−	146	64	−
Discontinued	−	128	214	−
Total nontaxable	606	624	474	803
All
Continuing	−	975	896	−
New	−	359	120	−
Discontinued	−	260	438	−
Total	1235	1334	1016	1692

Note: Pre-announcement is January 2014-December 2015, interim is April 2016-March 2018, and post-implementation is April 2018-March 2019 (January-March 2016 excluded; see the methods section for details). Continuing items are items that were purchased in the current and last periods; new items are items that were purchased in the current period but not last period; discontinued items are items that were purchased last period but not in the currentperiod, and are therefore excluded from a period’s totals. Only totals are displayed for the pre-announcement period because without prior data continuing and new items are indistinguishable and discontinued items are not observed.

Source: Authors’ calculations based on household purchase data from South Africa Europanel (January 2014-March 2019).

Table 2.

Reformulation of continuing beverage items by taxable category.

	Interim		Post-implementation
	N	%	N	%
Taxable
Sugar concentration unchanged	433	87.1	366	75.3
Lower sugar concentration	26	5.2	83	17.1
Higher sugar concentration	38	7.6	37	7.6
Nontaxable
Sugar concentration unchanged	438	91.6	362	88.3
Lower sugar concentration	29	6.1	25	6.1
Higher sugar concentration	11	2.3	23	5.6
All
Sugar concentration unchanged	871	89.3	728	81.3
Lower sugar concentration	55	5.6	108	12.1
Higher sugar concentration	49	5.0	60	6.7

Notes: Interim is April 2016-March 2018 and post-implementation is April 2018-March 2019 (January-March 2016 excluded; see the methods section for details). Reformulation defined as changes in sugar concentration (rounded to the nearest 0.1 g/100 ml) greater than 20%.

Source: Authors’ calculations based on household purchase data from South Africa Europanel (January 2014-March 2019).

Our dataset covered 1692 beverage items, roughly evenly split between taxable categories in each period. Of note, new beverage items accounted for smaller shares of the numbers of distinct items purchased in the post-implementation period (56 of 542 taxable items, or 10%, and 64 of 474 nontaxable items, or 14%) compared to the interim period (30% and 23%, respectively). By contrast, larger shares of beverage items were discontinued between the interim and post-announcement period (224 of 710 taxable items, or 32%, and 214 of 624 nontaxable items, or 30%) than between the pre-announcement and interim periods (21% and 20%, respectively).

Table 2 reports the frequencies of non-reformulated and reformulated continuing beverage items by taxable category and period. The pre-announcement period is omitted because without prior data, continuing and new items are indistinguishable and there is no information on past sugar concentrations. Changes in the sets of individual products that make up beverage items can lead to small changes in sugar concentration, so for the purpose of this table we defined reformulation as a change in sugar concentration greater than ± 20%, which is in line with current ranges allowed for margins of error in reporting sugar values in the United States (Center for Food Safety and Applied Nutrition, 2021). We also rounded sugar concentrations to the nearest 0.1 g/100 ml to mitigate the influence of large percentage changes from low bases on these frequencies.

Most continuing beverage items were not reformulated. However, the share of taxable items that were reformulated down (lower sugar concentration) increased from 5.2% in the interim period to 17.1% in the post-implementation period, while the share of items that were reformulated up (higher sugar concentration) remained at 7.6%. By contrast, the share of nontaxable beverage items that were reformulated down remained at 6.1%, while the share of items that were reformulated up increased slightly from 2.3% to 5.6%.

The sugar concentrations of taxable and non-taxable beverage purchases, as well as their volumes per capita per day and the resulting sugar contents per capita per day are available in Supplemental Table S6, with a breakdown by LSM level.

3.2. Decomposition results: all purchases

In what follows, a beverage or product means a beverage item as defined in the methods section. Keeping in mind the caveats associated with that unit of observation, the decomposition results for all households are reported in Table 3 and shown in Fig. 1. The sugar content of beverage purchases fell by 1.7 g/capita/day between the pre-announcement and interim periods (from 15.2 to 13.5 g/capita/day; Supplemental Table S6), and by 3.1 g/capita/day between the interim and post-implementation periods (from 13.5 to 10.4 g/capita/day), totaling a 4.9 g/capita/day overall decrease. Switching accounted for large proportions of these reductions: 44.8% (−0.8 g/capita/day) between the pre-announcement and interim periods, and 36.9% (−1.1 g/capita/day) between the interim and post-implementation periods. That is, in the absence of changes in the volume of beverage purchases, sugar concentrations, and the beverage mix, the same changes in consumer choices as were observed would have led to a 0.8 g/capita/day reduction in the sugar content of beverage purchases after announcement, and a further 1.1 g/capita/day decrease after implementation. Also important through both transition periods were changes in the volume of beverage purchases, which accounted for 38.8% (−0.7 g/capita/day) and 26.3% (−0.8 g/capita/day) of these reductions, respectively. By contrast, reformulation accounted for a relatively small proportion of the post-announcement change (17.6% or −0.3 g/capita/day), yet contributed the largest share of the change after the HPL came into effect (43.6% or −1.4 g/capita/day).

Table 3.

Changes in the sugar content of non-alcoholic beverage purchases (g/capita/day).

	Pre-interim		Interim-post		Cumulative
	Change	% of net change	Change	% of net change	Change	% of net change
Volume	−0.7	38.8	−0.8	26.3	−1.5	30.8
Reformulation	−0.3	17.6	−1.4	43.6	−1.7	34.3
Switching	−0.8	44.8	−1.1	36.9	−1.9	39.7
Net entry	−0.1	3.1	0.0	−0.1	−0.1	1.0
Cross terms	0.1	−4.3	0.2	−6.6	0.3	−5.8
Net change	−1.7	100.0	−3.1	100.0	−4.9	100.0

Notes: Pre-announcement is January 2014-December 2015, interim is April 2016-March 2018, and post-implementation is April 2018-March 2019 (January-March 2016 excluded; see the methods section for details). Each row represents the change in the sugar content of non-alcoholic beverage purchases attributable to the corresponding decomposition factor. ‘Cross terms’ captures the cross terms correcting for simultaneous volume and sugar concentration changes.

Source: Authors’ calculations based on household purchase data from South Africa Europanel (January 2014-March 2019).

Fig. 1.

Changes in the sugar content of non-alcoholic beverage purchases (g/capita/day). Notes: Pre-announcement is January 2014-December 2015, interim is April 2016-March 2018, and post-implementation is April 2018-March 2019 (January-March 2016 excluded; see the methods section for details). Decompositions of changes in the sugar content of non-alcoholic beverage purchases between the pre-announcement, interim, and post-implementation periods into parts attributable to changes in the volume of purchases, reformulation of beverage items (sets of closely related products as described in the text), consumer switching between beverage items, net entry of beverage items, and a correction for simultaneous changes in volumes and sugar concentrations. The pre-post decomposition is for the change between the pre-announcement and post-implementation periods. The cumulative decomposition adds each decomposition factor over time.

Source: Authors’ calculations based on household purchase data from South Africa Europanel (January 2014-March 2019).

On a cumulative basis, switching, reformulation, and changes in the volume of beverage purchases accounted for 39.7%, 34.3% and 30.8% of the total change, respectively. (These shares sum to more than 100% because each is derived assuming that the other shares are zero.) Probably due to our inability to track individual products over time, net product entry appears to have contributed little to these changes. The cross terms are positive, reflecting that both the volume and sugar concentration of beverage purchases declined. Notably, these reductions mostly came from purchases of taxable beverages (Supplemental Table S7), with only small changes in the sugar content of nontaxable beverage purchases. When we decomposed changes between the pre-announcement and post-implementation periods, but the results were qualitatively similar to cumulative changes.

3.3. Decomposition results by LSM level

Table 4 and Fig. 2 reveal considerable differences by LSM level. First, middle-LSM households (LSM 4–6) experienced larger declines in the sugar content of their beverage purchases than high-LSM households (LSM 7–10), in both absolute and relative terms and through both transitions periods. Between the pre-announcement and post-implementation periods, middle-LSM households saw the sugar content of their beverage purchases fall by as much as 6.4 g/capita/day (from 16.2 to 9.8 g/capita/day, or −39.3%; Supplemental Table S6), compared to 2.8 g/capita/day (from 14 to 11.3 g/capita/day, or −20%) for their high-LSM counterparts.

Table 4.

Changes in the sugar content of non-alcoholic beverage purchases by LSM level (g/capita/day).

	Pre-interim		Interim-post		Cumulative
	Change	% of net change	Change	% of net change	Change	% of net change
LSM 4–6
Volume	−0.8	33.7	−1.5	38.1	−2.3	36.4
Reformulation	−0.4	15.6	−1.5	37.7	−1.8	29.0
Switching	−1.4	56.2	−1.4	35.7	−2.8	43.7
Net entry	0.0	1.1	0.0	−0.6	0.0	0.1
Other	0.2	−6.5	0.4	−11.0	0.6	−9.2
Net change	−2.5	100.0	−3.9	100.0	−6.4	100.0
LSM 7−10
Volume	−0.5	64.5	0.2	−10.9	−0.3	11.6
Reformulation	−0.2	24.8	−1.2	62.2	−1.4	51.0
Switching	0.0	−2.2	−0.9	45.8	−0.9	31.5
Net entry	−0.1	10.8	0.0	1.2	−0.1	4.1
Other	0.0	2.1	0.0	1.7	0.0	1.8
Net change	−0.8	100.0	−1.9	100.0	−2.8	100.0

Notes: Pre-announcement is January 2014-December 2015, interim is April 2016-March 2018, and post-implementation is April 2018-March 2019 (January-March 2016 excluded; see the methods section for details). Each row represents the change in the sugar content of middle-LSM households’ and high-LSM households’ non-alcoholic beverage purchases attributable to the corresponding decomposition factor. ‘Other’ captures the cross terms correcting for simultaneous volume and sugar concentration changes and the stratification term correcting for discrepancies in purchases of continuing, new, and discontinued beverages across LSM groups.

Source: Authors’ calculations based on household purchase data from South Africa Europanel (January 2014-March 2019).

Fig. 2.

Changes in the sugar content of non-alcoholic beverage purchases by LSM level (g/capita/day). Notes: Pre-announcement is January 2014-December 2015, interim is April 2016-March 2018, and post-implementation is April 2018-March 2019 (January-March 2016 excluded; see the methods section for details). Decompositions of changes in the sugar content of middle-LSM households’ and high-LSM households’ non-alcoholic beverage purchases between the pre-announcement, interim, and post-implementation periods into parts attributable to changes in the volume of purchases, reformulation of beverage items (sets of closely related products as described in the text), consumer switching between beverage items, net entry of beverage items, and a correction for simultaneous changes in volumes and sugar concentrations. The pre-post decomposition is for the change between the pre-announcement and post-implementation periods. The cumulative decomposition adds each decomposition factor over time.

Source: Authors’ calculations based on household purchase data from South Africa Europanel (January 2014-March 2019)

Second, the factors underlying these reductions also differ. Among middle-LSM households, switching had the largest cumulative share of the total change at 43.7% (−2.8 g/capita/day), followed by volume reductions (36.4% or −2.3 g/capita/day) and reformulation (29% or −1.8 g/capita/day). By contrast, reformulation played the largest role among high-LSM households (51% or −1.4 g/capita/day), followed by switching (31.5% or −0.9 g/capita/day) and volume reductions (11.6% or −0.3 g/capita/day). Interestingly, while post-announcement reductions in sugar content attributed to volume changes and reformulation were similar across LSM levels, middle-LSM households switched towards lower-sugar beverages in both the interim and post-implementation periods, whereas high-LSM households only did so after the HPL came into effect. After implementation, volume changes, reformulation, and switching accounted for almost equal sugar content reductions among middle-LSM households but not among high-LSM households, who purchased more beverages overall, offsetting some of the reductions brought by reformulation and switching.

While most of these changes also came from taxable beverage purchases, it is worth noting that for both LSM groups, decreases in sugar content attributed to reductions in the volume of taxable beverage purchases were slightly offset by increases in the volume of nontaxable beverage purchases, although some reformulation of nontaxable beverages appears to have countered that effect in whole or in part (Supplemental Table S7). Interestingly, middle-LSM households tended to switch towards both taxable and nontaxable beverages that were lower in sugar, whereas high-LSM households tended to switch towards lower-sugar taxable beverages but higher-sugar nontaxable beverages. Finally, we did not find evidence that producers reformulated beverages purchased by middle-LSM households and beverages purchased by high-LSM households differently, or that the difference in sugar concentration between new and discontinued beverages differed by LSM level (Supplemental Table S8).

3.4. Sensitivity analyses

The sensitivity analyses results are shown in Supplemental Figs. S1–S4. Starting with the decompositions of annual changes in sugar content (Supplemental Figs. S1–S2), it is important to bear in mind that the sugar contents in the first period (pre-announcement for the decompositions by period, 2014 for decompositions by year) and the last period (post-implementation and 2018, respectively) are not equivalent, leading to slightly different cumulative changes. With that caveat, the cumulative changes in sugar content were slightly smaller in the decompositions by year. The cumulative shares for the full sample were similar, with switching, reformulation, and volume changes bringing decreases equivalent to 42% (slightly more), 38% (slightly more), and 25% (slightly less) of the net change, respectively. The shares did differ more substantially in the decompositions by LSM level. In the subsample of middle-LSM households’ purchases, the part of the total change attributed to switching increased to 51% (vs. 44%), while that attributed to volume changes decreased to 28% (vs. 36%) (reformulation did not change meaningfully). In the subsample of high-LSM households’ purchases, reformulation increased substantially to 84% (vs. 51%), volume changes decreased to 4% (vs. 12%), and switching changed sign, bringing increases in the sugar content of these purchases equivalent to 4% (0.04 g/capita/day) of the total change, compared to a decrease equivalent to 32% of the total change in the primary decomposition results. However, this opposite switching effect was concentrated in the 2014–2015 change, with switching acting to reduce sugar content in subsequent years. This could explain why this effect was not seen in the decompositions by period, since the pre-announcement period covers all of 2014 and 2015.

These analyses shed further light on the timing of reformulation processes, with very little to no reformulation prior to 2018, the year the HPL came into effect. It is also interesting to note that for both LSM groups, volume reductions impacted sugar content the most in 2017, the year between announcement and implementation.

The energy content results were largely consistent with the sugar content results (Supplemental Figs. S3–S4).

4. Discussion

We found that sugar content from beverage purchases fell from 15.2 g/capita/day before the HPL was announced to 13.5 g/capita/day after its announcement, and to 10.4 g/capita/day after it came into effect. For taxable beverage purchases alone, these figures are 13.1 g/capita/day, 11.3 g/capita/day, and 8.3 g/capita/day. To put this in context, a decrease in daily sugar intake of 5 g equates to a reduction in daily energy intake of 20 kcal. Simulation results from a dynamic model of human metabolism indicated that a reduction twice that amount could lead to an average weight loss of 1.8 kg over five years (Hall et al., 2011). Thus, if sustained, the observed reduction in sugar intake could at the very least help to prevent further weight gain. However, it is clear that significant reductions in the prevalence of overweight and obesity will require multiple policy instruments.

Previous analyses, albeit different in nature, of the same underlying data showed similar decreases in the average sugar content of taxable beverage purchases, in relative terms (Stacey et al., 2021). The same applies to findings by LSM. Diet recall data from 18 to 35-year-old middle LSM adults in Langa, Cape Town, before and a year after the implementation of the tax showed a 31% decrease in sugar intake from taxable beverages (Essman et al., 2021), which is also what we found for sugar content changes between the interim and post-implementation period for middle-LSM households (Supplemental Table S6). However, the relative change in sugar intake from nontaxable beverages was different: increasing by 35% (5.3 g/day), compared to virtually no change here. This difference may be due to a variety of factors, including different definitions of ‘taxable’ (we consider beverages in taxable categories and with under 4 g of sugar per 100 ml as taxable, while that study restricted ‘taxable’ to effectively taxed beverages), as well as differences between populations, purchase and intake measures (which may be starker for untaxed beverages, such as sweet tea prepared at home and some concentrates, which can be more or less diluted), and analytical approaches.

Decomposing these changes, we found that they were achieved by a mix of volume reductions, switching, and reformulation, particularly in the taxable category. Keeping in mind the descriptive nature of our analysis, it is worth noting that reformulation primarily occurred in 2018, year of the tax’s implementation. Media campaigns may explain consumer behavior before the tax came into effect (Essman et al., 2021; Murukutla et al., 2020). While it is unclear whether producers engaged in reformulation in response to the HPL or to changes in demand after it came into effect (e.g., as a result of higher SSB prices), these findings are consistent with existing evidence that national nutrition policies, including but not limited to taxes, can create incentives for producers to reformulate and reduce nutrients of concern if present in too high concentrations. For example, an analysis of packaged food and beverage products available in supermarkets in Santiago in Chile found evidence of reformulation following the implementation of front-of-pack nutrient warning labels and marketing restrictions for products high in nutrients of concern (Reyes et al., 2020). Indeed, options to revise Chile’s value-added tax structure are being discussed to further encourage reformulation, as well as signal to consumers to avoid products high in those nutrients.

Although not directly comparable to our findings given difference in measurement and sample, consumer choices were found to explain 71% (6.4 g/day) of the decrease in sugar intake in mostly middle-LSM 18–35 year old adults in Langa, Cape Town (Essman et al., 2021). Similarly, we found that volume changes and switching decreased the sugar content of middle-LSM households’ purchases of taxable beverages by 72% (2.8 g/capita/day) of the net change between the interim and post-implementation periods (Supplemental Table S7). In comparison, that figure is 43% (0.8 g/capita/day) for high-LSM households. Given the documented increases in taxable carbonates’ prices relative to pre-HPL trends compared to other beverages (Stacey et al., 2019), this difference is consistent with greater price sensitivity among middle-LSM households in South Africa (Burger et al., 2017).

On the producer side, we found that reformulation led to very similar decreases in the sugar concentration of beverage purchases (taxable and nontaxable) by both middle- and high-LSM households (−0.3 g/100 ml on a cumulative basis; Supplemental Table S8), indicating that despite clear differences in beverage purchases by LSM level (Supplemental Table S6), beverages purchased by each group were not reformulated differentially overall. However, it would be misleading to infer that exposure to alternative sweeteners after implementation was similar across LSM levels because there may be differences in the volume of reformulated beverages purchased. First, we did not examine alternative sweeteners directly, although reformulation is likely to have involved substitution of such sweeteners for sugar to mitigate taste alterations. Second, middle-LSM households purchased consistently more taxable beverages (and less nontaxable ones) than their high-LSM counterparts, in addition to relying more heavily on taxable beverages (Supplemental Table S6), likely because of higher average prices among nontaxable beverages. A decomposition focused on alternative sweeteners by LSM level would help to elucidate this question in future research.

This study has a number of limitations. First, as a descriptive study, it does not identify the causes of the observed reductions in the sugar content of beverage purchases, although it is reasonable to say that the HPL played some role. In particular, public debates and media campaigns about the HPL after it was announced may have contributed to these changes too, especially in light of some of the behavioral changes seen prior to implementation. Nevertheless, it is worth noting that most reformulations occurred after the tax came into effect. Furthermore, these reductions stand in sharp contrast to the steady rise in SSB purchases in South Africa, alongside overweight and obesity, in the period leading to the announcement of the HPL. Second, it was not possible to identify and track individual products due to data limitations, restricting our choice of unit of observation to sets of closely related products (same beverage category, brand, sub-brand, and taxable category). Although these sets of products were relatively homogeneous in terms of sugar concentration, net product entry and switching are likely somewhat understated, and reformulation somewhat overstated, since switching and net product entry within a given set of beverages that resulted in changes in the overall sugar concentration of the set would be captured as reformulation. Reassuringly, our findings align with the existing literature on the role of consumer and producer behaviors on changes in sugar intake from beverages a year after the HPL came into effect. Another limitation of this paper is that it did not examine potential compensation by means of increased purchases of sugary foods as substitutes for sugary beverages. However, available evidence suggests that such compensation may be limited. Evaluations of Seattle’s and Philadelphia’s SSB taxes in the US showed little to no substitution to sweet or salty snacks (Gibson et al., 2021; Oddo et al., 2021), in line with previous work on a hypothetical SSB tax in the US (Finkelstein et al., 2013). Cross-price elasticities estimates from Chile and the UK concord with this view (Guerrero-López et al., 2017; Smith et al., 2018). Similarly, our dataset only captures purchases brought home, raising the possibility of bias if purchases not brought home were systematically different. While it is important to be aware of that possibility, our results align well with findings from diet-recall data (Essman et al., 2021). Finally, our sample was restricted to households of middle to high LSM levels, lacking data on households representing 5–10% of all South African households, limiting somewhat the generalizability of our findings. Volume changes and switching may have played important roles at lower LSM levels due to price changes, although the price elasticity of demand for sugary drinks may also depend on consumption levels.

Despite its limitations, this study is the first, to our knowledge, to formally untangle producer and consumer factors behind changes in the sugar content of households’ beverage purchases as a tax on sugary drinks was announced and implemented. Indeed, it appears to be the first to examine these factors in relation to changes in the content (in grams per capita per day)—beyond concentration (in grams per 100 ml)—of food and beverage purchases in a nutrient or ingredient of concern, thereby incorporating changes in quantities purchased, either overall or in purchase groups of interest. It did so using a rich dataset of both urban and rural South African households’ beverage purchases matched to comprehensive nutrition facts data.

Future work could explore how the consumer and producer factors studied here compared in different segments of the sugary drink consumption distribution (see, e.g., (Ng et al., 2019; Valizadeh et al., 2020)). To help better understand how nutrition-related taxes impact population consumption patterns, further research could also apply the decomposition method used and enriched here to alternative tax designs (tiered, volumetric, ad valorem, or hybrid designs). Similarly, future work could apply these methods in other nutrition policy contexts where some degree of reformulation is incentivized, in particular warning labels and marketing regulations. Such policies have been implemented in Chile, Peru, Mexico, and Israel, and are gaining traction elsewhere.

5. Conclusions

The sugar content of South African households’ non-alcoholic beverage purchases fell by 4.9 g/capita/day as the HPL was announced and implemented (a 32% decrease). Taken in isolation, consumer switching and volume changes together led to a reduction equivalent to 71% of the total change, while reformulation accounted for a decrease equal to 34% of that change. Examining beverage purchases by LSM level, middle-LSM households experienced larger reductions than high-LSM households, which were mostly driven by larger changes on the consumer side. In particular, middle-LSM households switched to lower-sugar beverages after both the announcement and implementation of the HPL, whereas high-LSM households only did so after it came into effect. Moreover, volume reductions were larger and more consistent in middle-LSM households’ purchases. For both LSM groups, reformulation-led reductions mostly occurred after implementation, and most changes came from taxable beverage purchases. More work is needed to assess whether these reductions have been sustained, amplified, or attenuated, and the roles of consumer and producer factors in the longer term.