We performed a quantitative content analysis of South African newspapers, following previous methods developed for systematic search and coding of news media content related to SSB taxes (25,26). Our article search strategy, codebook development, and analysis are outlined below.
Sample Selection
Search strategy
We selected online news articles covering the HPL using two databases: (1) Nexis Uni and (2) ProQuest Central, both global databases that provide access to full text news articles. We searched for articles using the following search string:
("Sugary beverages" OR "sugar-sweetened beverages" OR "sugar sweetened beverages" OR “health promotion levy”) AND (Levy OR Levies OR Tax OR Taxes OR Taxation OR Legislat*) AND (“South Africa” OR “Eastern Cape” OR Free State OR Gauteng OR “KwaZulu-Natal” OR Limpopo OR Mpumalanga OR “North West” OR “Northern Cape” OR “Western Cape” OR Bhisho OR Bloemfontein OR Johannesburg OR Pietermaritzburg OR Polokwane OR Nelspruit OR Mahikeng OR Kimberley OR Cape Town OR Port Elizabeth OR Durban OR Rustenburg OR Soweto OR Pretoria OR “Mitchells Plain” OR Umlazi OR Katiehong OR Tembisa OR Khayelitsha OR Soshabguve OR Mamelodi OR Ibhayi OR Tshivhase OR Sebonkeng OR Mabopane OR Chatswork))
South African city search terms were included to potentially pick up smaller, local papers published in South Africa. We did not restrict to particular newspapers and included any South African newspaper source captured by the Nexis Uni or ProQuest Central databases. Supplementary Table 1 displays the study sample with circulation numbers, which includes most of the major English language newspaper publishers in South Africa (30).
Eligibility Criteria
Articles were included for analysis if they were published between January 1, 2016, one month prior to the initial announcement of a plan to tax SSBs by Finance Minister Pravin Gordhan, and June 30, 2019, when the article search was conducted. All articles were published in English, which is the primary language in South African education, journalism, broadcasting, and advertisements (31).
To be included in the search, articles must have included discussion of the South African HPL. Discussion of the HPL was defined as including at least one of the following topics: the potential effects of the HPL (either on health, economics, or SSB consumption), statements of support or opposition toward the HPL, explanations of the purpose of the HPL (to reduce consumption of unhealthy products or to improve health), or other statements that explain a purpose, goal, or likely outcome of the tax. Articles were excluded if they were duplicates of previous articles found in the search, if they were not relevant to the HPL (e.g. if they were about Value Added Taxes in South Africa or about general tax policy without any specific discussion of the HPL and its purpose or consequences), or if they were not published by a South African news source. Articles that discussed SSB consumption but did not discuss the HPL were excluded. Given our focus on news media, we also excluded reports from NGOs, law reviews and journals, and government documents. Articles could either be news articles written by journalists or opinion letters written to the newspaper and subsequently published, as publishing opinions was considered a view of the HPL presented by the newspaper. A flowchart for the article selection process is depicted in Figure 1.
Data extraction
Articles from the initial search were downloaded to the online software product Covidence (32). Two investigators (ME, FM) independently screened the article headline and first paragraph of the full text and excluded irrelevant articles using the inclusion and exclusion criteria. Disagreements were resolved after discussion of the inclusion/exclusion criteria between the two coders. Second, authors ME and FM independently screened the full text of relevant articles and eliminated those that did not fit the eligibility criteria. Articles were screened independently, with disagreements resolved after discussion between the two coders.
Final Study Sample
After the initial article search identified 571 articles, our final analytic sample included 193 articles published by the newspapers with the largest readerships in South Africa (Supplementary Table 1) (29).
Coding
To achieve inter-rater reliability, investigators (ME and FM) used articles written about the UK SSB tax to refine codebook definitions. This training set of articles from the UK context was used to decide which topics would be included in our study, and the codebook was updated in an iterative process until the two coders reached high agreement in understanding and using the codes. After the codebook training, the final set of definitions was established (Supplementary Table 2), and ME and FM coded a random subsample of 42 articles (22% of full sample) to ensure inter-rater reliability. We used Gwet’s AC1 as a measure for inter-rater agreement, as it has been showed to be a more stable measure of reliability than Cohen’s Kappa in instances of skewed distributions (in this case, many values of zero).(33,34) Items that achieved an acceptable Gwet’s coefficient were retained for the analysis, range = 0.77 – 0.97. After establishing acceptable inter-rater reliability and agreement, ME coded the remaining articles. Articles were coded by entering the data into a Microsoft Excel spreadsheet (35).
Outcomes: Topics mentioned
Topics analyzed included two major categories—health and economics, Health topics included statements that sugar consumption is (or is not) related to obesity, sugar consumption is (or is not) related to diabetes, sugar consumption is (or is not) related to NCDs, and the HPL will (or will not) improve health outcomes. Economics topics included statements that the HPL will (or will not) cause industry or economic harm, the HPL will (or will not) reduce health care costs, and the HPL will (or will not) economically harm the poor. We added two additional policy-relevant topics: changes in SSB consumption and SSB reformulation as a consequence of the HPL, as these are both important goals of the HPL. All topics used in our analysis are listed and defined in Supplementary Table 2.
Outcomes: Sources
We expanded upon the codebook used by Buckton and colleagues (26) by including the source attributed to each topic mentioned to be able to identify how different stakeholder perspectives were portrayed in the news media discussion of the HPL. We categorized sources as any person other than the journalist whom the journalist paraphrased or quoted as giving a statement about the HPL. Statements for which no source was given were attributed to the journalist. We categorized six source types as follows: industry, government, academics and medical experts, economists, non-governmental organizations (NGOs), or private citizens. Industry representatives included leaders of SSB companies, leaders of trade organizations with workers in either the SSB industry or sugar growing industry, or any other representatives of companies in the SSB production or sales supply chain. Government representatives were defined as members of the South African parliament or any other government job relevant to the HPL, such as the Minister of Finance or Minster of Health. Academics and medical experts (henceforth referred to as academics) included persons with an academic or public health research job at a college or university. Medical experts in this category also included medical doctors, nurses, or other health professionals. Economists were sources referred to as such in the article, or any employee of a research organization or other company conducting economics research. NGO representatives were defined as sources from a non-governmental organization such as the World Health Organization or other relevant NGOs. Private citizens were defined as South African citizens or members of the public who expressed a view about the HPL that did not belong in any of the other aforementioned categories. Sources were categorized based on how they were described in the article or by searching for biographical information about the source or author if no description was given. All sources used in our analysis are listed and defined in Supplementary Table 2.
Outcomes: Support or Opposition
Statements expressing support or opposition to the HPL were categorized in the same manner as topics described above, with a support or opposition statement and the source. We also coded articles for whether (1) individual mentions of support or opposition were present in the article and (2) if the article as a whole supported or opposed the HPL (article-level support or opposition). We recorded individual mentions of support or opposition because news media reports can often include a variety of perspectives, and we wanted to capture all unique views expressed in each article and by each source. However, in our analysis, we were also interested in the number of articles published over time that were primarily in support of or opposition to the HPL. We classified articles as being in support of the HPL if they contained a greater number of supporting mentions than opposing mentions. Articles with more opposing mentions compared to supporting mentions were classified as opposing articles. Articles with an equal number of supporting and opposing mentions were classified as balanced articles. Supporting mentions were defined as statements noting that obesity is related to SSB consumption; the HPL will improve health outcomes; the HPL will reduce SSB consumption; the HPL will reduce health care costs; the HPL will not harm industry; the HPL will benefit the health of the poor. Opposing mentions were defined as statements that obesity is not related to SSB consumption; the HPL will not improve health outcomes; the HPL will not reduce SSB consumption; the HPL will not reduce health care costs; the HPL will cause industry or economic harm; the HPL will economically harm the poor.
Outcomes: Proposed Solutions
Proposed solutions were recorded if (1) there was a health problem described that was linked to sugar consumption (e.g. obesity, diabetes, other chronic diseases) and if (2) any of our key sources proposed a solution to this health problem (e.g. government should tax SSBs, industry should voluntarily reformulate their products, individuals should exercise more).
Proposed solutions were classified into four levels at which interventions could operate: changes in individual action, changes in individual beliefs, changes in the nutritional composition of food (e.g. reformulation of products or the introduction or removal of products), or other food environment-related changes. Individual actions included diet changes, exercise changes, seeking education or counseling or other health information. Individual beliefs included interventions, initiatives, or structural responses proposed whose first steps are intended to change how individuals think about food, including public health campaigns, educational initiatives, marketing restrictions, and nutrition labels. Changes in nutritional composition of food included environmental or structural measures that would change the composition of foods available in the food supply, such as proposals to specifically incentivize reformulation or voluntary industry agreements for product reformulation. These proposed solutions for voluntary product reformulation differed from the topic code “reformulation as consequence” in that the voluntary solutions were proposed as a response to the health problem mentioned in the text, whereas reformulation as a consequence was mentioned as a direct consequence of the HPL. The fourth category included environmental or structural measures proposed that change the affordability, accessibility, or availability of foods, such as school food restrictions, taxes on unhealthy foods, subsidies of healthy foods, or restrictions on using financial assistance programs to purchase foods. The full definitions of our proposed solutions are provided in Supplementary Table 2.
We examined solutions by level of intervention, by source proposing the solution, and by the actor most responsible for carrying out the solution. For example, an academic source proposing SSB taxes as a means of reducing sugar consumption would be classified as an intervention related to changes in the food environment, suggested by academic source, and carried out by government regulation. Actors responsible for carrying out the proposed solutions included industry, government regulation, NGOs, and private citizens. This classification allowed us to separate an intervention into its component parts. Without this classification, government regulations would have all been categorized as the same.
Analysis
Supporting, opposing, and balanced articles published before and after HPL passed
First, to understand trends in article stance on the HPL over time, we conducted a descriptive analysis of the total number of supportive, opposing, or balanced articles published during our search timeframe, as well as before and after HPL implementation. A Pearson χ2 test was used to evaluate whether the proportion of pro, con, and balanced articles differed by whether articles were published before versus after the tax was passed. To contextualize trends over time, we identified the timing of publication relative to additional key events, including the announcement of the South African Government’s plan to tax SSBs (February 2016), the publication of a key research article showing the two year impact of an SSB tax in Mexico (February 2017), and the South African government passing the bill to tax SSBs (December 2017).
Topic mentions before and after HPL passed
Next, we analyzed the relative frequencies of topic mentions according to whether they were published before or after the HPL. Fisher’s exact tests were used to determine if the number of topic mentions differed based on being published before or after the HPL was passed, with a threshold for statistical significance set at 5% (α=0.05).
Topic mentions by source
Next, we examined whether the proportion of topic mentions and support/opposition for HPL differed by source. We stratified by the proportion of those topic mentions that came from our six sources in addition to journalist opinion (statement attributed to the journalist). A Pearson χ2 goodness-of-fit test was used to determine whether the distribution of sources within topic categories was different from the distribution of sources within all topics (i.e. the expected distribution assuming no relationship between topic and source), setting our significance level at 5% (α=0.05). We pooled all articles across time for this analysis because only 23% of our study sample was published after the HPL was implemented, which would leave many cell sizes small (less than 5), and therefore make statistical testing of differences between topics by source unreliable.
Supporting or opposing views of HPL by source
Because articles could contain more than one opinion of support or opposition from more than one source, we analyzed the total supportive or opposing views as the unit of analysis to characterize the perspectives presented from each of our six sources. Articles were analyzed according to total supporting and opposing opinions presented by each source. Fisher’s exact tests were used to determine if the number of supporting versus opposing views of the HPL within each source group were different, with a threshold for statistical significance set at 5% (α=0.05).
Proposed solutions: Actor responsible and level of action
To analyze the solutions proposed to solve health problems related to excess sugar intake, we first calculated the proportion of solutions that would be carried out by each actor as well as the sources that suggested that these actors should be responsible for carrying out the solutions. Next, we calculated the proportions of each level of action by the source suggesting the solution. Again, we used a χ2 goodness-of-fit test to determine whether these distributions differed significantly (α=0.05) from the expected distributions based on the total number of sources mentioned in the study sample. Post hoc Fisher’s exact probability z tests were used for pairwise comparisons between the percent contribution to solutions by source and the expected percent contribution based on overall prevalence of the source in the study sample.
All statistical analyses were performed in Stata 16 (36).