Article Text
PDF
Abstract
Background South Africa has since 1994 consistently increased the excise tax on cigarettes to maintain a total tax burden of 50% (1997–2003) and 52% (after 2004) of the average retail selling price. Between 1994 and 2004, the real (inflation-adjusted) excise tax increased by 249%, and the average real retail price of cigarettes increased by 110%. In addition, advertising and smoking bans were implemented in 2001. These measures, which we collectively refer to as tax-led, coincided with a 46% decrease in per capita consumption of cigarettes. No evaluation of South Africa's tobacco control policies has created a counterfactual of what would have happened if the tax-led measures had not occurred.
Objective (1) To create a credible counterfactual of what would have happened to per capita cigarette consumption if the tax-led measures had not happened. (2) To use this counterfactual to estimate their impact on cigarette consumption in South Africa.
Method We use a synthetic control method to create a synthetic South Africa, as a weighted average of countries (the ‘donor pool’) that are similar to South Africa, but that did not engage in large-scale tobacco control measures between 1990 and 2004.
Results Per capita cigarette consumption would not have continued declining in the absence of the tax-led measures that began in 1994. By 2004, per capita cigarette consumption was 36% lower than it would have been in the absence of the tax-led measures. These results, which we mostly attribute to tax increases, are robust to different specifications of the ‘donor pool’.
Conclusions Significant public health dividends can be obtained by consistently increasing the real tax on cigarettes.
- Economics
- Taxation
- Low/Middle income country
- Price
- Public policy
Statistics from Altmetric.com
Introduction
South Africa has since 1994 aggressively and consistently increased the excise tax on cigarettes so as to meet and maintain a total tax burden (including value added tax) of 50% of the average retail selling price. The target was met in 1997 and revised upwards to 52% in 2004. The real (inflation-adjusted) excise tax increased by 249% between 1994 and 2004, and by another 53% between 2004 and 2012. The tax increases have translated into substantial increases in the real retail selling price of cigarettes. For instance, the average real price per pack increased by 110% between 1994 and 2004 and by 190% if one extends the period to 2012 (see figure 1). In 2001, additional tobacco control measures (advertising bans and the prohibition of smoking in public places) also took effect.1 The increase in price and the introduction of other tobacco control measures coincided with substantial declines in smoking prevalence and cigarette consumption. Van Walbeek1 estimated that smoking prevalence declined from about 32% of the adult population in 1993 to about 24% in 2003, while aggregate and per capita cigarette consumption declined by 32% and 46%, respectively, over the same period.
Real excise tax per pack of cigarettes and real price per pack of cigarettes, South Africa 1960–2012.
An implicit assumption in the literature that evaluates South Africa's tobacco control measures is that preintervention levels in, say, prevalence and consumption would have continued in the absence of the intervention (Koch and Tshiswaka-Kashalala2 make a similar point). This assumption was also implied in Warner's evaluation of the antismoking campaigns in the USA in the 1970s.3 Given this, the effect of the intervention is typically assessed by comparing present-day consumption or prevalence with the magnitudes of these variables before the onset of the intervention. In the absence of a credible counterfactual (a ‘what-if’) scenario, and in an environment where cigarette consumption and prevalence were already decreasing before the onset of any tobacco control measures, for other reasons,1 ,4 their effect on cigarette consumption and smoking prevalence may be overstated.
This paper uses a transparent data-driven technique, the synthetic control method, developed and extended by Abadie and Gardeazabal5 and Abadie et al,6 to create a counterfactual of cigarette consumption in South Africa from 1994 to 2004. Using this counterfactual to create a synthetic South Africa, we estimate a ‘treatment effect’ of South Africa's tax-led tobacco control measures on cigarette consumption. We refer to them as ‘tax-led’ because: (1) increases in the excise tax were the first major tobacco control initiative in South Africa and (2) other than the introduction of small, non-graphic warning labels on cigarette packs and advertising material in 1995, tax increases formed the mainstay of South Africa's tobacco control measures between 1994 and 2004. A comprehensive advertising ban and the prohibition of smoking in public places were only implemented in 2001.
Prior to 1994, South Africa did not consciously aim to reduce the consumption of tobacco products on public health grounds. The relegation of public health concerns was likely due to the cordial relations that existed between the tobacco industry and the National Party that ruled South Africa from 1948 to 1994.1 For instance, the real excise tax on cigarettes, the main tobacco product in South Africa, declined by 70% and the real price per pack declined by 37% between 1961 and 1990.1 Annual per capita cigarette consumption increased from about 1000 cigarettes in 1961 to 1600 cigarettes in 1991.1
In the 1980s and early 1990s, the medical research community and the South African Medical Research Council published research showing that tobacco consumption imposed a net cost on the country.7–9 The publicity generated by these studies rallied the public health community and health-focused civil society organisations behind the common goal of getting the South African government to take tobacco control seriously. The momentum that had built up during the 1980s and early 1990s, along with the impending change of government, culminated in the passing of the Tobacco Products Control Act of 1993, which introduced warning labels for the first time.
The big turning point, however, came in 1994 when the democratically elected African National Congress-led government announced that the government would target a tax burden on cigarettes (including value added tax) of 50% of the retail price, to be phased in over a number of years.10 At the time of this announcement, the total tax burden (including value added tax) was 33% of the retail price. As a result of increasing the tax burden, the years 1994, 1995 and 1996 saw excise tax increments of 25%, 25% and 18%, respectively.10–12 In 1997, the Minister of Finance announced a 52% increase in the excise tax on cigarettes, a move that was expected to bring the total tax burden to 50% of the average retail selling price.13 After 1997, the annual increases on excise taxes on cigarettes have been aimed to maintain the 50% total tax burden, resulting in a highly predictable policy environment for all stakeholders, including the tobacco industry. In 2004, the total tax burden target was revised to 52% of the average retail selling price.14
South Africa's aggressive excise tax policy since 1994 has translated into substantial increases in the real price of cigarettes. From 1994 to 2012, the average real price per pack of cigarettes increased by 190% (see figure 1). Between 1994 and 2004, which is the period we evaluate in this paper, the real price per pack increased by 110%. In contrast, the period before 1994 saw considerable declines in the real price of cigarettes. It is this unprecedented increase in real cigarette excise taxes and prices (alongside other tobacco control measures), beginning in 1994, whose impact on consumption we seek to evaluate in this paper.
Method
Synthetic control method
This paper uses the synthetic control method to evaluate South Africa's tobacco control policies from 1994 to 2004. The technical aspects of the method are discussed in detail in Abadie et al 6 and in a working paper version of the current paper.15 We discuss the intuitive aspects of the method here.
The method involves estimating South Africa's counterfactual per capita cigarette consumption in the absence of the tax-led tobacco control measures that began in 1994. In other words, the method tries to recreate the trend in per capita cigarette consumption in South Africa if the tax-led measures had not happened. Synthetic South Africa, the counterfactual, is constructed as a weighted average of countries that are similar to South Africa but did not enact significant tobacco control policies over the period 1994–2004. The set of countries from which synthetic South Africa is constructed is collectively known as the ‘donor pool’ and the period after 1994 is known as the ‘treatment period’. The weights on each country in the donor pool are derived by comparing how closely each country matches pretreatment South Africa on the outcome variable (per capita cigarette consumption) and on those variables that influence consumption. Technically, the weights are calculated so as to minimise the pretreatment difference between actual South Africa and synthetic South Africa on the following variables: per capita cigarette consumption, real price of cigarettes, real per capita Gross Domestic Product (GDP), per capita alcohol consumption (in pure litres of alcohol) and the structure of the population (proportion of adults). The latter four variables are the standard predictors found in empirical specifications of the demand for tobacco products.16
The relationship between cigarette consumption on the one hand and prices and income on the other is well established in the literature. Recent work has also shown a positive and statistically significant association between cigarette consumption and alcohol consumption.17–19 The population structure variable captures the fact that cigarettes are mostly sold to adults. That is, the greater the proportion of adults in the population, the more cigarettes you expect to be bought, holding all other things constant.
The weights are obtained numerically using a Stata 12 user-written program called Synth, available at Jens Hainmueller's website.20
Once synthetic South Africa has been constructed, the observed per capita cigarette consumption for the ‘real’ South Africa is then compared to cigarette consumption under the counterfactual scenario to determine the effect of the tax-led measures in influencing consumption. By construction, there should be no difference in per capita consumption between the ‘real’ South Africa and its synthetic counterpart before treatment.
Our choice of conducting the evaluation over the 1994–2004 period is due to the WHO's Framework Convention on Tobacco Control (FCTC) coming into effect in 2005. The Treaty encourages countries to implement a wide array of tobacco control measures. Most of the countries in our donor pool began, from 2005 onwards, to implement a range of tobacco control measures, as part of their obligations under the FCTC. To the extent that these interventions were successful in reducing tobacco consumption, this would result in a downward bias in our treatment effect estimates. Further, Abadie et al 6 ,21 consider a 10-year period to be a sufficient timespan to properly evaluate the effects of a policy change.
Selection of the donor pool
The countries that are used to construct South Africa's counterfactual cigarette consumption are collectively referred to as the ‘donor pool’. First, the donor pool should consist of countries that did not engage in significant tobacco control initiatives over the evaluation period. We refer to these as ‘untreated’ countries. Second, the donor pool should consist of countries that are similar to South Africa in many respects. For instance, the donor pool should consist of countries at a similar level of economic development as South Africa.
We rely on the cigarette affordability literature to select a donor pool consisting of untreated countries. Two of the present authors propose a measure of cigarette affordability which compares the price of cigarettes to real per capita GDP.22 ,23 If a country experiences a decreasing ratio (ie, cigarette prices are decreasing relative to per capita GDP), this means that cigarettes are becoming more affordable, while an increasing ratio signifies declining affordability. Using this affordability concept, they classified 77 countries according to whether they experienced increasing affordability or decreasing affordability over the period 1990–2006.23 For these countries, they had complete and comparable data on real cigarette prices and real per capita GDP over the period 1990–2006. Of these 77 countries, cigarettes became more affordable in 37 countries over the period 1990–2006 (see Figure 3 in their paper). In 20 of the 37 countries, the real price of cigarettes decreased. In the remaining 17 countries, real per capita GDP grew faster than the increase in real prices, which resulted in cigarettes becoming more affordable.
We opt to use the increase in affordability over the 1990–2006 period as a proxy for the absence of treatment. That is, we regard countries where cigarettes became more affordable over this period as not having enacted significant tobacco control measures. This is obviously the case for the 20 countries where affordability increased as a result of declining real cigarette prices. We contend, however, that even for the remaining 17 countries where affordability increased due to real incomes growing faster than real prices, a conclusion of the absence of treatment is a reasonable one to make. This is because effective tobacco control measures require (1) real tax and price increases and (2) real tax and price increases that grow faster than the rate of growth in incomes.24 ,25
We recognise that the affordability concept might have some shortcomings in identifying whether a country has instituted tobacco control measures or not. For instance, a country may have adopted a wide set of tobacco control measures such as advertising bans and/or clean indoor air policies but neglected to significantly increase real cigarette prices. Our measure of treatment would consign this country to the pool of potential donor countries in spite of its tobacco control efforts. In as much as we recognise that tobacco control measures constitute more than just tax/price measures, the tobacco control literature recognises the primacy of tax/price policies in curbing demand.16 ,25 In any case, the estimates of the treatment effect would be lower bounds if the donor pool included some countries whose treatment status was misclassified in the manner suggested here.
Our criterion for identifying treatment correctly classifies many of the countries that are known for having instituted significant tobacco control measures over the period 1990–2006. For example, South Africa, the country of interest in this paper, is correctly classified as treated since cigarettes became less affordable over the period 1990–2006. Thailand, a country whose positive experience with tobacco control is often held up as a model for other developing and emerging countries,26 is also classified as having undergone treatment. Most high-income countries (especially in North America and Europe), whose tobacco control efforts predate the 1990s, are also correctly classified as treated. On the other hand, untreated countries are mainly low-income and middle-income countries, which is an expected outcome given these countries' slow progress in implementing effective tobacco control measures over the period 1990–2004.27 The full list of treated and untreated countries from Blecher and van Walbeek's 2009 paper are contained in online supplementary table A1.
Supplementary table
The final requirement is that South Africa and the countries in the donor pool are similar (this is technically known as the Convex Hull requirement).6 A transparent way of ensuring this is to use the World Bank's Country Classification System which is based on per capita income. We exclude all countries that were classified as high-income countries at the end of 2004.28 These countries are often perceived as being fundamentally different to low-income and middle-income countries. Finally, we drop from the potential donor pool countries without a complete set of data for all variables over the period 1990–2004 (the data and sources are described below). The final donor pool consists of 24 countries that are listed in the first column of table 1.
Donor pool and synthetic weights
It is worth noting that, even though we refer to the measures as tax-led, we attribute the treatment effect estimates presented in the Results section primarily to excise tax increases between 1994 and 2004. Several factors motivate our reasoning. First, the tobacco control literature has shown that tax and price measures are the single most effective tobacco control tools.16 ,25 Second, excise tax increases were, in the main, the only tobacco control measure between 1994 and 2004.1 Third, the advertising ban and the prohibition of smoking in public places were only implemented from 2001, 7 years after the policy of consistently increasing excise taxes had begun.1 We, however, recognise that the other tobacco control measures introduced from 2001 have also been shown to have had an effect on cigarette consumption in South Africa.29 We, therefore, acknowledge that our treatment effect estimates may overestimate the pure effect of tax increases on consumption in South Africa. This is especially the case for treatment effects estimates from 2001 to 2004.
Data
The weights on the respective countries in the donor pool are chosen on the basis of how closely each country matches pretreatment South Africa on the following variables: per capita cigarette consumption (the outcome variable), real price of cigarettes, per capita real GDP, per capita alcohol consumption (in litres of pure alcohol) and the structure of the population (proportion of adults in the total population). Data on the outcome variable, per capita cigarette consumption (in sticks), are taken from the World Cigarette Report published by the ERC Group.30 Cigarette price data are from the Economist Intelligence Unit's (EIU's) Worldwide Cost of Living Survey.31 Per capita GDP and data on the proportion of adults (16–64 years) in the population are taken from the World Bank's World Development Indicators database.32 Finally, data on per capita alcohol consumption (in litres of pure alcohol) are obtained from the WHO's Global Information System on Alcohol and Health.33
Results
Table 1 presents the combination of donor countries and their respective weights that are used to create synthetic South Africa. According to table 1, synthetic South Africa is a weighted average of Brazil, Argentina, Chile, Tunisia and Romania (in decreasing order of weighting). The combination of these countries, and their respective weights, nearly perfectly recreates South Africa's pretreatment characteristics, including per capita cigarette consumption (the outcome variable). Countries with zero weights do not share much similarity to South Africa's pretreatment characteristics.
The success of the synthetic control method in recreating South Africa's pretreatment consumption trend line can be seen in figure 2, which plots per capita cigarette consumption for South Africa and synthetic South Africa over the period 1990–2004. The difference in per capita cigarette consumption between South Africa and its synthetic counterpart before 1994 is negligible. This is by construction since the two countries are supposed to be identical in every respect before 1994. After 1994, the two countries are only distinguished by the fact that actual South Africa is treated (by the tax-led measures) and synthetic South Africa is not.
Cigarette consumption per capita, South Africa versus synthetic South Africa.
After the onset of treatment in 1994, the two lines in figure 2 diverge, with South Africa's consumption line being everywhere lower than synthetic South Africa's consumption line. South Africa's per capita cigarette consumption declines throughout the treatment period, whereas synthetic South Africa's trend line initially rises, then falls and eventually stabilises at around 800 cigarettes per capita from the year 2000. The difference between the two lines at each point in time is a measure of the cumulative effect of the treatment.
Table 2 presents actual estimates of the treatment effect calculated as the difference between South Africa and its synthetic counterpart. Between 1990 and 1993, the treatment effect is ∼0. By 1995, the first year after treatment begins, South Africa's per capita cigarette consumption is 38 cigarettes (about 4%) less than its synthetic counterpart. The treatment effect increases with each additional year so that by 2004, South Africa's per capita cigarette consumption is about 290 cigarettes, or 36%, less than its synthetic counterpart.
Actual estimates of treatment effects
One of the factors that might explain why per capita cigarette consumption stopped declining for synthetic South Africa after 1994 is the performance of the economy. Between 1980 and 1994, South Africa's real per capita GDP declined at the average rate of 1% per year.34 Since tobacco consumption in South Africa is positively related to income,1 ,29 ,35 ,36 the decrease in per capita GDP largely explains the decrease in cigarette consumption in the pretreatment period. On the other hand, between 1994 and 2004, per capita real GDP increased at the average rate of 2% per year.33 Therefore, the decline in consumption that was already underway by 1994 would likely have stopped, in the absence of tax-led measures, simply because incomes began to rise.
Robustness tests
In this section, we check whether the treatment effects are sensitive to the composition of the donor pool. We do this by first excluding, one at a time from the donor pool, the countries in table 1 that have positive donor weights, and then re-estimating the treatment effect. This is carried out to guard against the possibility that our estimated effects are being driven by a single donor country. A second robustness test uses a different criterion for identifying treatment status, namely the average annual percentage change in the real price of cigarettes over the period 1990–2004. This is carried out to guard against the possibility, highlighted earlier, that using the affordability concept might misclassify treatment particularly in cases where changes in affordability are driven by changes in income and not changes in price.
Figures 3 ⇓ ⇓ ⇓–7 present the results of successively excluding from the donor pool countries which earlier had positive weights. The pattern of the trajectories for synthetic South Africa is similar across the five figures and, more importantly, similar to the pattern in figure 2.
Robustness test excluding Argentina.
Robustness test excluding Brazil.
Robustness test excluding Tunisia.
Robustness test excluding Chile.
Robustness test excluding Romania.
Figure 8 presents the results of using the average annual percentage change in real cigarette prices as a criterion for identifying treatment status. A negative percentage change implies a lack of treatment and such countries then form our donor pool. In figure 8, the trajectory of synthetic South Africa after the beginning of treatment in 1994 is similar to that of synthetic South Africa in figure 2.
Robustness test using change in price to identify treatment status.
Discussion
South Africa was one of the first middle-income countries to use rapid increases in the excise tax to reduce cigarette consumption and thus significantly improve public health. In the late 1990s and early 2000s, South Africa was perceived as a model for other low-income and middle-income countries in tobacco control. Between 1994 and 2004, the real price of cigarettes increased by 110% in response to a 249% increase in the real excise tax. Other tobacco control measures, such as advertising bans and the prohibition of smoking in public places, were implemented from 2001. Aggregate consumption of cigarettes decreased by about 32% and per capita consumption by about 46% over this period.
This large decrease in cigarette consumption is not surprising. A very large literature has investigated the impact of changes in cigarette prices and cigarette consumption and concludes that, in most countries, a 10% increase in the real price of cigarettes decreases consumption by between 4% and 8%.25 This paper adds to this literature by confirming the results of the price elasticity literature, but it uses a novel approach, namely the synthetic control method. In essence, the method creates a counterfactual to what has played out in reality. The synthetic control method compares South Africa's cigarette consumption over time, with that of synthetic South Africa, where the latter consists of a weighted group of countries that are similar to South Africa, but differ in one crucial aspect: they did not follow South Africa's aggressive tobacco control policy over the study period.
We found that per capita cigarette consumption in South Africa was about 36% less in 2004 than it would have been had South Africa not followed this strategy. The public health impact of such a large decrease in cigarette consumption is substantial and, once again, illustrates the effectiveness of excise tax increases in reducing cigarette use.
The power of the synthetic control method lies in the fact that it can be applied in other contexts as well. For example, it could be used to evaluate the effectiveness of standardised (plain) packaging in Australia to reduce cigarette consumption in that country. We trust that other researchers will apply this technique to evaluate tobacco control interventions in their countries.
What this paper adds
The paper contributes to the literature on the rigorous evaluation of tobacco control interventions through the construction of a counterfactual (a what-if) scenario and uses South Africa as a case study.
Through mostly large increases in the excise tax, per capita cigarette consumption in South Africa was 36% lower in 2004 than in the absence of that intervention, again highlighting the important role that excise tax increases can play in reducing tobacco use.
Acknowledgments
The authors would like to acknowledge comments and suggestions made by attendees to the 2015 Biennial Conference of the Economics Society of South Africa (ESSA) held at the University of Cape Town. The authors also acknowledge comments received from anonymous referees of this journal and of the Economic Research Southern Africa (ERSA) Working Paper Series. The authors alone are responsible for the views expressed in this publication and they do not necessarily represent the decisions or the policies of the World Health Organisation.
References
Footnotes
Twitter Follow Grieve Chelwa at @gchelwa
Contributors GC conceptualised the project, did the analysis and wrote the paper along with CvW. EB supplied the data and contributed to the writing of the paper.
Funding This work was supported by funding from the Bill & Melinda Gates Foundation and from Economic Research Southern Africa (ERSA).
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.