Overall and income specific effect on prevalence of overweight and obesity of 20% sugar sweetened drink tax in UK: econometric and comparative risk assessment modelling study
We explain a paper that models the potential effects of a sugar sweetened drink tax
“Overall and income specific effect on prevalence of overweight and obesity of 20% sugar sweetened drink tax in UK: econometric and comparative risk assessment modelling study” by A Briggs and colleagues (BMJ 2013;347:f6189).
Objective—To model the overall and income specific effect of a 20% tax on sugar sweetened drinks on the prevalence of overweight and obesity in the UK.
Design—Econometric and comparative risk assessment modelling study.
Population—Adults aged 16 and over.
Intervention—A 20% tax on sugar sweetened drinks.
Main outcome measures—The primary outcomes were the overall and income specific changes in the number and percentage of overweight (body mass index ≥25) and obese (≥30) adults in the UK after the implementation of the tax. Secondary outcomes were the effect by age group (16-29, 30-49, and ≥50 years) and by UK constituent country. The revenue generated from the tax and the income specific changes in weekly expenditure on drinks were also estimated.
Results—A 20% tax on sugar sweetened drinks was estimated to reduce the number of obese adults in the UK by 1.3% (95% credible interval 0.8% to 1.7%) or 180 000 (110 000 to 247 000) people and the number who are overweight by 0.9% (0.6% to 1.1%) or 285 000 (201 000 to 364 000) people. The predicted reductions in prevalence of obesity for income thirds 1 (lowest income), 2, and 3 (highest income) were 1.3% (0.3% to 2.0%), 0.9% (0.1% to 1.6%), and 2.1% (1.3% to 2.9%). The effect on obesity declined with age. Predicted annual revenue was £276m (£272m to £279m), with estimated increases in total expenditure on drinks for income thirds 1, 2, and 3 of 2.1% (1.4% to 3.0%), 1.7% (1.2% to 2.2%), and 0.8% (0.4% to 1.2%).
Conclusions—A 20% tax on sugar sweetened drinks would lead to a reduction in the prevalence of obesity in the UK of 1.3% (around 180 000 people). The greatest effects may occur in young people, with no significant differences between income groups. Both effects warrant further exploration. Taxation of sugar sweetened drinks is a promising population measure to target population obesity, particularly among younger adults.
Why do the study?
Obesity is one of the biggest threats to public health in developed countries, and a leading cause is overconsumption of high energy foods including drinks sweetened with sugar. These drinks, including colas and other fizzy drinks, have no nutritional value and simply add calories to consumers’ diets. Governments are currently exploring how to reduce their consumption, and a tax at point of sale is one possibility. Taxes on alcohol and cigarettes reduce consumption of these harmful substances, so it seems an obvious step to consider a tax on sugar sweetened drinks. The authors of this study are trying to find out if a substantial tax would have any effect on the prevalence of obesity in the UK, a sensible move before changing government policy. They particularly want to find out if a tax would have any unwelcome side effects, such as hitting the poorest hardest.
What did the authors do?
The authors used well established modelling techniques to help predict what is likely to happen if the UK government introduced a 20% sales tax on drinks sweetened with sugar. They estimate the likely impact of the tax on consumption of sugar sweetened drinks, total energy intake, and finally body mass index across the whole UK population.
Mathematical models are an increasingly popular way to study large scale interventions aimed at improving population health, which are often fiscal policies such as taxation. They are well known to economists and meteorologists, but perhaps less familiar to doctors. The key thing to remember about modelling studies is that they are an artificial construct designed to predict the future, which is always a perilous undertaking. Researchers put data in, and get results out while readers rarely get to understand fully how the two are connected. A good modelling paper (such as this one) will include enough detail to give non expert readers a fighting chance. We don’t need to know exactly how the model works, so long as we can judge the quality of the data going in and check that all the assumptions made by the authors are valid.
These authors use two different models—a Markov Chain Monte Carlo (MCMC) simulation and a model called PRIME. Combining the two, they were able to estimate: firstly, how a price hike would change demand for different drink types (measuring their elasticities); how such a price hike would change consumption; how the change in consumption would change energy intake; and finally how the change in energy intake would affect body mass index.
Figure 1 describes a fairly straightforward chain of events, but there are built in assumptions at every stage. The authors helpfully give details of the main assumptions in the figure, so we can judge for ourselves whether or not they are reasonable. They plug in different data sources at each stage too, and use three different national surveys: the Living Costs and Food survey (2010), the National Diet and Nutrition Survey (2008-10), and National Health Surveys for England and Scotland (2010). These data are the backbone of the study: if you don’t trust the sources, then you shouldn’t trust the findings. No amount of statistical sophistication can make up for bad data. A good modelling paper should tell you precisely where all data came from so interested readers can do it all again and check the findings. 1
The MCMC simulation was used to estimate the elasticities (first stage) and the PRIME model was used to link changes in energy intake with changes in body mass index (last stage). The authors generated results from their modelling and put credibility intervals around the headline figures to reflect the uncertainties in the model. Credibility intervals work a bit like confidence intervals around results in randomised trials. If you ran the models 100 times, 95 of the results would be within the quoted credibility interval. You can check that the bottom and the top numbers still represent a worthwhile effect. Very wide credibility intervals indicate uncertainty around the results, and readers should know to interpret with caution. These authors also tested the consistency of their figures by running the model twice—once for a 20% tax on sugar sweetened drink, then again for a 10% tax. As expected, the effect of a 10% tax on overweight and obesity was roughly half the effect of a 20% tax.
Modelling studies can be a useful tool to help guide policy. They give us a hint about the possible impact of an intervention on populations. But they all come with a health warning. Like weather reports, modelling studies can be miles out. Partly because populations are complex, and authors rarely have the detailed data they need to account for all eventualities. They must do their best with what they have, using more or less robust assumptions. No one can predict the future, and the only way to find out precisely what will happen if the UK government taxes sugary drinks is to do it in a controlled way and measure what happens very carefully in intervention and control jurisdictions. Modelling gives policy makers some idea about whether to go ahead, and what to measure if they do. The editorial linked to this paper (p 8) urges policy makers to introduce the tax, and gives us some “real world” data to go on. Gazing into crystal balls can take you only so far.
What did the study find?
These authors estimate that a 20% tax on sugar sweetened drinks would cut the prevalence of obesity among UK adults by 1.3% (95% credible interval 0.8% to 1.7%), or 180 000 people (110 000 to 247 000), and would cut the prevalence of overweight by 0.9% (0.6% to 1.1%) or 285 000 people (201 000 to 364 000). The credibility intervals are reasonably narrow. The lowest numbers are small but still suggest an effect worth having. They found no consistent pattern between income groups and so conclude that the tax wouldn’t be regressive and hit the poorest third of the population harder than the richest third. Finally they estimate that a 20% tax at point of sale would generate £276m (€333m; $450m) a year for the treasury. The authors think a taxation policy looks promising, but it is not a panacea for obesity in the UK.
Strengths and weaknesses
The main strength of this modelling study is its clarity. Non-experts can read it, understand what was done, and begin to judge the robustness of the findings. Transparency is an important aspect of any modelling study. The more you know about the inputs and assumptions, the more you can judge the outputs, even if the model itself remains something of a black box.
As always, the main weaknesses lie in the data. These authors used three surveys to derive their estimates, and had to extrapolate from samples to the whole population of the UK. For example, they had to use the Health Survey for England to estimate the baseline prevalence of obesity and overweight in Wales and Northern Ireland, where prevalence may be different.
Price elasticities aren’t easy to calculate with certainty and tend to vary depending on the methods used. The calculation of price elasticities are the first stage in their model (linking price changes with changes in demand), so weakness here can have knock on effects further down the chain. The authors had to assume that elasticities were the same for people of all age groups, which may not be the case. Young adults drink more sugary drinks than older adults, so may be less sensitive to changes in price. Perhaps the tax would work less well for the highest consumers.
The authors also mention a mismatch between what people said they were consuming in national surveys and sales data for sugar sweetened drinks. The industry sells three to four times more than people claim to be drinking. If sales data are more accurate (which seems likely, since people are notoriously deluded about their unhealthy diets), then a tax would have a bigger impact on energy intake and obesity than reported here.
Importantly, the authors explore these limitations and several others in a self critical discussion section—another sign of a modelling study that you can trust.
What does the study mean?
The findings suggest that a substantial tax on sugar sweetened drinks could have a modest but measurable impact on the prevalence of obesity and overweight in the UK. Additionally, such a tax might at the same time generate a similarly modest income for the treasury, which might be deployed to fund other measures to improve population health.
The main message for students is that modelling studies are here to stay. Don’t be afraid to tackle them, and look for a clear, detailed, transparent, and self critical account of the methods and findings.Alison Tonks, associate editor, BMJ
Competing interests: I have read and understood the BMJ Group policy on declaration of interests and declare: I am a research editor at the BMJ and help to select papers for publication.
Provenance and peer review: Commissioned; not externally peer reviewed.
- Block J P, A substantial tax on sugar sweetened drinks could help reduce obesity. BMJ 2013;347:f5947.
Cite this as: Student BMJ 2013;21:f7256