Hoping no doubt to capitalise on New Year temperance sentiments, David Cameron chose last week to announce a review of alcohol pricing. Westminster officials have been ordered to develop ‘a scheme in England to stop the sale of alcohol at below 40p to 50p a unit in shops and supermarkets’, akin to similar but further advanced Scottish plans. The idea behind these neo-Prohibitionist proposals is the so-called ‘single distribution theory’, expressed rather alarmingly some time ago in the title of a seminal BMJ paper as ‘The population mean predicts the number of deviant individuals’. The notion is that, not only can the tail wag the dog, so too can the dog wag the tail. Reduce overall alcohol consumption, and the number of deviant (i.e. heavy) drinkers will also reduce. Like most neo-Prohibitionist clap-trap, it is pure poppy-cock.
Almost exactly two years ago – attacks on boozers do indeed seem often to appear at this time of year – Dr No did a demolition job on Rose and Day’s deviant individuals paper, and its fore-runner, Rose’s ‘Sick Individuals and Sick Populations’. Although the single distribution theory was not new, these two papers became the seminal works on the dog’s body wags the tail idea. Those who care to can read Dr No’s earlier post, but in summary the fatal flaw in Rose and Day’s analysis is that it relies on cross-sectional data – that is, snapshot in time data – and, as any epidemiologist worth his salt, not to mention malt, will tell you, association does not prove causation. A study across countries may well show mean alcohol consumption is positively associated with the prevalence of heavy drinking – as indeed it is – but that does not prove that a high mean consumption of alcohol causes a higher prevalence of heavy drinking.
Rose is clever with the wording in his papers, which no doubt helped his ideas to gain currency, by stating that the population means predict, and are related to, numbers of sick/deviant individuals. In a strict mathematical sense, these statements are true: given a mean alcohol consumption, one can predict the number of heavy drinkers, because of the mathematical relationship, demonstrated by the snapshot data; but Rose’s error is subtly to extend this correct cross-sectional observation to a wider more general assertion that interventions to reduce population means will necessarily reduce the number of sick/deviant individuals.
This extension, unfortunately, is not only logically unjustified; it also lacks substantial empirical support. Despite the passage of two or more decades from the original papers, and notwithstanding some gambling research – for example, this rather cute Norwegian one armed bandit study here, and a similar British National Lottery study here – which do nonetheless suggest a possible causal relationship between mean population gambling spend and problem gambling, Dr No is not aware of a single study that convincingly demonstrates that an intervention to reduce mean population alcohol consumption will necessarily reduce the prevalence of heavy drinking.
Instead, both the Scottish and Westminster governments rely heavily on statistical modelling work – hypothetical Excel-style ‘what-if’ calculations, done, notably, by the University of Sheffield. Unsurprisingly, the University’s reports are rather upbeat about the beneficial effects of minimum pricing, but what we have to remember here is that not only is this upbeat message based on hypothetical calculations, it is also totally dependent on the assumptions on which those calculations are based. This kind of ‘research’ – the quotations marks justified by the fact it isn’t real research, just a series of ‘what if’ calculations – is unavoidably susceptible to what the ‘researchers’, who are after all only human, just happen to use as assumptions that – well – make their ‘results’ more upbeat.
Now, if the single distribution theory is correct, in broad brush terms the price elasticity of alcohol needs to be – so the tails move by the same amount as the body – reasonably similar across all drinkers. The Sheffield researchers do indeed appear to make an assumption that price elasticity is relatively similar across moderate, hazardous and harmful drinkers for general price increases (an elasticity of around –0.4% : that is, a 10% increase in cost will reduce consumption by around 4% – see here; and while there, we may note the superfluous addition of the 25% bars – they are simply the 10% bars multiplied by 2.5). However, when it comes to minimum pricing (the right-hand-most three groups of bars), they appear to assume that heavier drinkers have higher elasticities – that is, they change their consumption more in response to changes in prices than moderate drinkers.
Having known a few boozers in his time, Dr No finds the suggestion that heavier drinkers have higher price elasticities than moderate drinkers counter-intuitive. Boozers like their booze, and one might suppose they are less likely to reduce consumption, even when prices are hiked. Yet the Sheffield reports suggest otherwise. So Dr No decided to look in more detail at the Sheffield reports – and what he found does suggest, ahem, that the Sheffield researchers are indeed ‘all too human’.
The key paragraphs are on page 51 of the Sheffield modelling report. In summary, they say we got the opposite results to everyone else – but are going to use our contrary data anyway. And when their figures do agree with those reported by others, they simply say “these high-level estimates are provided for reference only and are not included in the model”. As someone once said: why let data get in the way of a good conclusion?
There are other flaws in the Sheffield data, with perhaps the most significant being that they rely on the EFS (Expenditure and Food Survey), which not only has a worryingly low response rate (55% – ie almost half of households did not provide data); even worse, there is evidence that, notwithstanding applied corrective weighting, that the results are biased towards the more moderate, ‘responsible’ households – in other words those that exclude the very households – those perhaps with more heavy drinkers – that we are most interested in.
So the Sheffield model is very likely based on wrong assumptions. If it is indeed true that heavy drinkers are relatively price inelastic, then Rose’s population mean predicts the number of deviant individuals assertion starts to look decidedly shaky. If we use more widely accepted elasticities, for example from this mega meta-study (and interestingly the dumped Sheffield high level estimates are very similar), then we find an elasticity of –0.44 for ‘total alcohol’ (ie all drinkers), falling to –0.28 for heavy drinkers (the related dumped Sheffield figures being moderate drinkers –0.47, harmful and hazardous drinkers combined –0.21).
What this means, in broad brush terms, is that for moderate (and so ‘average’, or ‘mean’) drinkers, we might expect them to reduce consumption by around 4.5% for every 10% increase in price, while hardened boozers will only reduce consumption by some 2.5% given the same price increase. The body, clearly, is not wagging the tail as strongly as Rose would like us to believe.
Nor, if we convert those percentages into actual amounts drunk (and not drunk) do the results impress. The heavy drinker consuming say 50 units a week is likely to respond to a 10 percent increase in price by a 2.5% reduction in consumption – a paltry 1.25 units – from 50 units to 48.75 units…hardly, Dr No suggests, an impressive result.
As a doctor interested in epidemiology, Dr No has naturally focused on the science for and against the effectiveness of minimum pricing as a means to reduce heavy drinking: and he is not in the slightest bit convinced that a case has been made. Others have pointed out the libertarian objections, and then there is the regressive nature of any such policy: the poor will be disproportionately hit. And – if Dr No is correct, and minimum pricing will have precious little effect on heavy drinkers – then there may turn out to be serious adverse consequences. Heavy drinkers will continue to drink heavily, but will spend more, and amongst poorer heavier drinkers, that can only mean less to spend on other things: less to spend on food, on their children, and their children’s food.
And then – to add insult to injury – the extra cash raised (note that the negative elasticities mean that sales will decline by less than the increase in price, such that overall turnover will increase) by minimum pricing will, unlike tax revenues, go mostly into the pockets of the supermarkets and drinks manufacturers. Small wonder, then, that some supermarkets and drinks manufacturers are putting their weight behind minimum pricing.
The primary modelling reports on which both the Scottish and Westminster governments rely on appear deeply flawed, and the likely unintended consequences of minimum pricing for alcohol are all too predictable. It seems to Dr No that instead of good government tending to good people, we have sick politicians tending to sick notions.