Thursday 27 November 2014

A belated reply to the minimum pricing modellers (part one)

Quite some ago I wrote a short paper with John Duffy about the minimum pricing predictions coming out of Sheffield University. It was published by the Adam Smith Institute. Some of the Sheffield researchers later wrote a response to our points which, for one reason or another, I never got round to replying to.

It's a little late to pen a response now, but as John Holmes keeps saying that our critique has been 'rebutted', I'd like it to be known that it rebuts nothing. In this blog post and the follow-up post, I will address the points they raise. Although some of the views will, I hope, be shared by John Duffy, I should stress that I am speaking only for myself.

Price, consumption and harm in practice

The Sheffield response begins by stating that "it is consistently the case that when prices go up, consumption goes down" and that "it is consistently the case that when prices go up, overall levels of harm go down." The former statement is basically just the law of demand, whilst the latter is the law of demand plus the temperance assumption that harm is tied to per capita consumption (AKA the Total Consumption Model).

The former is a quite reasonable assumption upon which to base an economic model, but it is not an iron law. The second is much more debatable. Taking both assumptions as gospel and using them to make absurdly specific claims about minimum pricing reducing alcohol-related deaths by exactly 521 a year is just silly.

From the outset, my reason for wanting to write about the Sheffield model was my exasperation at the way in which its estimates were treated as settled science when they were really just elaborate guesses conjured up by people who support the policy. It is not that the above assumptions are necessarily wrong, just that there was a reasonable chance that they were wrong and there was a good chance that at least one of them was wrong. (And, of course, even if they are both right that doesn't mean their estimates are right, as we shall see.)

The hubris of the Sheffield model is exhibited in their response to us. Contrary to their bald assertions, it is not "consistently the case" that x follows y in the way the Sheffield team claim. You only need to look at the UK in the last ten years to see that a decline in alcohol consumption does not consistently lead to a commensurate decline in alcohol-related harm. I gave various other examples of x not following y in the ASI report, saying:

Moreover, the belief that reducing the affordability of alcohol will inevitably reduce both alcohol consumption and alcohol-related harm has frequently been confounded. Alcohol consumption has fallen in most of Europe and the USA (though not the UK) in recent decades despite rising incomes which have made alcohol more affordable. When the Institute of Alcohol Studies compared alcohol-related disease rates and alcohol prices across European countries, it found “no discernible relationship between affordability and harm” (Institute of Alcohol Studies, n.d.)

Dramatic reductions in the price of alcohol in Scandinavia in 2003-04 provided a natural experiment for price elasticity models to be put to the test, but the results were surprising. Denmark reduced the tax on spirits by 45 per cent in 2003 without experiencing any increase in alcohol consumption (Mäkelä, 2008; Grittner, 2009). Instead, there was a decline in alcohol-related problems (Bloomfield, 2010). Prior to the Danish tax cut, it was predicted that alcohol consumption would soar in southern Sweden because Swedes would cross the Danish border to buy cheap booze. Many did, and yet alcohol consumption in the south fell overall while consumption in the distant north rose (Gustafsson, 2010).

I then gave examples of price increases being associated with lower consumption and less harm before concluding that:

It would be quite wrong to assume that pricing has no effect on alcohol consumption, only that the effects of price interventions are highly unpredictable

That is the simple truth. The inherent unpredictability of people's alcohol consumption has profound and obvious implications for those who try to predict what will happen after any alcohol policy is introduced, let alone a policy that has never been tried anywhere in the world.

Total consumption model

We wrote a bit about the Total Consumption Model//Lederman Hypothesis/Single Distribution Theory in the ASI paper (we later wrote a whole report about it). Simply put, this theory says that if per capita alcohol consumption declines, heavy drinking and alcohol-related harm will also decline. In their response, the Sheffield team say that our reference to this theory is irrelevant.

Duffy & Snowdon are wrong when they assert and discuss over numerous pages that "at the heart of SAPM's projections is the 'single distribution' model, a theory first advanced by... Lederman in 1956". Our research is not based on this theory, which assumes a direct relationship between alcohol consumption in a population and rates of alcohol-related harm...

They go on to say that it is "bemusing" that we write so much about the Lederman Hypothesis/Total Consumption Model/Single Distribution Theory. Whilst it is possible to argue for minimum pricing without reference to this theory - and it is certainly true that the Sheffield model is far more complex than the simplistic Lederman hypothesis - the problem for the authors of the rebuttal is that Tim Stockwell, one of the world's leading proponents of minimum pricing and co-author of Sheffield University alcohol pricing research, is on the record explicitly stating that the Sheffield Model is fundamentally rooted in the Total Consumption Model. He writes:

How was the Sheffield Model constructed?

... The model is based on two fundamental elements that are well established in the much larger literature on the relationship between alcohol consumption and alcohol-related harms:

(i) When the price of alcohol increases consumption by most drinkers goes down including, critically, consumption by hazardous and harmful drinkers;

(ii) When alcohol consumption in a population declines, rates of alcohol-related harms also decline.

All subsequent debate about the Sheffield has centred on the degree of certainty regarding the size of the effects.

As if to avoid any ambiguity, Stockwell makes the Sheffield model's debt to the Lederman Hypothesis even more explicit:

High quality reviews confirm that when total consumption of alcohol in the population declines, consumption among heavier drinkers is reduced and, further, rates of alcohol-related mortality also decline. The Sheffield Model applied these general principles specifically to the UK and provided numerical estimates of the benefits.

If Stockwell thinks the Total Consumption Model is at the heart of the Sheffield model, it's safe to say that it probably is.


The Sheffield researchers use the supposed 'Canadian experience' as real world evidence that minimum pricing saves lives:

A form of this policy exists in Canada and recent studies have begun to evaluate its effects. These studies have shown that increases in minimum prices are associated with falls in alcohol consumption and alcohol-­related deaths.  

This is a reference to a statistical analysis of data from British Columbia conducted by Tim Stockwell (yes, him again). Stockwell claimed that there was a large drop in wholly alcohol-attributable deaths in 2006-07 which roughly coincided with some (fairly minor) increases in the minimum price of some drinks.

Alas, this is entirely inconsistent with the established facts. Official statistics show that the alcohol mortality rate in British Columbia rose from 26 per 100,000 persons to 28 per 100,000 persons between 2002 and 2008. As the graph below shows, neither mortality (solid line) nor per capita alcohol consumption (dotted line) fell during this period.

Between 2002 and 2011, the number of deaths directly attributed to alcohol in British Columbia rose from 315 to 443 with the largest annual death rates occurring after the minimum price rises of 2006. Between 2006 and 2008, when further minimum price rises occurred, the number of deaths rose from 383 to a peak of 448. Moreover, the rate of hospitalisations for both alcohol-related ailments and acute intoxication both rose during this decade. According to the Centre for Addictions for BC - which is Stockwell's own institute:

‘Alcohol consumption in BC has been above the Canadian average for the last decade. The rates of hospitalizations in BC for conditions related to alcohol have shown a significant increase since 2002, reflecting an overall increase in alcohol consumption in the province’.

Quite how Stockwell manipulated these figures to portray a positive effect from minimum pricing remains a mystery, but there was never a fall in alcohol-related harm, there was never a fall in alcohol consumption and - for that matter - there was never much of a rise in the minimum price.

Heavy drinkers

One of our main criticisms of the Sheffield model is that they use price elasticities that have little support in the economic literature. Most economic studies show that heavy drinkers are less price sensitive than moderate drinkers and that addicted drinkers are - as common sense would dictate - least price sensitive of all. The Sheffield model assumes that the reverse is the case - that price increases will make heavy drinkers reduce their consumption more than moderate and light drinkers. This assumption means that their model shows a more profound effect on alcohol-related harm than is realistic.

In their response, they say this:

Since we first published our findings in 2008, we have observed that a common tactic used by those wishing to misinterpret the alcohol policy evidence base is to begin a sentence with one subject before subtly shifting to another subject. We observe a classic case when Duffy & Snowdon say "it is heavy drinkers who cause and suffer the most alcohol-related harm, but can we really assume that someone with an alcohol dependency is more likely to be deterred by price rises than a more casual consumer?" Note here the conflation of heavy drinkers (ie. those drinking above the NHS guidelines) with dependent drinkers (those who are addicted to alcohol).

Leaving aside the assumption of bad faith that runs through these (and many other) sentences, there was no attempt on my part to conflate heavy and addicted drinkers. I was using a dependent drinker as one example of a heavy drinker. Not all heavy drinkers are addicted but all addicted drinkers are heavy drinkers. By assuming that heavy drinkers have a more elastic demand, the Sheffield team are - by definition - assuming that addicted drinkers have a more elastic demand.

Note the attempt at putting words in my mouth by defining a heavy drinker as someone who exceeds the government's risible guidelines. That's not my definition. I was talking about genuinely heavy drinkers in the top five per cent (or so) of the distribution.

All of this is a diversionary tactic to avoid addressing the important point which is that heavy drinkers (whether addicted or not) are less price sensitive than moderate drinkers.

In the ASI report, I quote directly from one of the Sheffield reports which notes that "the 95th percentile of drinkers have an elasticity not significantly different from zero". This, in turn, is a reference to Manning et al. (1995) which concluded that "the very heaviest drinkers" may have "perfectly price inelastic demands". Although the Sheffield researchers acknowledge that this sort of evidence exists, they make the opposite assumption in their model, saying:

By contrast, the elasticity estimates generated here tend to show own-price elasticities with greater magnitude for hazardous/harmful drinkers compared to moderate drinkers.

This breezy assumption goes against the weight of evidence and has major implications for the numbers that pour out of the Sheffield model. As I explained in the ASI paper:

By wrongly assuming that heavy drinkers are more sensitive than the general population to changes in the price of alcohol as a product category, the Sheffield model not only overestimates the putative health benefits to be derived from minimum pricing, but also overestimates the drop in overall consumption that is likely to take place (since heavier drinkers consume a disproportionate quantity of alcohol). Moreover, it underestimates how much poorer heavier drinkers will be as a result.

In response to this, the Sheffield researchers say that they have conducted "further sensitivity analyses on these estimates (including analyses where heavy drinkers are assumed to be less responsive to price changes)". Taken together, the various Sheffield publications run to many hundreds of pages and it is true that if you look hard enough you can find alternative scenarios being modelled. These alternative figures have had no impact on the public debate, however. They have never featured in the various press releases and they have never been quoted in the media or by politicians. Nor do the researchers themselves ever quote them. They might as well be footnotes.

The figures quoted from the Sheffield models are invariably the headline figures from the mainstream model which assumes - wrongly - that heavy drinkers are more price sensitive than moderate drinkers and, therefore, that minimum pricing will have a larger and more positive effect than is likely.

Part two coming soon.


Christopher Snowdon said...

I did wade through the hundreds of pages generated by the Sheffield group back in 2012 and wrote a short piece for Liberal Vision on just some of the inconsistencies it contains. One of the most misleading elements is the claim relating to price elasticity and heavier drinkers that you highlight above. At the time I wrote:

"Faced with the fact that their methodology produces results
inconsistent with other findings for price elasticity in heavier drinkers they do provide an analysis that is consistent with the literature…

'To enable more direct comparability with the estimates in the literature we have also generated elasticity estimates for total alcohol purchasing from the EFS, shown in Table 11. These
are in broad agreement with the literature, showing that - at the highest level of aggregation -hazardous and harmful drinkers (combined elasticity of -0.21) are less price elastic than moderate drinkers (elasticity of -0.47).”

….but then ignore it.

"Note that these high-level estimates are provided for reference only and are not included in the model.' "

They are still ignoring it and it seems getting quite irate with anyone who points out this massive flaw in their modelling and its consequences for their ludicrously precise claims.

Christopher Snowdon said...

I'm trying to be totally objective here, but I would have thought that heavy / dependent drinkers would be the least deterred by price. The effect would be for those who found themselves priced out of the market to either 1). turn to criminal activity to fund their now unaffordable habit, or 2). find an alternative source of their preferred oblivion and turn to (doubtless newly created due to demand) the 'alternative' market of illegally produced high-strength alcohol which has enjoyed none of the normal QC checks. Or taxes.

Either way, it looks to me to be a disastrous policy.

Christopher Snowdon said...

Yup, they are least deterred by price.

Christopher Snowdon said...

You really don't need a degree to work that out but Nicola Sturgeon disagrees.

Christopher Snowdon said...

Nicola Sturgeon is a politician with a background in community law centres. There's no reason to expect her to have a clue.
But it's depressing that the Sheffield modellers, Stockwell and most of alcohol 'research' crowd aren't any better - in fact they're worse, as they're just as epistemologically challenged but pretend to know what they're talking about.