Alcohol wage puzzles: Churchill-Stalin edition

There's a longstanding alcohol wage puzzle: drinkers earn more than non-drinkers even after correcting for a bunch of stuff. Chris Auld found that moderate drinkers earn 10% more than non-drinkers and that heavy drinkers earn 12% more than non-drinkers; plenty of other studies have found similar effects.

One not-unreasonable explanation is that drinking together builds trust among co-workers, making them jointly more productive (and higher paid) despite the occasional productivity-reducing hangover.*

Radio New Zealand provides a nice bit of anecdotal evidence.

Josef Stalin and Winston Churchill had an all-night drinking session in Moscow that lasted until 3am during World War II.

Relations between the Russian and British leaders were stiff until Churchill arranged a late-night banquet in August 1942 with Stalin, according to files held by Britain's National Archives.

The event was recorded by Foreign Office permanent under-secretary Sir Alexander Cadogan.

The mood was "merry as a marriage-bell," he added, though Churchill was complaining of a "slight headache" when Cadogan came to find him at 1am.

The two men did not engage in much military talk during the meeting, which went on until 3am.

The evening was dubbed a success by Cadogan, as the two men got on.

"Certainly Winston was impressed, and I think the feeling was reciprocated," he wrote in a letter.

"We broke up soon after 3(am), giving me just time to get back to the hotel, pack, and leave for the aerodrome at 4.15(am)."

The BBC reports the letter was among almost 600 government files dating from World War II and the early years of the Cold War, released by the National Archives.

Do all-nighters really end at 3am?

I wonder whether the anti-alcohol activists of 1942 would have bemoaned the wartime losses caused by Churchill's slight hangover the next day while ignoring that getting on well with the Soviets was rather more important.

* Note that Google Insights for Search provides reasonable evidence that hangovers are concentrated on the weekend. I mean, look at this and tell me that the folks who take average weekly hangovers as a measure of productivity losses aren't just a bit mischievous.

A global phenomenon

So it isn't just New Zealand's Ministry of Health who seem to apply different standards of evidence on the health effects of alcohol depending on whether, for any particular disorder, alcohol consumption seems to ameliorate or worsen the condition. I'd written:

The New Zealand Food and Nutrition Guidelines for Older People [Updated: see here] talks a whole lot about the risks and very very little about the benefits. Recall that the J-curve is strongest for middle aged and older people.

Alcohol causes a range of adverse effects on health, including cirrhosis of the liver, pancreatitis, endocrine disorders, cardiomyopathy, gastritis, high blood pressure, haemorrhagic stroke, and cancers of the mouth, pharynx, larynx, oesophagus, breast and liver. It also contributes to death and injury on the roads, drowning, suicide, assaults and domestic violence, other non-traffic related mortality and morbidity, and some mental health disorders and sexual health problems. High levels of alcohol use are also associated with alcohol dependence and abuse (Ministry of Health 2008a). Moderate alcohol consumption may have some benefits for older people, but further research is needed to fully understand the potential benefits.

Every risk is conclusively proven; every benefit needs further study.

There's an interesting letter in the latest issue of The Lancet. Michael Roerecke and Jurgen Rehm recently put out another metastudy on the cardioprotective benefits of low to moderate alcohol consumption. They don't recommend that people start drinking for heart health, but they find the cardioprotective benefits awfully hard to deny.

Well, Tim Stockwell tries to deny it. He's the one who recently suggested that minimum pricing for alcohol in British Columbia strongly reduced alcohol-related mortality rates despite that the aggregate B.C. time trend showed no particular decline in mortality; I suppose the counterfactual had to have been a stronger worsening.* Anyway, in a letter to Addiction, he keeps wondering if it's possible ever to derive any evidence of protective benefits from population epidemiological studies because of potential uncontrolled confounding. Except, as you'll recall, that most of the evidence we have of the harms of heavy alcohol use hardly come from experimental studies where we tell the treatment group to drink a fifth of whisky a day while telling the controls to abstain; it's the same kind of epidemiological evidence which is also subject to potential confounding from that heavy drinkers would not be like moderate drinkers even if they didn't drink: the things that predispose you to alcoholism seem likely to predispose you to all kinds of other negative outcomes. Confounding is everywhere.

And so I just loved Roerecke and Rehm's reply to Stockwell in the latest issue. They write, in a beautiful glove slap:

Given the above points and further points made in [1], it seems that some researchers in the field may be using different standards in assessing the cardioprotective effect of alcohol vs. its detrimental effect. Consider two examples. One is the effects of alcohol on colon cancer [12]. Would the same arguments used to judge the relationship between alcohol and ischaemic heart disease not hold for this relation as well? The other example is the more than 200 other risk relations between alcohol and disease and injury outcomes [13]. Of course, this is not a good argument against scrutinizing the cardioprotective effect of alcohol, but we sense a desire by some in the field to apply tough standards on protective effects and more lenient standards on other effects, where sometimes the responses to very simple survey questions such as ‘Did your partner’s alcohol consumption contribute to your marriage problems?’ are accepted as causal evidence.

 It seems that I'm not the only one frustrated by non-truth-seekers.

* The study is here; I've not had a chance to assess it in much depth. I'd be curious whether running the CPI in place of minimum prices in their series would yield similar effects to those found - there isn't that much change in the minimum price over the period, and they use CPI adjusted prices. That has to be most of their price variation over the interval. I'd also be a bit nervous about using a panel estimator where my main regressor of interest had no panel variation. And, I'm not sure that I've ever before seen this line in an empirical paper, or anything like it:

Estimates for acute AA deaths prevented are not provided due to both counter-intuitiveness and the lack of stability in the lagged associations for these short-term outcomes.

I'm also not quite sure what to think about their using lengthy lags of prices where most of the time series price variation is just CPI. Surely lagged consumption would have a better basis in theory for predicting current acute injury than would lagged prices where most price variation is CPI. I also note that while Chris Auld, an applied econometrician whose work I trust, had worked with this group on price elasticity estimates a while ago, his name isn't associated with the group's more recent adventures. So I'd have to check the group's new work in reasonable detail before I'd believe it.

Social Costs of Healthy

Chris Auld was prescient. If we require corrective policy to internalise pecuniary costs running through the public health system, then we have to do it across the board. And what if it turns out that healthy people wind up costing more because they have a longer retirement in which they consume lots of subsidized health services? He wrote:

If healthy behaviors wind up increasing lifecycle health care costs, we should either subsidize less than we otherwise would, or perhaps even tax, healthy behaviors. Healthy behaviors in this scenario benefit the person exhibiting the behavior but impose costs on everyone else, and this logic demands that we discourage healthy behavior relative to whatever policies we would otherwise have enacted.

This argument does not sit well with me.

A new article in PLOS Medicine finds that the lifetime health costs of the healthy are indeed highest:Non-smokers of moderate BMI imposed the highest lifetime costs. [HT: @Dick_Puddlecote, who points to the Telegraph's report on the study.] Update: I'd missed the date on the PLOS online study; it's from a few years ago. So Auld's prescience may have been overstated. Thanks to Chris Snowdon for the correction.

From the article's conclusion:

In this study we have shown that, although obese people induce high medical costs during their lives, their lifetime health-care costs are lower than those of healthy-living people but higher than those of smokers. Obesity increases the risk of diseases such as diabetes and coronary heart disease, thereby increasing health-care utilization but decreasing life expectancy. Successful prevention of obesity, in turn, increases life expectancy. Unfortunately, these life-years gained are not lived in full health and come at a price: people suffer from other diseases, which increases health-care costs. Obesity prevention, just like smoking prevention, will not stem the tide of increasing health-care expenditures. The underlying mechanism is that there is a substitution of inexpensive, lethal diseases toward less lethal, and therefore more costly, diseases [9]. As smoking is in particular related to lethal (and relatively inexpensive) diseases, the ratio of cost savings from a reduced incidence of risk factor–related diseases to the medical costs in life-years gained is more favorable for obesity prevention than for smoking prevention.

Here's the graph of the expected costs and benefits of smoking and obesity prevention, imagining a costless intervention that would switch the obese or smokers into normal-weight non-smokers. For the first few decades after the assumed-costless intervention, all's great. And then...

If it were free to turn smokers into non-smokers, and if we ignore tobacco excise revenues entirely, they say the costless intervention only passes a fiscal cost-benefit analysis for discount rates higher than 5.7%. If we remember that tobacco excise revenues are heavily front-loaded, being paid often decades before the health costs obtain, then you're not going to find a discount rate where the costless intervention saves the government money.

They warn that they've only focused on health care costs and have left aside productivity costs. But the bulk of productivity costs are borne by the drinker, smoker, or eater himself: they're reflected in lower wages.

Smokers remain the benefactors of the rest of us - voluntarily paying ridiculous levels of tax and then dying before taking much out of the superannuation system.

It's a mistake to model governments as unitary agents. The zealot parts of government want to ban tobacco; the sane parts worry about revenue consequences. I wonder, as does Lionel Shriver, whether worries about lost excise motivate government antipathy towards electronic cigarettes. The zealot side of government doesn't like them because they let smokers keep having fun without moral consequence; the fiscal side doesn't like the lost revenue: bootleggers and baptists in different government departments.

Meanwhile, the British Labour Party contemplates bans on sugary breakfast cereals. I agree entirely with Alex Masse's piece at the Spectator. Alex writes:

Yet one of the features of our society is the steady accumulation of influence – and increasingly of power too – of what might be termed the Government-Health-Security Complex*.

Sometimes slippery slopes really do exist. Some folk warned that the public health industry – that is, the Government-Health-Security Complex – would never be satisfied with its battles against tobacco and alcohol and that it would, in time, launch fresh offensives against fast food, soft drinks, and all things salty an sweet.Don’t be silly, we were told. That’s different. Well, who looks stupid now?

Like so much else this is also, in the end, a question of power and class. The NHS – treated as some kind of secular religion – is to be used as a means of shaming the population (especially the bestial lower orders) into behaving in a more comely, acceptable fashion. The class prejudice inherent in all this is rarely far from the surface. The common people are revolting. Their pleasures must be taxed or, wherever possible, suppressed entirely (see extending the ban on smoking in working-class clubs for example).

And, always, the message is simple: the people – poor, lardy, wheezing, sods – are too stupid to make their own choices and it is government’s role to save them from themselves.

Next step then is plain packaging for breakfast cereals....

Check your sources

Doug Sellman in the opinion section of today's Christchurch Press claims to have had his numbers right [not yet online]. Let's check the history here.

Here's Sellman and Connor's original press release:

Who is advising the Prime Minister on alcohol reform?

Mr Key announced today that he doesn’t believe that minimum pricing for alcohol will change the amount people drink.

“This is contrary to the scientific evidence base about alcohol pricing in general and minimum pricing in particular” said Prof Jennie Connor, medical spokeperson for Alcohol Action NZ.

“Mr Key states that what typically happens is people move down ‘the quality curve’ and still get access to alcohol. Where does this information come from? On the contrary, minimum pricing specifically targets the very cheapest alcohol options and is predicted to reduce average consumption by removing high-alcohol low-cost products from the market.”

“A recent Canadian study has shown that a 10% increase in the minimum price of alcohol reduces its consumption by 16% relative to other drinks”. [emphasis added]

“And these latest data are consistent with the scientific literature which indicates that increasing the price of alcohol has a positive impact on reducing heavy drinking”.

This is very clearly saying that there are very large price effects of increasing the cost of the lowest-priced alcohol. I had initially found the Newstalk ZB report and wondered whether she'd been misquoted, before finding the press release.

I wrote:

Connor has to have been misquoted here or the journalists left out a couple of subsequent clarifying sentences. The error is in the press release. Oh dear.

The link there is now deprecated, but the Scoop link still works.

In today's press, Sellman says that they had it right all along:

On July 3, 2012 we issued a press release recommending the government enacts a minimum price per standard drink of alcohol – to eliminate ultra-cheap drinks favoured by binge drinkers, young drinkers and heavy drinkers – and pointed to a Canadian study that showed ‘‘a 10 per cent increase in the minimum price of alcohol reduces its consumption by 16 per cent relative to other drinks’’. Our wording was based closely on the paper’s wording: “Longitudinal estimates suggest that a 10 per cent increase in the minimum price of an alcoholic beverage reduced its consumption relative to other beverages by 16.1 per cent (p0.001).”

Two days later Crampton wrote a damning critique of the press release on his personal blog, using the same arguments he later used in this Press article. However, it appears Crampton based his critique on a short Newstalk ZB news report of the press release, which quoted Jennie Connor saying, ‘‘studies show a 10 per cent increase in the minimum price of alcohol reduces consumption 16 per cent’’. Note: the reporter had cut off the words ‘‘relative to other drinks’’, which would indeed be wrong if she had said it.

Crampton’s blog piece, and his later Perspectives article, ridiculed the presumed mistake: ‘‘Can a 10 per cent increase in the minimum price of alcohol really reduce total alcohol consumption by 16 per cent?’’ he wrote. ‘‘No’’. But if he had taken the basic precaution of checking the primary source, our press release, he would have seen the words ‘‘relative to other drinks’’ and realised that we had not misquoted the Canadian study at all.

Thus Crampton’s main argument in the Press article was based on his own simple and avoidable mistake, which seems careless for a senior lecturer.

The problem isn't that the Newstalk piece left off the words "relative to other drinks" but that Sellman and Connor used that estimate as though it were relevant to average consumption and where "other drinks" would be interpreted as something other than other categories of alcoholic beverages.

Further, they might have noted that my post of 10 July quoted the press release accurately; my post of 5 July had cited the NewsTalk reporting.

If we look a bit further down the Auld paper, we see pretty clearly what Sellman and Connor had missed:

"The estimates indicate that a 10% increase in the minimum price of a given type of beverage reduced consumption of that type by about 16.1% relative to all other beverages, and a simultaneous 10% increase in the minimum prices of all types reduced total consumption by 3.4% (p<0.01 in both cases)."

Sellman and Connor were building a case in their press release that the Prime Minister was way off base in claiming that raising the minimum price would not have large effects on drinking. Whether "relative to other drinks" is included or not is irrelevant where the context suggests that "other drinks" means drinks other than alcohol.

And so I sent the letter below to the Press this morning:

Doug Sellman in Monday's Press claims to have had his numbers right all along. In his press release of 3 July, he and Jennie Connor wrote:

“Mr Key states that what typically happens is people move down ‘the quality curve’ and still get access to alcohol. Where does this information come from? On the contrary, minimum pricing specifically targets the very cheapest alcohol options and is predicted to reduce average consumption by removing high-alcohol low-cost products from the market.”

“A recent Canadian study has shown that a 10% increase in the minimum price of alcohol reduces its consumption by 16% relative to other drinks”.

The rather obvious interpretation of their release, which was highly critical of the Prime Minister's claim that minimum prices would not greatly affect consumption, was that we should expect a sixteen percent reduction in consumption of alcohol relative to other drinks were the minimum price of alcohol to rise by ten percent.

The paper on which their analysis was based does indeed have a quote that reads a lot like Sellman and Connor's. But, it refers to the effects you get if the price of one category of alcohol - like beer, wine, or spirits - rises relative to other categories of alcoholic drinks. It isn't talking about the consumption of alcohol as compared to fruit juice. This is obvious if we read the second clause of the sentence, where Auld and his coauthors write:

"The estimates indicate that a 10% increase in the minimum price of a given type of beverage reduced consumption of that type by about 16.1% relative to all other beverages, and a simultaneous 10% increase in the minimum prices of all types reduced total consumption by 3.4% (p<0.01 in both cases)."

In Monday's Press, Sellman claims not to have misquoted the Canadian study and that they had, all along, meant "relative to other drinks" to refer to other categories of alcohol. If so, it seems odd to have chosen that figure as being relevant to the argument they were building. It could be relevant if we were estimating the likely reduction in consumption of premixed "alco-pops" relative to other alcoholic beverages, but surely the total amount of alcohol consumed matters more than whether it is consumed in one type of alcoholic beverage rather than another. And, for total consumption, the 3.4% figure is the rather more relevant one.

I strongly encourage readers to read the paper on which Sellman's claims are based and to judge for themselves, rather than trusting either of us. An ungated version of it is available here: http://www.vsnews.fr/etudes/Does-Minimum-Pricing-Reduce-Alcohol-Consumption.pdf . Or, go to scholar.google.com and type "Does minimum pricing reduce alcohol consumption?" You will find that the authors there, like Sellman, favour minimum prices. I worry more about harms imposed on lower income moderate consumers of lower cost alcohol. How we weigh the tradeoff between reducing harms from heavy drinkers and reducing consumption benefits from poorer moderate drinkers is a fairly important discussion. But the case for a minimum price for alcohol ought not be based on an estimate of its effects that is roughly five times larger than that which can be supported by the evidence.

About that Canadian study...

Last week, anti-alcohol advocacy group Alcohol Action NZ put out a press release where the University of Otago's Jennie Connor was quoted:

"A recent Canadian study has shown that a 10% increase in the minimum price of alcohol reduces its consumption by 16% relative to other drinks".

I got in touch with one of the authors of what has to be the study to which she's referring.

Chris Auld reported that the -1.6 price elasticity figure indeed only refers to a measure of own-price elasticity. Except it isn't quite own-price elasticity. Because the estimation technique doesn't correct for substitution effects, it combines the own-price elasticity with cross-price elasticity from other products. Quoting from Chris, with his permission:

Suppose we have two types, 1 and 2. Demand for type 1 is x_1( m_1, m_2 ), presumably decreasing in own min price m_1 and increasing in the min price, m_2, of the other type. The panel models recover d(x1 - x2)/d(m_1), so they are not estimates of own-demand slopes. For example, we might estimate -1.6 if the own-elasticity is -0.9 and the cross-elasticity is +0.7. Since we are not controlling for the cross-price, nothing can be said from these models about the effect of increasing both minimum prices---it could be that total consumption is almost invariant to min prices, but we could still generate big estimates from these models if various types of alcohol are strong substitutes. Test statistics against the null that the total effect is zero are still valid, but it's easy to misinterpret what the estimates mean

Chris also confirms that the -0.34 estimate is the one that best reflects the expected effects of an across-the-board price increase like minimum pricing, but notes that the standard kinds of time series problems makes that estimate rather less robust than he'd like.

I think it's [the estimate] probably too high, although it may be in the ballpark - a variety of evidence does suggest that min price changes are quite effective in targetting heavy drinkers.

Wagneaar found -0.28 among heavy drinkers, so I'm less worried about potential lack of robustness around Chris's estimate; if every estimate of this sort has similar robustness issues, then we might worry about systematic overestimation of demand elasticity with publication bias.

Heavy drinkers who consume cheap alcohol will be targeted with minimum prices, but so too will moderate poor drinkers who choose cheap alcohol.

Chris says he's doing some theory work showing that:

the central planner would always like to impose minimum pricing but reduce conventional taxes when confronted by consumers who are heterogeneous in an underlying demand parameter and when externalities are nonlinear in consumption---because there is an externality on the quantity but quality choice, the planner would like people to drink less alcohol, but drink higher quality alcohol.

I agree with Chris on this one - I'd posted a pretty similar point last week. The New Zealand Drug Foundation should perhaps pay attention to this one: having a minimum price should be coupled with excise reductions, not increases; NZDF has been pushing for both a minimum price and an increase in excise. If the ex ante excise were seriously below the optimum, then I'd expect a model like the one Chris is likely working up to say instead that minimum pricing lets us increase excise less than we otherwise would. But aggregate excise here isn't far out from actual external harms from alcohol. And I still worry about effects on moderate drinkers of lower income. The policy seems likely to be severely regressive.

Jennie Connor really should retract her press release or issue a correction. It leads people to believe that a minimum price will have far more effect on harmful drinkers' consumption than can be supported by the evidence. Otherwise, how much weight should anybody place on any "fact" claimed by Jennie Connor in her press releases?

In other scorekeeping, Ross Bell is right and John Key is wrong: a minimum price will increase the average quality of drink consumed, not reduce it. Here's Key:

"Instead of buying a $10 bottle of wine that might go to $15, they'll buy a $5 bottle of wine that'll cost $10. Their outlay is the same, the quality of what they're buying is worse," John Key said.

Competition among retailers, distributors, and producers ensures that drinkers get at least the minimum price's worth of value for the drink they're consuming except where there are other restrictions in the system that allows agents to accumulate rents.


But in that same article, Bell underestimates the number of standard drinks in a bottle of wine; this has the effect of reducing the perceived effect of a minimum price. Bell writes:

If the Government were to set the minimum price for alcohol at $1.50 - a reasonable and workable price - it would mean a seven-standard-drink bottle of wine could not be sold for less than $10.50.

Most bottles of wine are closer to 8 standard drinks than to 7. I had a quick flip through our wine rack. A Pegasus Bay riesling came in at 6.6 standard drinks. Nothing else in the rack rounded to 7 - everything else rounded to 8, except a few Aussie reds that rounded to 9. 8 standard drinks at a $1.50 minimum price is $12, not $10.50. And, though I'm a moderate high-income drinker, I do often buy bottles of wine in the $10-$12 range. The Montana Classics range on special for $9 is typically great value; I never feel bad about using third of a bottle in cooking at the price, and a glass while cooking is generally decent too.

Bell cites a Scottish government study suggesting that moderate drinkers won't reduce their consumption by much in absolute terms while heavy drinkers will have massive reductions in absolute consumption. If that one's based on Sheffield, and if Sheffield there is assuming constant elasticity across moderate and heavy drinkers, I wouldn't put much weight on it. Sometimes Sheffield estimates differential elasticities, sometimes they just assume constant elasticities. I'm not sure what they're doing in this particular one. And I also worry too about differential patterns in how people reduce their consumption. The Australian study I'd cited last week showed that most of the action in price increases is in reducing the number of days of light drinking rather than reducing the amount of heavy drinking, though there are other studies suggesting reasonable price elasticity of binge drinking. 

Finally, Bell cites BERL's (adjusted) figure on alcohol-related harm: $4.4 billion. That's disappointing. Ross, please remember that that study is just terrible. Again,

• They count the VSL from lives lost while simultaneously counting the total value of forgone production from premature mortality. The Ministry of Transport, who puts out the VSL measure, never does this when they tabulate the social costs of car crashes. They count the value of lives lost in accidents, the cost of injuries, and the value of production forgone due to injuries, but they don't count forgone production from those who die. The measure of the value of a statistical life is inclusive of the measure of forgone production. BERL says that VSL costs are $1.52 billion and that labour costs, mostly from deaths, is $1.48 billion. 


• Where their model study, Collins and Lapsley, counted both the health benefits and the health costs of drinking and took a net measure of costs to the health care system, BERL took a one-line assumption that harmful drinking can never have any health benefits as justifying a move zeroing out any of the aetiological fractions where alcohol reduced costs. This was absurd and points strongly to that they just wanted to give the Ministry of Health the very very large number that the Ministry of Health wanted. Even drinking that is on net harmful can have a mix of underlying positives and negatives. 


• They everywhere conflate private and social costs. Ross, you probably want to include all the costs that drinkers impose on themselves. And that's fair enough where you accurately characterise those total costs as mostly consisting of costs drinkers impose on themselves. But neither of the points above have anything to do with that. It's just poor method designed to inflate reported costs. And repeating the "costs New Zealand" line without the qualification makes people think that you're referring to a cost to the taxpayer through the health system rather than a cost drinkers impose on themselves - it's misleading; I hope not purposefully so.

Peter Dunne seems pretty sensible on this one, even if I do curse his name each and every time I want to get cold medicine that works.

"To say that we'll have a minimum price of $12 for a bottle of wine because people who can't afford to pay $12 shouldn't pay a lesser price, but Chardonnay socialists who can pay $25, $30 for a bottle of wine will still be able to get their wine. I think that's a really elitist and ridiculous argument."

The policy would reduce some harmful consumption, but it would also reduce some reasonable consumption from lower income drinkers - and from a few higher income cheapskates like me. We either need more serious work showing that the harms prevented outweighs the harm imposed by the policy, or at least coupling the policy with transfers to those negatively affected.