Friday 25 April 2014

Increasing alcohol excise is great, if you assume the right things

Imagine that hazardous drinkers really really cared about the price of alcohol. If you increased the price of alcohol just a little bit, they'd stop drinking harmfully. Imagine further that moderate drinkers didn't respond very much to prices: what does it matter to the rich Chardonnay-sipping set if a bottle is $8 or $40? If that were the true state of the world, we would have a very simple solution to alcohol problems: hike excise taxes. Harmful drinkers would stop drinking and would stop doing alcohol-related harmful things; moderate drinkers would pay more but that would just be tax revenue for the government. Since they wouldn't change their consumption by very much, deadweight costs would be pretty small relative to the harms avoided. Yay taxes!

Unfortunately, the world don't quite look like that. Our best evidence on it remains Wagenaar's metastudy showing that heavy drinkers respond to a 10% price hike by reducing consumption by 2.8%; average consumption drops by 4.4% with the same price increase. Moderate drinkers respond more to price increases than do heavy drinkers.

Even worse, Byrnes et al show that heavy drinkers' price responsiveness mostly comes from their reductions in drinking on low-drinking days: they basically save up to be able to continue binging on the weekend. So their reduction in consumption is in the part of their consumption that does the least harm. Boo taxes!

Enter the NZ Government report on excise and minimum pricing. Fortunately, the Minister has more sense than her Ministry and hasn't gone ahead with minimum pricing; hopefully, she's not looking at excise. What's the problem with the report? They started by assuming that heavy drinkers are more responsive to prices than are moderate drinkers.

And they know it's wrong. Here, at Table 5, they show the general consensus of the international literature: heavy drinkers don't respond to prices nearly as strongly as do moderate drinkers.

The Wagenaar numbers vary a bit from the -0.44 that I tend to cite as average; I usually go for the weighted measure. Bottom line: heavy drinkers are roughly half as responsive to prices as are moderate drinkers. That's page 20. And they cite Byrnes accurately at page 21.

But then what do they go and do? They started by trying to get SHORE to estimate NZ elasticities, but something went wrong there: the elasticities were completely out of whack with reality. Reading between the lines at page 25, it looks like SHORE was using the increase in purchases of products on special at supermarkets as part of its price elasticity estimation, and that just ain't right. If you switch brands because something's on special and buy more of it than you otherwise would have, that isn't the same effect as you'd expect for across-the-board price changes you get with excise or minimum pricing.

As the report rather bluntly puts it:
"It was decided that the significant reductions in consumption estimated using NZ elasticity estimates are not a realistic representation of what is likely to happen in reality and are contrary to all international evidence of the responsiveness of alcohol consumers to changes in price."
Rather than discard the completely nuts NZ numbers, they let those figures stand and added alternative numbers as robustness checks. Those big headline estimates you've been seeing in the papers about just how awesome excise is? They're based on the numbers that, according to the report, "are not a realistic representation in reality and are contrary to all international evidence of the responsiveness of alcohol consumers to changes in price."

Example? A 133% excise hike means about a 40% increase in the cost of low-priced beer, a 44% increase in the cost of low-priced wine, a 45% increase in the cost of low-priced RTDs, and a 103% increase in the price of low-cost spirits. The heavy drinkers SHORE estimated a 61% reduction in harmful consumers' consumption with that tax hike. So they're saying that harmful consumers are more than unit elastic. That's just not right.

The other problem with the SHORE numbers is that they couldn't distinguish between heavy and moderate drinkers. Everybody's elasticity was equally overestimated.

Because those numbers were so out of whack, they also ported in some Sheffield elasticity estimates from the UK. Problem with those estimates is that, while they're about right on average, they've messed up the relative elasticities: they have harmful drinkers being more price responsive than moderate drinkers. I'm not sure why they didn't pull in the 2008 Sheffield estimates noted at Table 5. At Table 25 they say, using Sheffield, that a 133% excise increase would reduce low risk consumption by 18.6% while reducing harmful consumption by 21%. If that 133% excise increase corresponds to about a 45% price increase, then Wagenaar's estimates would say we'd get only a 12.6% decrease in heavy consumption and a 28% decrease in moderate drinkers' consumption (using the figures from Table 5). Heavy drinkers' real-world price responsiveness is 60% of that advertised, while moderate drinkers' responsiveness is 150% of their figures.

Now, why does this matter? The more responsive are moderate drinkers to price measures as compared to heavy drinkers, the more expensive is any bit of harm reduction in terms of harms imposed on moderate drinkers. And the folks who wrote the report know this matters too! Here's what they wrote at page 85:
Overall it appears that excise increases have a greater impact on harmful drinkers than low risk drinkers, based on University of Sheffield elasticity estimates. This is driven by the greater own-price elasticities, particularly for spirits. However, this result is inconsistent with findings in studies such as Wagenaar et al (2009), which found that heavy drinkers are much less responsive to price changes (with an elasticity of -0.28 compared to -0.62 for all drinkers). The University of Sheffield also found that harmful drinkers are much more price inelastic compared to low risk drinkers when total alcohol consumption was considered, rather than consumption by beverage type.
We also do not have separate elasticities for per occasion drinking, and recent evidence indicates that people are much less price responsive during drinking occasions (Byrnes et al, 2012). Therefore there is a risk that the effects on purchases could have been over-estimated for per occasion purchases. 
Therefore we cannot conclude with confidence that excise increases will have a greater impact on harmful drinkers. More research is needed to confirm this, which could be done once revised University of Sheffield elasticity estimates are available.
Here are but a few of the the ways this will affect their analysis:

  • Crime reduction benefits are overestimated both because heavy drinkers are modelled as being far more price responsive than they really are and because crime seems more responsive to binge drinking than to longer term heavy drinking, and it's the latter that seems more affected than the former among heavy drinkers. 
  • The health effects will be overestimated where harmful drinkers are modelled as sharply curtailing consumption and where we underestimate by how much moderate drinkers shift into non-drinking and miss out on the health benefits of moderate drinking.
  • The deadweight costs facing moderate consumers are underestimated in the Sheffield figures that understate moderate drinkers' consumption elasticity.
  • All the other benefits that require consumption reductions among heavy drinkers will also be overstated.
Tables 40 and 41 have a whole whack of sensitivity checks. What if population growth is lower or higher? What if we use a higher or lower discount rate? Nowhere in the 21 sensitivity analyses is the one that really matters: "What if we use a sane, international-consensus measure of relative price elasticities?"

Whether this is enough to overturn their net benefits finding - I can't tell without a pretty extensive bit of work. Just linear extrapolations from the changes in harmful and moderate consumption won't do it: deadweight costs will increase extraproportionately, and harms are likely exponential in heavy consumption.* But I'm pretty sure that Collins was dead right in not relying on this stuff to justify imposing minimum pricing: the case hasn't been made. 

A few other potential problems on a cursory reading:
  • They estimate the productivity costs of those showing up to work with hangovers under the assumption that those prone to showing up to work with hangovers have average productivity characteristics but for their propensity to show up for work with hangovers. I rather suspect that less conscientious workers are more prone to showing up unfit for duty, and that this affects more than just hangovers.
  • They estimate the productivity costs of taking a day off for a hangover under the assumption that the kinds of people who take a sickie for a hangover wouldn't have used up that free sick day for some other recreational purpose later in the year. This also seems implausible. 
  • Their analysis of alcohol-related unemployment is bereft of consideration of comorbidity between alcoholism and depression and the independent effect of underlying characteristics on employment.
  • Frictional costs to employers of replacing fired or deceased workers seem predicated on an assumption that the employer would never otherwise have to have replaced that worker.
  • Their crime estimates count as an alcohol-caused crime any crime committed by someone arrested within 12 hours of offending and presenting as at least moderately intoxicated. While this may underestimate things by excluding those who aren't caught until much later, it also says that every one of those moderately intoxicated arrestees would never have committed the offence but for the alcohol. Not sure which way things will cut on that one, but I'm awfully sure that they'd do better by just porting in some decent estimates of the elasticity of crime with respect to alcohol out of something like Carpenter. 

I'll be interested to see what they come up with when they have a second go at this, with some sane elasticity figures. 

On the plus side, Footnote 31 points to my blog post on likely effects of minimum prices on producers. I'd have recommended this one instead.

* At Table 37, they give a $268,185,000 cost of a 133% excise hike. If moderate consumers are 150% as responsive as they're reckoning, then a linearisation would put that cost up to $402 million. At Table 30, they give a value of harm reduction in Year 1 of $740,344,000. If heavy consumers are only 60% as price responsive as they're reckoning and if harms were linear, that knocks the benefits down to $444m. So we're down to a net benefit of about $40 million, without considering that deadweight costs will be higher than implied by linearisation, and without considering the other problems noted above. Whether the linearisation on harms here overstates or understates things gets complicated: for any individual drinker, harms are strongly non-linear, so if we had a pile of really really heavy drinkers estimated to become moderate drinkers, but instead only become heavy drinkers, then my linearisation overestimates how much the report overstates harm reduction: the reduction in harm in moving from really really heavy drinking to heavy drinking is bigger than the reduction in harm from moving from heavy to moderate drinking. But if instead we get the harm reduction from greater proportions of heavy drinkers shifting into moderate drinkers than really would, then the linearisation would be about right. And if prices are really bad at hitting acute drinking that contributes most greatly to harms, then even my linearisation understates the extent to which the report overstates harm reduction. It's complicated, and I haven't access to the guts of this machine. 


  1. Can't like this enough. Thanks for this thorough and consistent analysis.

  2. "crime seems more responsive to binge drinking than to longer term heavy drinking"

    Do you have a citation for this one?

  3. this is not good Eric,some people, even me possibly say you are in cohort with alcohol company, best write about other statistic, maybe the value of Canadians in NZ, see what happens.

  4. It is the Ministry paper's assumption at Section 6.4

  5. Ah, thanks. I misread your wording to refer to results of empirical studies.

  6. yes, have we had enough yet he does go on does he not, is he being paid is he alcohol industry, is he like bryce, just another tool

  7. Perhaps a similar logic would also apply for the availability of on-premise alcohol?

    If you close bars earlier the heavydrinkers wont respond as much as moderate drinkers because they will be more motivated (stubborn to the hint?) and be more likely to shift their drinking to earlier in the night. So, the heavy drinkers' habits will be condensed in a shorter time period which could have the effect of increasing the risk of alcohol-related harm (of self or others) for the specific group of people who are the highest risk.

  8. Or to paraphrase, "whaaa whaaa! He is right but I dont want to admit it, even to myself! whaaa whaaa!"

    You promise more later, but you needn't bother. You have completely embarrassed yourself already.

    How does everyone else know Eric is right? Even ignoring the facts and rational logical argument he provided and the report authors declined, it is probably sufficient just to see that YOU think he is wrong!

  9. It will be really interesting to see whether we can get decent data out of changed local alcohol policies for testing this.