Thursday 5 July 2012

Price elasticity of alcohol demand [Updated]

Otago's Jennie Connor cites some numbers on the elasticity of alcohol demand with respect to price that just seem off [HT:].
But Professor of Preventive and Social Medicine at Otago University, Jennie Connor, says price does influence alcohol consumption.
She says studies show a 10 percent increase in the minimum price of alcohol reduces consumption 16 percent.
Professor Connor says it's a very efficient and effective way of reducing problems from alcohol, because it targets heavy drinkers the most.
I've been citing the Wagenaar meta-study results showing an aggregate alcohol demand elasticity of -0.44; she's saying it's -1.16 -1.6. There's a bit of a gap there.

When I look at Table 1 of Wagenaar's metastudy of 112 different studies of the price elasticity of demand, I can't find a single one that has an aggregate elasticity lower than -0.92 at the lower end of the confidence interval; the lowest point estimate is -0.84. Recall that you need to hit -1 to have a product that's price elastic. But those are aggregate estimates; maybe there's just something different about elasticity at the bottom end of the price distribution.

So let's have a look at a very recent Australian issues paper examining the case for minimum pricing. This one comes from the Australian National Preventative Health Agency. What do they say about demand elasticity at that end of the distribution?

First, we have to be awfully careful in distinguishing between own-price and aggregate category effects. If you look at the effects of a price increase in one category of product, you'll likely overestimate aggregate price responsiveness because price-sensitive shoppers will flip to another alcohol product category if one category's prices change.

Who does the Australian National Preventative Health Agency cite? Wagenaar, like I do: Wagenaar reports aggregate elasticity across all alcohol categories. They cite two other meta-studies, one of which, Gallet, also reports an aggregate category elasticity that's pretty much identical to Wagenaar's. Where Wagenaar gets -0.51, Gallet gets -0.52.* The last one only reports elasticity by category, which is next to useless for reckoning changes in response to minimum prices or cross-category excise increases, but within-category findings are about the same as Wagenaar and Gallet were getting in those kinds of estimates - more elastic than aggregate elasticity, but still nowhere near -1. Wagenaar gets -0.8 for spirits; people shift from spirits to other products more quickly when the price of spirits jumps.

Next, ANPHA cites some Sheffield figures on aggregate price elasticities for overall alcohol consumption: -0.47 for moderate drinkers and -0.21 for hazardous and harmful drinkers. They note too that, when we look at own-price elasticity within product categories, hazardous and harmful drinkers are more price elastic than moderate drinkers: they're more likely to shift product categories. But that tells us zilch about what harmful drinkers do in response to a price increase for the entire product category; it would be misleading to use this kind of data to claim that harmful drinkers are the most price responsive. They're most price responsive when their preferred brand or product changes in price but they're also least responsive to aggregate changes in alcohol prices.

Finally, ANPHA look at some evidence from Canadian experiments with social reference pricing. At page 16, they cite evidence from British Columbia where a 10% increase in alcohol's minimum price resulted in 3.4% reduction in aggregate consumption: again, a finding consistent with the numbers I'm citing, and not consistent with Connor's.

ANPHA cite a bunch of other Sheffield simulation results showing relative price inelasticity but greater simulated total consumption changes among harmful drinkers than among moderate drinkers. Simulation results here are going to be pretty sensitive to parameter estimates, but it's also the case that a smaller percentage consumption reduction among heavy drinkers than among moderate drinkers can easily generate larger total consumption decreases among heavy drinkers.

The ANPHA issues paper doesn't make policy recommendations; it just looks at issues around minimum pricing. But they do include in their appendix a paragraph noting their prior recommendation: that alcohol move to a tiered volumetric tax that looks an awful lot like New Zealand's current system where spirits are more heavily taxed per unit alcohol than beer and wine, which are more heavily taxed per unit alcohol than low-alcohol products. I'm not endorsing the ANPHA paper or its recommendations,** but its survey around the price elasticity of demand is pretty much what I've been finding in my own searches through the literature. And I can't find anything in it that would support suggestions that alcohol consumption is relatively price elastic to regulated minimum prices.

I'd love to know where Connor's getting her elasticity estimates. They are completely outside of any plausible range. Again there is not a single paper among the 112 papers cited by Wagenaar that comes within cooie of the numbers she's suggesting would here apply; not a single one finds that aggregate alcohol demand is relatively elastic. O'Connor's suggesting an absolute price elasticity of demand of 1.16 1.6. Zero of the 112 cited by Wagenaar finds an absolute elasticity greater than 1. ANPHA does not cite a single finding suggesting an absolute elasticity greater than 1. Maybe there's a study out there somewhere that supports it, but I'd be awfully uncomfortable being cited in the press using number that are that far out of whack from the rest of the literature.

Finally, Connor says that minimum pricing targets heavy drinkers the most. It's true that they'll take the biggest dollar hit from an increase in minimum prices. But they are still less responsive relative to their consumption than are moderate drinkers. Imagine, for instance, claiming that rich fat people are most affected by a 100% tax on food because they wind up having the biggest increase in what they spend on food. Yeah, it's true. But the smaller absolute reduction in the quantity consumed by poor thin people matters too, especially if they cut their consumption by a much larger proportion than do the rich folks.

UPDATE: I've found the study Connor is citing, or at least I think I have. And, I think it's the same one that ANPHA is citing. Remember how I noted at the start that you have to be careful to look at aggregate changes rather than just own-price? Well, ANPHA was careful about that. Connor is citing the figure from the same study that looks only at what happens to, say, beer consumption if you hike the price of beer while leaving the price of wine constant. Suppose Honda increased the price of its cars 10% and we saw a 16% drop in Honda sales as consumers shifted to Toyotas; it would be a bit nuts to sell that as saying that a 10% tax on all cars would reduce aggregate car purchases by 16%. Here's the original paper:
The estimates indicate that a 10% increase in the minimum price of a given type of beverage [EC: eg, spirits, beer, wine] 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). The first estimate may overestimate minimum price effects because  it incorporates compensatory increases in consumption of all other beverages. The estimate of the effect of across-the-board changes in minimum prices on total consumption will be biased to the extent that the extra structure we imposed on the model is unrealistic. 
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.


*Note that I've been citing -0.44: mean elasticity across all the estimates is -0.51, but Wagenaar gives -0.44 for total alcohol after adjustment for study characteristics.

** In particular, arguments around increasing the price of the lowest-cost products in Australia, where the WET means that cask wine can be very cheap, just don't translate to the New Zealand environment where we already have a tiered volumetric excise regime. I also am far more sceptical of results coming from Sheffield's simulation work than they seem to be.


  1. Isn't it a price elasticity of demand of 1.6 not 1.16? Even further from what the evidence suggests

  2. Yes. Stumblefingers. Will fix when at a proper computer.

  3. Good post. Two points: 1. Whatever the total price elasticity of alcohol, since alcoholism is defined as an addiction, it would be highly counter-intuitive (to put it mildly!) to suggest that alcoholics would be more price-sensitive than moderate drinkers. To support such a proposition, you would certainly need some evidence, which, as you point out, isn't there in the literature. It is a lot more reasonable to expect that minimum prices or tax increases will reduce alcohol consumption from moderate drinkers (thus reducing their mostly harmless fun) much more than from problem drinkers. Sounds like poor policy. 2. I'm no expert on alcohol policy, but it seems like a lot of commentary on the issue seems to conflate two problems: binge drinking and alcoholism. While there is a lot of overlap, it seems that they are fundamentally two distinct things. The former is an acute issue of public order and safety (since it is a major cause of violent crime, car crashes not to mention spread of STDs, unplanned pregnancies etc.) whereas the latter is a more chronic health issue (affecting morbidity, labour force participation etc.). Eric -- what is your impression of which is the main focus of public policy discussion in NZ? And how do you think effective policies dealing with those two issues might differ?

  4. On 1: agreed.

    On 2: These get conflated in the studies as well; harmful or hazardous drinking categories will have a mix of binge drinkers who get into the category by virtue of binging and constant heavy drinkers.

    The problem for liberal policy is that while getting drunk increases the drinker's likelihood of committing crimes, driving drunk, or otherwise imposing costs on others, it's only probabilistic. We then have to weigh up the effectiveness of targeting externality-generating outcomes directly by doing things like criminalizing drink-driving or penalizing people who engage in externality-generating activities while drunk versus trying to address the behaviour that increases the risk of bad outcomes among some drinkers.

    I'm not sure that we've really tried hard enough in New Zealand at addressing the actual externality generating activities to give up on that side and go for things that might hit binge drinking. Heck, just going back to having drunk tanks with medical supervision for those drunk and disorderly could go a long way towards curbing problems.

    It's also hard to identify policies that curb binge drinking but that don't also penalize moderate drinkers who don't binge. Price measures hit everybody. One of the best direct measures - requiring that bars not serve those who are obviously intoxicated - also can push people to drinking where they're away from a bartender's supervision.

  5. Actually, the real elasticity is lower. Given the tendency to overinflate demand elasticities (publication bias), the elasticity is probably half of what Wagenaar claim it is.