## Wednesday, 31 October 2018

### Excise to reduce road crash costs?

I often like to borrow Ed Stringham's analogy between harmful alcohol use and speeding. We don't use petrol excise to try to deter speeding, we use road patrols and traffic cameras. Does it really make sense to use alcohol excise to go after problem drinkers?

Linda Cobiac, Anja Mizdrak and Nick Wilson argue the case for an alcohol excise hike to prevent car crashes. The article is not linked in any of the news stories on it, but I think it's this one. Those without institutional subscriptions can find it on SciHub. The study was funded by the Health Research Council, so it's not like you didn't already pay for it.

Before we get to the new work, let's get some context. The old BERL study, using 2005/6 figures, claimed about \$700 million in road crash costs. Now that study had a lot of problems, including double counting. And it counted a pile of costs falling only on the drinkers themselves which would have to be weighed against benefits in a better analysis. But let's leave all that to one side and call it the maximum implausible estimate of the social costs of alcohol-related car crashes in NZ in that year.

Since then, alcohol-related road crash injuries and fatalities have dropped by about 20%. So a current maximum implausible figure should be about 20% lower, but CPI adjusted. Taking those together, a current value would be \$685m (I used a 2005-2015 CPI adjustment after reducing the 2005 costs by 20% to reflect the drop in the crash numbers). This is rough-and-ready rather than precise, but should be in the ballpark if you wanted an update from the old figure. Cobiac et al's work is baselined to 2011. The maximum implausible figure for 2011 would be \$777 million.

Cobiac et al argue for a 15 cent (\$0.15) per standard drink increase in excise. A cheap bottle of wine has about 8 standard drinks and sells for \$8. So a standard drink at the lower end, current-excise-inclusive, costs about \$1. They're then arguing for about a 15% increase in prices at the lower end.

Their modelling work says this would reduce social costs by about \$240 million, plus QALY savings that would bump that up to about \$280 million.

But recall that the maximum implausible estimate of road crash costs for 2011 is about \$777m.

Does it seem plausible that a 15% increase in the price of alcohol at the lower end of the cost scale would reduce the social costs of road accidents by about 30%? Either they have a much higher overall cost estimate for road crashes (and recall that BERL's was implausibly high), or something strange in the modelling.

Unfortunately, I cannot find the online supplemental materials where details on the elasticity figures used are meant to be hiding. The main paper explains the main method, but doesn't have the elasticity figures used.

The modelling work they describe first estimates the effects of an excise increase on consumption. They do not report whether they use lower price elasticity estimates for the binge and heavier drinkers they take as the population likely to drink drive; I expect the supplementary tables would have more detail.

But they then scale up all consumption "to account for survey under-reporting". Basically, survey measures don't line up with sales figures - less is reported by survey respondents than gets reported in the tax data. Some of that will be under-reporting, but some of it will be that New Zealand gets a ton of tourists, and tourists are going to drink while here but don't show up in the alcohol use or household expenditure surveys.

After getting an estimated effect of excise on consumption, which will overstate things if they haven't used a lower elasticity for binge drinkers, they then estimate the effect of lower consumption on drink driving:
Modelling health effects of changes in alcohol consumption

From the change in alcohol consumption, we determined the change in risk of motor vehicle and motorcycle road deaths and injuries using a population impact fraction approach (online supplementary text S1). In these calculations, we applied dose–response relative risk curves for high-risk (binge) drinkers.19 These relative risks were adjusted to reflect the proportion of the day spent at increased risk, using a function that predicts exposure time based on estimated rates of alcohol clearance by the liver.5

To model the future health impact of alcohol consumption changes in the New Zealand population, we used multistate life-table modelling methods used previously to model the health impacts of alcohol taxes in Australia20 and Denmark,21 and previously adapted in New Zealand for modelling health impacts of tobacco taxes.22 In these analyses, we focused on modelling the impact of changes in alcohol intake on motor vehicle and motor cycle road transport injuries.
So they use that to go from a modelled reduction in binge drinking to a modelled reduction in motor vehicle accidents. The modelled reduction in costs comes out of that modelled reduction in accidents. But at the end of that chain of modelled links, they wind up claiming an implausibly large reduction in road accident costs from a 15% increase in the cost of cheaper alcohol.

They claim that their results are robust to different assumptions about elasticities of consumption with respect to price, but all that meant was that, in the absence of NZ estimates, they first applied UK estimates and then applied Australian estimates. It doesn't say how robust their figures are to different assumptions about the relative elasticities of binge and regular drinkers.

Anyway - colour me more than a little bit sceptical. If I were looking for best-buys for reducing drink driving, I'd be looking hard at South Dakota's 24/7 and Hawaii's HOPE programmes.

For a fun alternative way of approaching the problem, here's Robert McClelland and John Iselin of the Urban Institute and Brookings Institution's Tax Policy Center. They use a synthetic control method to estimate the effects of two large excise increases in Illinois. The synthetic control method looks for states whose consumption patterns were a lot like Illinois's patterns prior to the tax change, then use the combination of those states to build a synthetic version of Illinois - what consumption in Illinois would have looked like absent the tax change. The effect of the tax is then estimated against that counterfactual. They find a temporary reduction in accidents in counties that don't border other states, but even that washes away pretty quickly. There'll be other studies that find reductions in accident rates, but I kinda like this synthetic control method for establishing a counterfactual.