Wednesday, 28 July 2010

Drink driving: a research proposal

I'm rather pleased that National's ruled out a drop in the drink driving limit pending further research. I'm skeptical that a drop from 0.08 to 0.05 would pass cost benefit, but here's what it would take to convince me. It would be nice if the folks horribly aggrieved by the decision would similarly put their cards on the table: what would you deem a sensible research design, and what would it take for you to change your position?

Here's my go at it.

First, the government's already running random breath testing stations where everyone driving by point X is breath tested for blood alcohol. Start keeping data logs on all of these. We need to know the blood alcohol level of drivers on the roads at different times of the day. That gives us a base rate.

Next, collect data on all accidents: every driver after every accident attended by police is breath tested. Use this, in combination with the base rate, to get the accident rate for folks at different blood alcohol levels. But, be sure to correct for time of day: if a lot of drink driving accidents are at night, we need to know how much of this is due to drink and how much of it is due to driving at night. If folks with blood alcohol between 0.05 and 0.08 tend to be on the road late at night, and if more accidents happen late at night even for sober drivers, then just comparing overall accident rates without adjusting for these time of day effects will overstate the effect of alcohol.

Use the above to put a cost on excess accidents involving folks with increased levels of blood alcohol.

Next, we need to find the elasticity of those costs with respect to potential law changes. Because, of course, we don't know for sure how drivers will respond to the change in legislation. I would expect not much movement in the part of the curve above 0.1 and a reasonable drop in the part of the curve from .085 or so downwards, and a reasonable drop below the 0.05 level as risk averse people further reduce consumption to avoid the now lower limit. It's always annoyed me that the anti-alcohol folks on the radio harp on and on about how you can drink way more than most folks' rule-of-thumb drinking level and stay below 0.08; they figure that's evidence of how folks underestimate just how drunk people can be and still drive. I chalk it up instead to risk aversion. The rule of thumb of a couple of drinks followed by a drink an hour will keep you under 0.08 with a good padding for odd stuff like a drink being poured heavier than you'd expected, dinner having been lighter than you'd have liked, or just random variables that affect your alcohol uptake. It's like inflation targeting: you ought to target the middle of the band to avoid going over the upper bound rather than targeting the upper bound (shouting at RBNZ of 2005...).

So, how can we estimate that elasticity? Seamus suggests one mechanism: allow local authorities to set a different limit for their local jurisdiction, then just track things as per above. Another option would be to try and find comparable jurisdictions that kept similar data around the time of an alcohol limit change. The strong disadvantage of the latter approach is that those changes tend to be accompanied by stepped-up enforcement, which itself can affect drink driving rates independently of the limit. MacDoctor notes this problem as well as problems with the relative risk curves used in this kind of analysis.

Next, use the elasticity of costs with respect to legislative changes to estimate the reduction in costs with the law change.

But finally we have the really important bit - the part that the anti-alcohol campaigners always manage to leave out. We need to estimate the lost consumer surplus with the reduced drink driving limit. Lots of folks enjoy sharing a bottle of wine with dinner. A lot of them would decide to give wine a pass entirely, or to have less of it, with the rule change. Since they chose previously to consume it, they must value it. And the constraint necessarily then destroys some of that enjoyment. How do we estimate reduced consumer surplus of this sort? I'd start by looking at what happens to the restaurant and bar sector: by how much is their traffic reduced? I'd also want to look closely at data on on-licence and off-licence alcohol sales. If there's no change from trend in folks' restaurant and bar purchases but just a big increase in taxicab use, then the cost would just be the difference in cost between driving and cabbing multiplied by the increase in taxi use. If we see a drop in restaurant and bar activity, we could back out a measure of lost consumer surplus from prior estimates of price elasticity of demand for both of those, and add in the change in taxi usage as well. If we see a drop in restaurant and bar activity combined with an increase in off-licence sales, we'd need to net out the increased surplus from drinking at home from the reduced surplus from nights out. Get a total value for the surplus forgone as consequence of the limit reduction.

The policy change is ONLY worthwhile if the cost reduction from accidents forgone exceeds the reduction in surplus accruing to social drinkers whose nights out are less fun or more costly than before. And, the reduction in the drink driving limit also has to be the most efficient way of achieving those cost reductions. The government is putting in place a lot of other policies that will do a fair bit to reduce drink driving costs while not unduly burdening social drinkers over the age of 20 (but while sticking it to those lousy kids...shaking fist and zimmer frame at them...). Ignition interlocks for repeat drink drivers, zero alcohol limits while driving for folks who've demonstrated that they have a hard time saying no once they've started: these are the kinds of targeted policies that ought always be the first cut at solving a problem. I'd want to start collecting the above data after folks have had time to adjust to the more targeted measures.

I'm not sure what the Minister was reckoning would be adequate data collection. But this is how I'd start going about things if I were the guy at Treasury or Transport charged with handling things. I especially hope that they put weight on losses the policy would impose on social drinkers.

If the method above showed that the reduction in benefits to social drinkers were less than the drink driving costs avoided by reducing the drink driving limit, and if it achieved those cost reductions efficiently relative to other policies, I'd sign on for the flip to 0.05. I'd be surprised if that were the case; I'd put maybe a 20% chance that the study would come out showing the reduction to be efficient. And so I'm against it. But I'm happy to update based on evidence.

Oh: it's also interesting to note that the Cabinet insider who did very well on iPredict trading on potential minimum wage changes didn't trade on this one: a reduction in the drink driving limit was trading at iPredict at over $0.80 prior to Joyce's announcement, at which point it plunged to zero. I was moderately short at the time of announcement and so am pleased; I'd have been even more pleased had I been logged in and trading when Joyce made his announcement!


  1. My biggest problem with many of those who argue for lowering the limit is that they use examples of horrific incidents caused by drivers who were well over the limit as it is currently set. If a blood alcohol limit of 0.08 didn't deter these folk from driving drunk, lowering it to 0.05 will similarly have no effect.

  2. @Lats: The other side would argue that lowering the limit makes it less likely that folks would hit the really high drink driving levels. Empirical question. I'm doubtful it has that big of an effect, but it's possible.

  3. That's brilliant, looks good to me. What I'd like to know is what are the odds on reputable government agency actually doing the research as suggested. Is there anyway we could frame an iPredict stock on the research being performed? Perhaps use a purpose clause and give admin leeway to make a judgment call? If so we could do another one on research for the gender wage gap.

  4. It would be a big job. I'd guess they'd want to outsource it. Under no circumstances should anyone other than Treasury or MED be allowed to set the terms of the contract and to evaluate the results. And it ought to be peer reviewed. Folks outside of Treasury / MED aren't competent to evaluate the difference between a standard economic method and ... less standard methods; worse, they may prefer the non-standard methods.

    Very slim chance that anybody would follow my above proposal. Slight chance they'd go for a better design; reasonable chance they go for a weaker design. 10% chance a method's used that matches or betters the one listed above. Mostly because it would be a big job, but partially because a lot of folks just don't want to take seriously the welfare losses to moderate drinkers.