Thursday 20 December 2012

Finding the denominator

We simply cannot tell whether changing the drink driving limit to 0.05 from 0.08 is a good idea unless we have some idea what proportion of drivers on the road at different times of day are in that range. We can talk a lot about the numerator - how many people involved in accidents have had something to drink. But without the denominator, we cannot make sense of the numerator. 20 accidents could be a huge or a tiny proportion of drivers on the road who have similar blood alcohol readings.

Most likely, drivers in the 0.05 to 0.08 range would be over-represented in the accident stats after correcting for time-of-day and day-of-week effects. But that doesn't tell us whether reducing the drink driving limit would be a good idea. Rather, we have to weigh up the likely reduction in consumption in that range given a change in the law, and the likely consequent reductions in accidents in that range, against forgone consumer surplus among those who would have otherwise been driving in the 0.05-0.08 range without adverse incident. I don't know which way this would turn out, which always makes things more fun.

The New Zealand Police told me that it's impossible to have any automated gathering of that data - their machines are not set up to keep those running tallies. And it didn't look like anybody else was collecting the data. So I tagged along with an alcohol checkstop unit on Saturday night to scope out what would be needed for a survey - how many research assistants would need to tag along to have a decent chance of getting accurate data while still staying out of the way. Over about an hour on Victoria Street, 173 drivers returned no alcohol; 13 were in the 0-250 mcg range (under 0.05); 2 were in the 250-400 range (0.05-0.08); 2 were sent on for evidential readings as they seemed to be above 400 mcg. The officers noted it seemed to be a pretty quiet night - traffic was light and few drivers were in the 250-400 range as compared to other nights. And, surprisingly, one of the officers reported that she was sure that someone else had been collecting data of this sort.

I subsequently heard back from the right person at the Ministry of Transport that they have run a short survey assessing the proportion of drivers on the road in the various BAC ranges, or at least tallying the numbers registering 0, 0-250, and 250-400. And, even better, they've promised me the raw data mid-January. It's limited in that they were only looking at Friday and Saturday nights, but so long as I can sort out the proportion of accidents that happen at the same times, I can deal with it.

I'm very glad I'm not going to have to re-create the wheel on this one. The data collection looked like it wasn't going to be the easiest thing, and I was starting to have nightmares about just what the University administration paperwork was going to be needed to get permission to send RAs out with the Police, even were I to have found external funding.

Many thanks to the Officers who let me tag along for the night, the Canterbury Road Policing Manager, and Mike McCosker for getting me in touch with the right people to get things going. I think it's going to be pretty useful knowing how this data gets generated when I go to play with it in the early new year.

4 comments:

  1. Fantastic - this is exactly what we need to have a sensible decision that doesn't rely on some public health types asserting "people are drinking, and some of them have accidents, clearly the problem is the drinking."

    I think there are a few possible answers that could come out:
    1. There are no people driving in the 0.05-0.08 range. Unlikely



    2. There are people driving in the 0.05-0.08 range, and their proportion of accidents is less than their overall proportion of the drivers. In this case, we should force everyone to drink 0.05-0.08 before they drive. :-)

    3. There are people driving in the 0.05-0.08 range, and their proportion of accidents is roughly the same as their overall proportion of the drivers. In this case, we probably should do nothing about the law.

    4. There are people driving in the 0.05-0.08 range, and their proportion of accidents
    is higher than their overall proportion of the drivers, but the increased accidents (assume 1-2 per annum) is lower than the benefits that the drivers (assume many hundreds of thousands) get from being able to have one or two drinks (including their health benefits). In this case, we probably should do nothing about the law, but it's hard to explain to many people (would it be rude to describe them as stupid people?)

    5. There are people driving in the 0.05-0.08 range, and their proportion of accidents is higher than their overall proportion of the drivers, and the increased accidents (assume more than 2 per annum) is higher than the benefits that the drivers (assume many hundreds of thousands) get from being able to have one or two drinks (including their health benefits). In this case, we probably should change the law to reduce the limit.

    Other things to watch out for:
    - selection bias. Do the police select for "random breath tests" those who are more likely to have been drinking? (i.e. does everyone get tested, or some subset that the Police select)

    - is there an upstream impact - does pushing down the limit 0.05-0.08 also drag down those drinking at higher levels? And if so, is it morally reasonable to change the rules for those who are doing nothing wrong so as to impact those who are already breaking the law?

    ReplyDelete
  2. I'm also expecting that #4 is what obtains. But who knows. That's what makes it fun.


    The night I spent out with the police was really instructive on how that data gets collected; I'd have been a fool not to do it even had I known that MoT was already gathering the data. My current worries:
    1) The first-round breath screening devices do not seem all that sensitive for low-level drinking. Many drivers admitting to have had something to drink showed up on those initial screens as having had no alcohol (presence of alcohol on the initial screen has you blow in the tube for the finer reading of 0, 0-250, 250-400, >400). Most of those drivers would have been in the 0-250 mcg range, but maybe some were in the 250-400 range. That would have me underestimating the number of drivers affected.
    2) MoT's data gathering was on Friday and Saturday nights - relatively high drinking times. But so long as I restrict things so the range of accidents (day/time) matches the days of data collection, I should be ok.
    3) Agreed on upstream impact, but I also worry about downstream impact. Suppose that I target 240 trying to avoid accidentally hitting 400. What do I target to avoid hitting 250? There will have to be a reasonable consumption drop among those in the 0-250 range, but it's not going to be easy to reckon how much.

    ReplyDelete
  3. Downstream impact is interesting. I agree with you, that's a negative impact on people who were law abiding even under the mooted new law.

    On selection, my concern is that at high volume times the police in some places appear to pull over only a subset of the vehicles. What isn't clear to me is whether this is genuinely random - so they do all the cars in their queue, then they stop the next 20 cars irrespective of who they are, or whether they pull out specific cars based on "intuition" as to who looks dodgy. If the latter, then if we presume that police intuition has some sort of basis in fact, they are more likely to be pulling over cars of people who have been drinking or are otherwise dodgy.

    ReplyDelete
  4. When I was out, the only non random bits were that taxis were waived through and everyone parking or turning quickly on seeing the checkstop were hit. Otherwise it seemed based on stuff orthogonal to drinking.

    ReplyDelete