Monday 19 August 2013

Longer opening hours, fewer road accidents

This is a welcome unintended consequence of the liberalisation of licensing laws, if true. From The Times (also reported in the Daily Mail)...

Late-night drink laws saved many young lives

Hundreds of young people have escaped death on Britain’s roads after laws were relaxed to allow pubs to open late into the night, a study has found.

The number of crashes reported to police fell by 13 per cent, or 1,643 a month, after licensing laws were reformed in 2005, according to the first such analysis of official data.

The decline has been felt most sharply among drivers aged between 18 and 25 and has been most marked on Friday and Saturday nights. Accidents involving young drivers at those times have fallen by 33 per cent. 

There are plausible reasons why extending opening hours might reduce road accidents but there are also plausible reasons why it might increase road accidents. The authors discuss them at the start of the study (PDF).

First, extended hours may influence the intensity of drinking (the number of impaired bar patrons). Thus, if patrons spend longer at the bar and drink the same total amount as before the longer hours, the intensity could decline as they no longer drink to beat the clock.

Alternatively, the longer period could allow a larger share of drinkers to become impaired and so intensity could increase. The second influence works through the amount and timing of driving among the impaired patrons. As we’ve suggested, the later closing hour may reduce the driving to illegal pubs or home parties as patrons stay in the pub.

Moreover, if the bar no longer goes instantly from full to empty at the early closing hour, there may be fewer impaired drivers simultaneously on the road, even if the total number of impaired drivers is the same. In addition, the impaired drivers may now drive later in the night when fewer other drivers are on the road. While these influences may suggest fewer accidents, it could be the case that those who leave the bar very late have less access to public transit and are more likely to drive.

In the end, we find the arguments each way at least plausible and view the direction of the influence as an empirical issue.

To get to the empirical meat, they use data from the Department of the Environment, Transport and the Regions "that contain all motor vehicle accidents reported to the police from 2002 to 2008 for all 416 local jurisdictions in the three parts of Britain."

We know the type of accident (whether it caused either serious injury or a death), the date and time of the accident, location of the accident and the age of the driver of any vehicle involved in the accident. Critically, the data allow us to accurately assign accidents to the pre and post policy period.

The results are striking and there are several reasons to think that the correlation with the Licensing Act implies causation. Firstly, there is a sharp reduction in road accidents in the months immediately following the start of so-called '24 hour drinking'. Secondly, the rate of accidents remains significantly lower than the pre-liberalisation period thereafter. Thirdly, the data from Scotland—used as a control—doesn't follow the same pattern at all. On the contrary, Scotland saw rates falling before the enactment of the legislation (which affected England and Wales, but not Scotland) and then flattening out. This suggests that there is no unidentified third variable that reduced road accidents in the UK as a whole.

The authors conclude that "the policy is associated with a 9.8% decrease in the number of accidents per jurisdiction." The results are statistically significant and remain robust despite the researchers throwing numerous tests at them. Indeed, the results become stronger as they dig further into the data. It becomes clear that the decline in road accidents is concentrated late at night, at weekends and amongst young people.

When we look at night hours the decrease is slightly higher at 14%. These night hours can be further divided into those on the weekends and during the weekdays. The weekday decrease is 12% as shown in row 3 while the Friday and Saturday decrease is 18% as shown in row 4.

... We next examine the hours of 9:00 at night to 3:00 in the morning on Friday and Saturday nights. As shown in the fourth row of Table 3, the estimated decline remains statistically significant and represents a 23% decline. Similarly, an even narrower window of 11:00 at night to 3:00 in the morning on Friday and Saturday nights shows a nearly 24% decline in accidents. Thus, at the very hours when one might anticipate patrons would otherwise be simultaneously streaming out of bars and into cars, the new legislation allows them to remain in the bar and spreads out their dispersal with a large apparent decline in traffic accidents.

... The estimates reveal that the decrease in accidents associated with longer hours is concentrated almost entirely among young drivers... As a percentage decrease, a drop of .380 represents a 32.5% decrease for younger drivers on Friday and Saturday nights. While the coefficient is negative for middle age drivers, it is not statistically different from zero. The coefficient for older drivers is not significant, vanishingly small and actually positive. Thus, the clear pattern is that the age group most influenced by the policy change is the youngest drivers.

It's interesting to compare this study with the crude ecological studies that I have often written about in the past (such as the English heart miracle, the Canadian minimum pricing scam and, most recently, the Colorado casino miracle). These studies use aggregate data such as hospital admissions and attribute the reported decline to legislation (invariably, legislation that the researcher supports). One problem with this approach is that there are all sorts of other variables that could confound the results. Another problem is that some outcomes (such as heart attacks) are in long-term decline and would have fallen with or without the legislation.

In the case of road accidents in the UK, there is no doubt that they have been in decline for decades, albeit with significant fluctuations (they have risen recently, for example, and they were rising immediately before the Licensing Act came into force). The authors acknowledge this and their conclusions are based on a clear and sharp drop that coincided with the liberalisation of the licensing laws.

But aside from taking the same basic approach, this study has nothing in common with the policy-based ecological studies that I have criticised in the past. The real problem with those studies is that they either cherry-pick a small sample of the population while ignoring the national data (eg. here, here and here) or that they are based on a computer model that bears no relationship to the actual data (eg. here, here and here). Sometimes they even pretend the legislation started when it didn't just to make the facts fit (eg. here and here). Oh, and they hardly ever use a control group.

This study doesn't pull any of those tricks. It examines all road accidents in England and Wales over a seven year period and compares them to a control group (Scotland) that is directly comparable. It shows a clear decline in rates directly following the enactment of the legislation and it shows that this decline was followed by a consistently lower rate thereafter.

Trends in England and Wales (top) compared with trend in Scotland (bottom)

They show all this using real—not modelled—data and they compare the results in England and Wales with real data from another real country, Scotland, not with a hypothetical model of what might have happened in the absence of legislation.
Moreover, the lead author is an economist who has written about a broad range of issues, not a single-issue campaigner with a known bias. Indeed, he has previously authored a study that reported a negative effect of the Licensing Act (ie. absenteeism). He kicks the wheels pretty hard with a series of tests to make sure the findings are robust. It could, of course, still be a coincidence, but the evidence that extending drinking hours in England and Wales had a positive effect on road accidents looks strong.


Colonel Shotover said...

One might also suggest that one is more likely to take a taxi when intending to come home at 2am rather than 11pm.

Christopher Snowdon said...

Someone on Twitter made the same point, with a graph.

nisakiman said...

What I find most interesting is that here we have what actually appears to be a positive outcome as a result of relaxing legislation. One wonders how much more could be achieved if we threw out all the self-righteous lobbyists and the restrictive laws they've engendered. Life in the UK might actually start to be enjoyable again! :)

James Burr said...

"Life in the UK might actually start to be enjoyable again! :)"

And, as you say, safer.

Smoking rates were in steady decline until the ban came in. And I can only wonder what not saying how much tar is in a fag or what having loads of untested fire retardant chemicals in a fag will do.

Or indeed, banning packs of 10 (there have been times in the past when I had given up for months when I just had a craving. A couple of fags would have done the trick and then I would have binned the pack and given up again. But I wasn't going to chuck 18 fags! Especially at £8.00 a pack (another Tobacco Control brain wave)). Indeed, I can't think of a single TC measure that hasn't resulted in HARM rather than encouraging health. It's telling that the only time I gave up for a long time (18 months) was when I lived in Spain and fags were 80p a pack and at the time you could smoke anywhere. I knew I was giving up because I wanted to, not because I was being hectored into it.

It's also telling that I haven't wanted to give up since 2007. ASH - promoting ill-health one step at a time.

BrianB said...

Sorry Chris, but I am not convinced that this shows anything more conclusive than the 'miracle' studies of the anti-tobacco zealots.

Yes, there was a "significant" fall in the trend between two 42-month periods, but so what? Looking at the trend in figure 1, it seems to have been falling more steeply before the legislation than after. It doesn't help having only 5-point moving averages, when there is clearly a huge seasonal (winter/summer) effect going on - 13-month moving averages would have been more useful.

I do take your point about the large drop immediately after the introduction of the legislation, but it follows an apparently out-of-trend increase over the preceding period. The end result is that, over the whole 7 years the trend seems remarkable consistent, and any abrupt changes are merely down to whatever was going on in late 2004/early 2005. I would have been happier if they had shown the prevailing trend for a longer period before the legislation too, at least to see whether the prevailing trend was a long-term one. If it was, then nothing appears to have changed. If it wasn't, then where is the analysis of potential confounders (ie other possible reasons for the change)? There isn't one.

I am also quite deeply unhappy that they have used a 'fudge factor' of value 1 as a proxy for 'measuring' the existence of the legislation. This trick has recently been used by Glantz and copied by one of the UK anti-tobacco wimmin (I can't remember which one - Bauld, I think). All they are really doing is setting the value of 1 for any period after a specific date - it is no way a measurement of the efect of the legislation per se and, as such, cannot be used to prove any relationship, other than coincidence over time, which, as I suggested before, looks like it was happening anyway.

All in all, I see no methodological difference between this study and the Pell/Glantz/Gilmore garbage. Both groups can claim to have achieved statistically significant results, but that is not surprising - there is always a mathematical comparison technique somewhere that will show significance. It just needs to be searched for in the software packages used by these people.

Anyway, why should this be seen as any different to the 'miracle' studies? "Study proves that state legislation is effective" is hardly a rallying cry to admire, and equally applies to both types of study.

In my (low) opinion, economists are one of the three categories who should not be allowed anywhere near statistical methodologies - the others being epidemiologists and climate scientists (sic). They are all charlatans, and, equally, are usually wrong.