Seat belts in cars save the lives of occupants in cars, but do they lead to lower deaths for all road users?

Question

Reading and contributing to the skeptics debate about the effects of cycle helmets reminded me of the much more significant debate on seat belts in cars.

Most countries seem to have accepted that they should be compulsory. The UK government claimed they have saved more than 60,000 lives on the 25th anniversary of the law. But there have always been skeptics. John Adams (one of the originators of the idea of risk compensation) has made strong statistical arguments (full content here) that the benefit cannot be clearly seen in any country's accident statistics.

Clearly, as with many safety issues, the consensus is against him. And it seems obvious to many that seat belts must be good. But his argument on risk compensation:

Because seat belts are undeniably effective at reducing death and injury in crashes there is, or was, a mystery. Why in country after country that mandated seat belts was it impossible to see the promised reduction in road accident fatalities? The most plausible explanation is “risk compensation”. It appears that measures that protect drivers from the consequences of bad driving encourage bad driving. The principal effect of seat belt legislation has been a shift in the burden of risk from those already best protected in cars, to the most vulnerable, pedestrians and cyclists, outside cars.

suggests we need to be really careful which statistics we use. If one effect of seat belts is to increase risky driving we might see more accidents and a shift of the deaths to other road users. These effects may not be based on conscious or rational decisions. The question is can we see them in the statistics? The logic of seat belt compulsion is fine, but is that logic backed by experimental evidence?

So what do other skeptics think? What other evidence can we bring to the debate?

Answer 1

The evidence with respect to seat belts suggests that the law had no effect on total fatalities but was associated with a redistribution of danger from car occupants to pedestrians and cyclists.

Which evidence?

"In Britain, as in Australia, and as in the eight European countries examined by Isles, in the year that the wearing of seat belts became compulsory the numbers of pedestrians and cyclists killed increased, by 8% and 15% respectively. The numbers of pedestrians and cyclists killed by heavy goods vehicles and public service vehicles (categories not covered by the law) decreased following the law."

Which Isles?

Isles, J.E. 1981. Seat belt savings: implications of European Statistics. London, UK Department of Transport. This report was suppressed, i.e. not published, and not available to Parliament when it debated the seat belt law. A leaked copy was reported four years later in New Scientist (7 February 1985). The contents of the report and its conclusions have never been denied. The study is available here: http://john-adams.co.uk/wp-content/uploads/2007/01/isles%20report.pdf

This actually already contains some interesting comments:

the variation in trend between countries is greater than a 'seat belt law effect' ... Any beneficial effect could be concealed by rejecting certain countries So basically, by choosing the right countries, any effect could be proven, and the discussion which countries have to be chosen is not very strict: what if a law is not enforced? Or partially enforced (only some users, eg with recent cars, only some areas, eg outside the city centre).

This comment should also apply to Isles own study. Here is the graph that should support the fact that there is a jump in pedestrians killed the year the legislation is passed:

It shows the number of pedestrian casualties, scaled to 100 3 years prior to the law. This means that there is no confidence interval that year (everything scaled to 100). Interestingly enough, one country (Belgium) was excluded in the rest of the analysis of Isles, because their model did not fit well (!)*, this already means that all other confidence intervals are underestimations of the true uncertainty. But even disregarding these facts, it is clear from the remaining confidence intervals that their width is a lot bigger in the 'law year'. which makes the above comments more relevant again: if one county is included/excluded, the effect prior to the law would be small, but after the law would be big. The reason that the interval suddenly gets large is due to 2 things: the fact that scaling was done 3 years before (and autocorrelation is becoming less important), but also the fact that car usage was seriously changed in some in the same years due to the oil crisis.

I also don't think the presented model fits the data nicely after the law year. The large jump in the 'exponential with jump' line seems too large in the law year, but it is that effect which is used to prove that there is a 8% increase.

(* Being Belgian, I can imagine that they somehow changed the way that statistics were gathered and that that caused the strange numbers. If however no reason is found it would be plainly wrong to exclude the data because they don't fit an exponential trend seen in other countries.)

---

Apart from that, and that is the core of his message:

The claims made for seat belt laws in all these other jurisdictions rested on a deduction which assumed no risk compensation effect.

It is true that this assumption was not proved. But the reverse, that the risk compensation effect is not negligible is also not proven. Even if a correlation would be present proving that more fatalities occur after introduction of the law, it still does not prove a causal link.

---

So let's look at articles that cite him:

A study of driver behavior before and after a mandatory seat belt use law in Newfoundland found that the benefits of such legislation are not reduced by riskier driving, as has been suggested by some theorists.

http://onlinelibrary.wiley.com/doi/10.1111/j.1539-6924.1984.tb00130.x/abstract

A time-series analysis of Japanese data for the period 1966-85 estimates the effectiveness of seat belt usage and government safety policies on various types of traffic-related fatalities. The results indicate that these policies reduce the risk of death for both motor vehicle occupants and non-occupants.

The efficacy of government safety policies on traffic-related fatalities: Empirical estimates from Japan McCornac, DC Applied Economics [APPL. ECON.]. Vol. 25, no. 3, pp. 409-412. 1993.

Answer 2

Overall fatalities have been steadily decreasing:

Source: Image from Wikipedia based on data from US NTSB

This clearly contradicts "risk compensation" theory, with which you'd expect a flat line.

---

Now, as for the quoted article, there quite a few problems:

• causality: article suggests that in case of Denmark introduction of belt laws caused increase in fatalities. If you look at plot he provides to illustrate that, seems that belt laws were introduced after the sharp increase started:
  
  Image source: Jonh Adams "Britain’s Seat Belt Law should be Repealed"

• detection rate vs actual rate: article suggest, that huge increase in number of drunk drivers in UK is consequence of belt laws. But the fact is, that increase coincides with the dates when evidential breath testing was introduced in UK, thus it's more than likely that detection rate went up.

• doesn't consider increasing number of cars: the statistics above deal only with absolute numbers, failing to provide comparison with increasing number of cars and growing population.

• it's based on idea that has already been debunked: namely Smeed's law.