The CycleChat Helmet Debate Thread

[Justinslow]

Where do I start?

Risk doesn't mean the rate of accidents, it's a combination of the rate and the severity. As the URL indicates, the word "risk" appears large in the reporting of the study (press release probably prepared by the authors' institution). The review isn't about the odds of a head injury given injury, it's about the odds of a head injury given admission to hospital. Mathematically and in other ways it's plausible that helmet wearers are more likely to turn up in hospital, but once there are less likely to have a particular kind of injury. The evidence on the question of whether helmets reduce the odds of a head injury is still contentious.

Based on that series of errors I'm not going to click on the link.

Why's that then? I've clicked on all the links showing "evidence in Australia" etc, you frightened something won't fit with your opinion?

 

[swansonj]

Where do I start?

Risk doesn't mean the rate of accidents, it's a combination of the rate and the severity. As the URL indicates, the word "risk" appears large in the reporting of the study (press release probably prepared by the authors' institution). The review isn't about the odds of a head injury given injury, it's about the odds of a head injury given admission to hospital. Mathematically and in other ways it's plausible that helmet wearers are more likely to turn up in hospital, but once there are less likely to have a particular kind of injury. The evidence on the question of whether helmets reduce the odds of a head injury is still contentious.

Based on that series of errors I'm not going to click on the link.

Just one teensie weensie correction to one aspect of that - in epidemiology, "risk" does in fact mean probability or rate of occurrence without reference to severity.

 

[swansonj]

Why's that then? I've clicked on all the links showing "evidence in Australia" etc, you frightened something won't fit with your opinion?

I do want to read the full paper. But I'm b'd if I'm going to pay thirty quid for the privilege, so I'm waiting till I can get to a library where I can download it for free. A classic example of the iniquity of the conventional model of scientific publishing (which @Flying_Monkey have discussed before) - the authors can put out a press release and generate the alarm and concern, but no-one can check the data for themselves without contributing to the coffers of a publishing house, albeit in this case an academic one rather than a purely commercial one.

 

[srw]

Why's that then? I've clicked on all the links showing "evidence in Australia" etc, you frightened something won't fit with your opinion?

If you read an extract of something and you can find half a dozen elementary errors in it will you bother reading the rest?

 

[srw]

Just one teensie weensie correction to one aspect of that - in epidemiology, "risk" does in fact mean probability or rate of occurrence without reference to severity.

Which, in "normal" epidemiology, probably makes sense - the discipline is classically applied to disease, where one case of malaria or cancer or flu is much like the next. But in this instance it's actively unhelpful. The impact on the owner of the skull of a (vanishingly rare) blow is rather different depending on whether it's a light tap or a massive lorry-sized hammer-blow.

 

[swansonj]

Which, in "normal" epidemiology, probably makes sense - the discipline is classically applied to disease, where one case of malaria or cancer or flu is much like the next. But in this instance it's actively unhelpful. The impact on the owner of the skull of a (vanishingly rare) blow is rather different depending on whether it's a light tap or a massive lorry-sized hammer-blow.

And epidemiology does have metrics that measure impact as well as incidence, though as you suggest, less often used than simple incidence or mortality. But it has different terms for those, and it uses the word risk for a pure probability, and it's therefore just a tad harsh to imply that the authors are at fault for using the term in exactly the correct way.

I think we're all agreed what we think the real problem with this paper is - that the quantity it measures is not particularly relevant to "risk" (in any sense) as experienced by cyclists and therefore to decisions about whether to wear a helmet or not, and quite probably has not much to do with the effectiveness of helmets at all. And that the conclusions of the paper, and even more so the press release, obfuscate these issues.

 

[srw]

And epidemiology does have metrics that measure impact as well as incidence, though as you suggest, less often used than simple incidence or mortality. But it has different terms for those, and it uses the word risk for a pure probability, and it's therefore just a tad harsh to imply that the authors are at fault for using the term in exactly the correct way.

But my comments about "risk", and the other call-outs, referred to the treehugger.com quote that GC produced, not to the paper that kicked off this discussion. Like you I'd love to read that paper, but my academic maths is probably too rusty to do it justice.

I think we're all agreed what we think the real problem with this paper is - that the quantity it measures is not particularly relevant to "risk" (in any sense) as experienced by cyclists and therefore to decisions about whether to wear a helmet or not, and quite probably has not much to do with the effectiveness of helmets at all. And that the conclusions of the paper, and even more so the press release, obfuscate these issues.

I suspect there's more to it than that. My memory is that among papers of its type only the original Rivara and co article ever claimed an incidence reduction of as much as 70%, and even Rivara and co reduced their incidence reduction on re-analysis. So how in the name of mathematics can a re-analysis of the entire universe of papers come up with an incidence reduction as high as 70%.

 

[swansonj]

...Like you I'd love to read that paper, but my academic maths is probably too rusty to do it justice....

You don't need much specialist maths to do epidemiology, trust me . In fact you need precious little specialist knowledge of anything but you do need clear logical analytical thought ... which you don't always find in published papers, sadly.

 

[winjim]

But my comments about "risk", and the other call-outs, referred to the treehugger.com quote that GC produced, not to the paper that kicked off this discussion. Like you I'd love to read that paper, but my academic maths is probably too rusty to do it justice.



I suspect there's more to it than that. My memory is that among papers of its type only the original Rivara and co article ever claimed an incidence reduction of as much as 70%, and even Rivara and co reduced their incidence reduction on re-analysis. So how in the name of mathematics can a re-analysis of the entire universe of papers come up with an incidence reduction as high as 70%.

Google the title of the paper and you will find a pdf of what appears to be a pre-publication presentation of Olivier and Creighton's analysis. It gives the studies used and odds ratios for each category of injury. For serious head injuries, the odds ratios range from 0.03 to 1.17, so an odds reduction of 97% to -17%. So one study found that helmet wearing increased the odds of serious head injury, although that one has a rather small sample size.

Notable by its absence is the odds ratio data for each study used in the analysis of fatal head injuries. I think this is probably because the meta-analysis is based on only two studies, which would explain the size of the confidence interval. So even though the Graun article calls the fatality data significant, to me it looks like the sample size is too small to extract any meaningful data. This in turn leads me to reason that the odds of sustaining a fatal head injury when cycling, helmeted or not, are really very small indeed.

 

[hatless]

Wrong.

This would only be true if you - and every cyclist - hit their heads in exactly the same angle, every single time. Except of course, this is untrue. You can fall off and hit any part of your head, at any angle. This is obvious, is it not? You need to stop thinking about just one instance, which (say) someone strikes their head at an angle of 58 degrees to the anterior and consider another accident, where the angle is 42 degrees, and the next, and the next. Take one thousand instances and you'll approach the full 360 degrees. You cannot predict which angle impact will happen. Hence the requirement to model impact probability with area. Which, again rather obviously, is why helmets are round(ish).

If someone threw pebbles towards me, the chance of one hitting my helmeted head would be proportional to the cross-sectional area visible to them. If I fall off my bike, the chances of my helmeted head hitting the ground is different. It depends on whether I manage to stop my head in time, perhaps by letting another bit of me absorb the impact. Making my head bigger will make this more difficult, and this will vary with the additional radius that the helmet gives to my head, though probably not in a simple ratio.

Think about a long pendulum swinging in a small room. The swings can be of varying amplitude. You can also increase the size of the bob. Whether it hits a wall will depend on the size of the swing and the radius of the bob.

A head on a springy neck attached to a falling body behaves rather like a pendulum. Land on your back, shoulder or outstretched arms and immediately after contact with the ground, your head will continue moving towards it. If the fall is gentle and you are fully conscious, you will probably be able to keep your head from hitting the ground hard. If your fall is faster, your head may have too much momentum for your neck muscles to be able to stop it before it hits the Tarmac. In that case, how hard it hits the Tarmac depends on how soon it hits the Tarmac, in other words, on how thick your helmet is.

 

[Wobblers]

If someone threw pebbles towards me, the chance of one hitting my helmeted head would be proportional to the cross-sectional area visible to them. If I fall off my bike, the chances of my helmeted head hitting the ground is different. It depends on whether I manage to stop my head in time, perhaps by letting another bit of me absorb the impact. Making my head bigger will make this more difficult, and this will vary with the additional radius that the helmet gives to my head, though probably not in a simple ratio.

Think about a long pendulum swinging in a small room. The swings can be of varying amplitude. You can also increase the size of the bob. Whether it hits a wall will depend on the size of the swing and the radius of the bob.

A head on a springy neck attached to a falling body behaves rather like a pendulum. Land on your back, shoulder or outstretched arms and immediately after contact with the ground, your head will continue moving towards it. If the fall is gentle and you are fully conscious, you will probably be able to keep your head from hitting the ground hard. If your fall is faster, your head may have too much momentum for your neck muscles to be able to stop it before it hits the Tarmac. In that case, how hard it hits the Tarmac depends on how soon it hits the Tarmac, in other words, on how thick your helmet is.

Sigh. You're still not thinking in three dimensions.

Someone's head does not behave like a pendulum, at all, in a collision. The impact forces involved are magnitudes of order greater than the restoring force (muscle tone in the neck muscles). Any model involving simple harmonic motion is therefore incorrect. You have also neglected the possibility that someone may hit the ground head first. But even were we to assume for a moment that an inverted pendulum is a reasonable model, it predicts that impacts will happen along a line - close to a circle in fact. A very brief examination of the many photos of damaged impacts should swiftly disdabuse you of that notion - impact damage is distributed along the entire surface of the helmet. Your model cannot account for the physical evidence, therefore it must be rejected as being incorrect.

It is instructive to consider why what you've suggested does not accord with reality. Your example assumes that the room is the same size in all impacts therefore the impact occurs at the same point along the curve. Project that through all possible impact directions and you get a circle. But you've assumed that the impact geometry is identical for all given impacts. This is incorrect. An impact can occur randomly at any point along the curve swept out by the pendulum. Plot all those arcs out in all directions and you will get a hemisphere. What is the area of a hemisphere?

 

[jefmcg]

Because wearing helmets has become normalised behaviour in the state, they do not have any negative effect on the growth of riding.

On what basis do they say this? I'm in Melbourne at the moment and shocked at how few cyclists there are. I've gone down to the shops on a lovely sunny day, and have literally seen no other bikes either ridden or locked up but the car parks are full. And I was by the beach yesterday between 4 & 6 pm, at what is a major commuter route for driving and a nice low traffic route for cyclists and there were really very few cyclists, though the road was at a standstill with barely moving bumper-to-bumper cars. Yesterday was not a nice day, with intermittent rain, but to a regularly cycle commuter, not really any bother.

And I find it telling that Melburnians think cycling is much more dangerous than Londoners do (most of my friends live in either Melbourne or London). Melburnians always ask me if I wear a helmet, and look worried. Londoners tend to say, "that's a long way" with no concern for my head.

(of course, it maybe Londoners are just heartless)

 

[david k]

I was in London last week I counted the first 25 riders, 24 wore helmets. Not large scale research I know, but strikes me wearing helmets hasn't had a negative effect on cycling numbers.

There is more than just helmets that affect the amount of cyclists

 

[benb]

I was in London last week I counted the first 25 riders, 24 wore helmets. Not large scale research I know, but strikes me wearing helmets hasn't had a negative effect on cycling numbers.

There is more than just helmets that affect the amount of cyclists

If you saw a majority of helmeted cyclists, then surely "wearing helmets hasn't had a negative effect on cycling numbers" is precisely what that observation doesn't show.

 

[EnPassant]

I was in London last week I counted the first 25 riders, 24 wore helmets. Not large scale research I know, but strikes me wearing helmets hasn't had a negative effect on cycling numbers.

I was trying to cross the road in Gloucester today, I had to wait for 5 minutes whilst a long stream of cars went by. Not large scale research I know, but it strikes me that the rising cost of petrol hasn't had a negative effect on car numbers.*

The trouble is, in isolation this tells us next to nothing about numbers as they stand, and absolutely nothing about where they might have been if you changed factor X, Y or Z.
You need something like the Australian data mentioned more than once already in this thread. Statistics is a complicated business and proving causality more slippery than a bucket of eels covered in olive oil in situations like this.

There is more than just helmets that affect the amount of cyclists

Sure. But you have to use rigorous statistical analysis to know exactly what affects which things. And even then the experts still get it wrong.

*This is not meant sarcastically nor unkindly, although I read it back and it reads somewhat so I'm afraid. I can't seem to find a more succinct way of doing it without losing the point in 15 paragraphs of TL DR.