Have London's roads become more dangerous for cyclists? Statistical study says something has changed on London’s roads

It’s not by chance that six cyclists were killed in the capital in 14 days in November, say statisticians

by Sarah Barth   January 4, 2014  

Ghost bike for Brian Dorling at Bow Roundabout (©John Stevenson:road.cc)

‘Black November’ of 2013, when six London cyclists were killed in the space of 14 days, was a sign of increased danger on the capital’s roads, a new study has found.

Published in Significance, the Journal of the Royal Statistics Society, the statisticians Jody Aberdein and David Spiegelhalter looked at the deaths of six cyclists who were killed while cycling in London.

Their piece poses the question: “as an urban cyclist, should you hang up your hat and fluoro? Should you forsake two wheels for four, or more, in order to preserve life and limb?

“Many will have started their journeys to work this month with more trepidation that they did in November.”

They then attempt to answer the question by considering how unusual six sequential deaths might be.

As they note: ‘Randomness is clumpy, and humans are excellent at pattern recognition.

“The issue with rare events which are subject to chance is that it is very hard as an observer to know whether there is an underlying pattern.

“We have, for a variety of reasons, a great propensity to attribute cause to chance occurrence.

“This tendency is never more active than when there is an emotive and high-stakes outcome.”

However the pair go on to study data from road traffic accidents for the whole of the UK for the last several years, concluding that there are historically on average 0.6 cyclist deaths per fortnight over the period 2005 to 2012.

Even plotting the deaths out over calendar months, in 8 years the statisticians find only five fortnights with 3 deaths in them.

Aberdein and Spiegelhalter go on to conclude that “It turns out that over 8 years the chance of our seeing six deaths in a 14-day period is around 2.5%.

“The usual level of “significance” in such analyses is set, arbitrarily, at 5%. In other words, the chance of observing an occurrence at least as extreme under the baseline assumption is less than 5%.

“Hence six deaths is a significant finding.”

20 user comments

Oldest firstNewest firstBest rated

Never in the field of human commuting has so much been said to say so little.

FATBEGGARONABIKE's picture

posted by FATBEGGARONABIKE [555 posts]
4th January 2014 - 11:12

like this
Like (32)

I am glad they have undertaken such an in depth analysis. Anecdotally there is a particular type of vehicle involved driven in a particular fashion that is disproportionately responsible for the deaths and yes I am pointing the finger at tipper trucks driven by impatient, inconsiderate time focused drivers. Deal with this category and you will significantly alter the statistical outcome.
Sure there is an element of cyclist training (not going up the left side of lorries) but this will require a total overhaul of infrastructure design which predominantly sends you up the left side of all traffic at junctions so the novice naturally assumes this is what should be done.
What is upsetting is that politically no one feels responsible enough to take simple obvious actions to prevent further needless loss of life on our roads.

posted by arfa [438 posts]
4th January 2014 - 12:29

like this
Like (28)


Published Nov13.

Add to this chart a line showing the increase in the number of cyclists on London roads and you'll find the answer to the question posed in the headline is "No".

The Human Cyclist A blog. Try it, you might like it...

sm's picture

posted by sm [332 posts]
4th January 2014 - 13:14

like this
Like (27)

Break it down by vehicle type involved in KSI's and you will find a disproportionate statistical outlier.

posted by arfa [438 posts]
4th January 2014 - 13:42

like this
Like (17)

I did a quick poisson analysis a while ago, 6 deaths is indeed unlikely due to chance (as the article says, you expect clumpiness - if you had an average of one death a month, it would very rare to experience a year where exactly once cyclist died each month).

The paper seems, however, to assume death rate is constant throughout the year (can't tell this for sure, don't you just love academic paywallls? -but that's a rant for elsewhere). That seems a dodgy assumption - I'd be explicitly looking for a consistent spike just after the clocks have changed, for example - and the fact that overall deaths this year were within normal variance suggests this is not exceptional at all unless the rest of the year was exceptionally safe.

It was coincidence, probably in addition to a seasonal spike.

posted by nuclear coffee [105 posts]
4th January 2014 - 15:31

like this
Like (13)

Not even wrong. Although the question "Should you forsake two wheels for four, or more, in order to preserve life and limb?" does imply some dubious assumptions which would tend to indicate they're well on the way.

Saying something is statistically significant is of no use other than a bald, lonely statement of statistical likelihood. It says nothing about correlation with or causality from other factors - weather, time of accident, physical infrastructure, legal & criminal justice environment, rider training & behaviour, driver training & behaviour, changes from BST to GMT, and so on. There's a capital letter at the start of this sentence, but it says bugger all about grammatical structure of language other than that I start sentences with a capital letter. Less headline grabbing, more analysis plz.

posted by Argos74 [256 posts]
4th January 2014 - 16:46

like this
Like (13)

I think this must be a commentary on the misuse of statistical significance rather than on road safety for cyclists. Taking an exceptional event and then testing it for significance is simply wrong. I can't believe that an RSS publication would say that this was "a sign of increased danger on the capital’s roads".

Baffling.

thegibdog's picture

posted by thegibdog [70 posts]
4th January 2014 - 19:18

like this
Like (11)

More details on the statistical methods used are available here:

http://understandinguncertainty.org/when-cluster-real-cluster

I'm not sure why some commentators are railing against the findings, but I think that Professor Speigelhalter is adequately qualified to make the statements that he does - more so than the armchair statisticians that appear to have popped up.

posted by synoptic [8 posts]
4th January 2014 - 20:28

like this
Like (15)

synoptic wrote:
More details on the statistical methods used are available here:

http://understandinguncertainty.org/when-cluster-real-cluster

I'm not sure why some commentators are railing against the findings, but I think that Professor Speigelhalter is adequately qualified to make the statements that he does - more so than the armchair statisticians that appear to have popped up.

Even professional statisticians will acknowledge that the word 'significant' doesn't have the same meaning in statistics as it does in colloqual use.

So I'm sure the guy is entirely qualified to say what he says, but the question is, are _we_ qualified to know what he says actually _means_?

(And I refer you to the many studies that have claimed to find something 'significant', based on one thing correlating with another thing to a supposed 'signifance', which nevertheless turned out to be irreproducable (and probably down to chance after all). Seems to crop up particularly in medical and nutrition fields.)

posted by FluffyKittenofT... [618 posts]
5th January 2014 - 12:07

like this
Like (8)

synoptic wrote:
More details on the statistical methods used are available here:

http://understandinguncertainty.org/when-cluster-real-cluster

I'm not sure why some commentators are railing against the findings, but I think that Professor Speigelhalter is adequately qualified to make the statements that he does - more so than the armchair statisticians that appear to have popped up.


Thanks for posting this. I had misunderstood the "It turns out that over 8 years the chance of our seeing six deaths in a 14-day period is around 2.5%" statement in the road.cc article. The linked post says "the chance of getting at least 6 deaths in any 2-week window over 8 years is estimated to be 2.4%" which (to me anyway!) is quite different. I had thought the percentage chance related to the two weeks rather than the 8 years; it makes much more sense now!


I still don't see how the conclusion of this being a "sign of increased danger on the capital’s roads" comes about though, I am assuming this is a road.cc interpretation.

thegibdog's picture

posted by thegibdog [70 posts]
5th January 2014 - 12:38

like this
Like (9)

Typically with statistics it isn't really saying anything except that November was an unusual month. A very high mortality rate for
Cyclists, without saying why....and that is the crucial point.

I went to the casino the other day. I was betting on roulette. Whenever I put higher stakes bets on they won more than my low stakes bets (I made random and arbitrary bets). Statistically using the model above the high stakes bets look significant, but if you put it over the course of all my roulette bets then it looks like the significance of the bets is more reduced. Statistics can exceed in applying meaning to meaninglessness. In fact that is the problem with statistics, it is applied frequently to give meaning to events which have none. And even once it has done so there are arguments about the truth in the meaning. I'm sure mathematically the results are fine, but I would guess that the sample size the statisticians were working with and breath of data was too small to add any real quality of meaning. In fact their conclusion is 'significantly' vague.....quo bono?

posted by Colin Peyresourde [1059 posts]
5th January 2014 - 12:45

like this
Like (6)

Actually just read the article on which this is based and the reporting here is very misleading.

They actually conclude that cycling in London is safe and that the likelyhood of dying is quite small.

They themselves seem to say that while the events of November are significant they cannot say why and that more analysis is due. Which is pretty much what some said at the time.

posted by Colin Peyresourde [1059 posts]
5th January 2014 - 13:43

like this
Like (5)

A 5% chance is two standard deviations away from the mean occurrence and tends to be the point where data becomes "statistically significant" and 2.5% even more so (can't remember my bell curve analysis off the top of my head). Bottom line, it was very unusual and even more so if you narrow it by vehicle type involved.
What really irks me is that it is beyond obvious if you observe certain driving behaviours as alluded to above and it could be vastly improved at a pen stroke - change the terms for tipper tricks to day rates and not per load (most of them are crossrail which is a government contract). Insisting on cycle training for all the drivers before they can enter the low emissions zone might not be a bad idea as well.

posted by arfa [438 posts]
5th January 2014 - 14:40

like this
Like (8)

FluffyKittenofTindalos wrote:

Quote:
Even professional statisticians will acknowledge that the word 'significant' doesn't have the same meaning in statistics as it does in colloqual use.

So I'm sure the guy is entirely qualified to say what he says, but the question is, are _we_ qualified to know what he says actually _means_?

That's a fair point, although I would say that the onus is on the reporting journalist to provide the context of the information s/he presents - especially where the original paper was written for a specialist audience.

posted by synoptic [8 posts]
5th January 2014 - 18:01

like this
Like (6)

@nuclear-coffee w/r/t the paywall, yes terrible, open access forever, etc, but in this case the pdf is free, there's a link to it here http://onlinelibrary.wiley.com/doi/10.1111/j.1740-9713.2013.00715.x/abst...

posted by alsothings [19 posts]
6th January 2014 - 9:51

like this
Like (3)

arfa wrote:
A 5% chance is two standard deviations away from the mean occurrence and tends to be the point where data becomes "statistically significant" and 2.5% even more so (can't remember my bell curve analysis off the top of my head). Bottom line, it was very unusual and even more so if you narrow it by vehicle type involved.
What really irks me is that it is beyond obvious if you observe certain driving behaviours as alluded to above and it could be vastly improved at a pen stroke - change the terms for tipper tricks to day rates and not per load (most of them are crossrail which is a government contract). Insisting on cycle training for all the drivers before they can enter the low emissions zone might not be a bad idea as well.

You talk about tipper trucks, but in the cases of the 6 deaths I'm not sure that many were involved. In fact I think buses were more often at fault. I'm sure you have some statistics which prove your theory though....

posted by Colin Peyresourde [1059 posts]
7th January 2014 - 14:42

like this
Like (3)

The study is more complex than the report, as any listener to the great More or Less will have heard (http://downloads.bbc.co.uk/podcasts/radio4/moreorless/moreorless_2014010...) because you need to consider not just the consecutive 14 day periods but the non-consecutive ones in order to address whether these deaths were statistically significant. The conclusion: there isn't any significant increase, allowing for increased cycling in London, and that cycling remains life lengthening rather than reducing.

Of course, if we allow drivers to add Google Glass to their armoury ...

posted by adaminbristol [2 posts]
8th January 2014 - 10:36

like this
Like (2)

Colin Peyresourde wrote:
arfa wrote:
A 5% chance is two standard deviations away from the mean occurrence and tends to be the point where data becomes "statistically significant" and 2.5% even more so (can't remember my bell curve analysis off the top of my head). Bottom line, it was very unusual and even more so if you narrow it by vehicle type involved.
What really irks me is that it is beyond obvious if you observe certain driving behaviours as alluded to above and it could be vastly improved at a pen stroke - change the terms for tipper tricks to day rates and not per load (most of them are crossrail which is a government contract). Insisting on cycle training for all the drivers before they can enter the low emissions zone might not be a bad idea as well.

You talk about tipper trucks, but in the cases of the 6 deaths I'm not sure that many were involved. In fact I think buses were more often at fault. I'm sure you have some statistics which prove your theory though....

I should have been a little clearer that I was referring to longer term figures. Data available for you here
http://www.bbc.co.uk/news/uk-england-london-24999302

Bottom line, HGV's disproportionately over represented as a % of London cycling fatalities when looking at overall traffic share

posted by arfa [438 posts]
8th January 2014 - 13:00

like this
Like (3)

..

posted by arfa [438 posts]
8th January 2014 - 13:02

like this
Like (2)

It's all pretty simple. Cycling in London has increased and there are more cyclists amongst more dense traffic than anywhere else!

I really think that denying the obvious is not serving anyone!

'Have London's roads got more dangerous?' Well, apart from pot holes,
how? So the increase in cycling since 2012 has nothing to do with it?

Road safety 'experts' are often folk who's CV doesn't cut the mustard.

posted by Sedgepeat [61 posts]
8th January 2014 - 13:06

like this
Like (2)