Viewing 19 posts - 1 through 19 (of 19 total)
  • London cycle deaths – stats analysis
  • Ogg
    Full Member

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

    “6 cyclists were killed in London in a 2 week period between 5th and 13th November 2013. Should we be surprised that this happens at some point in recent history?”

    mattjg
    Free Member

    without reading the link – no, and neither should we be surprised if there is a period with no incidents.

    the worse mistake (Deming said) in understanding a system is confusing common causes with special causes.

    most likely the casualties are systemic (the fix is to fix the system) rather than special (where each incident is treated as a ‘special case’)

    http://en.wikipedia.org/wiki/Common_cause_and_special_cause_(statistics)

    brooess
    Free Member

    Whoever worked that out is definitely cleverer than me 😀

    Anecdotal experience in recent weeks suggests the media hullabaloo is leading to more considerate driver behaviour in London and its environs. Certainly Tfl and Andrew Gilligan are at pains to demonstrate they’re dealing with the poor infrastructure.

    Although I’d like them to focus more on making bullying and incompetent driving socially unacceptable… they blame Tube passengers for delays (don’t hold the doors open) rather than the infrastructure but don’t blame drivers for killing cyclists – they blame cylists and the infrastructure… 😯

    Fresh Goods Friday 696: The Middling Edition

    Fresh Goods Friday 696: The Middlin...
    Latest Singletrack Videos
    mattjg
    Free Member

    Anecdotal experience in recent weeks suggests the media hullabaloo is leading to more considerate driver behaviour in London and its environs

    Good news.

    I think ghost bike installations are potentially a very powerful tool, they humanise a statistic in a very immediate way for the viewer I think.

    Although I’d like them to focus more on making bullying and incompetent driving socially unacceptable…

    My hunch would be that, nationally, driver incompetence, inconsideration and aggression account for many more casualties and near misses than the left hand undertaking and so on that are common in London.

    Also deaths make headlines but I’d bet for every death there is a much higher number of life-changing serious injuries.

    TiRed
    Full Member

    Nice piece of analysis from David Spiegelhalter. He’s professor of risk at Cambridge and and all round good guy when it comes to Bayesian statistics. It was an analysis I’d looked at already, but I wasn’t certain of the underlying Poisson rate. To be honest, I suspect deaths are negative binomial with over-dispersion, but haven’t I don’t have the actual dates for deaths.

    EDIT: and a previous analysis by Spiegelhater on murders in london (taken from the paper) for analogy

    Back in 2008, on a single day, July 10th, four
    people were murdered – all stabbed – in separate,
    unconnected incidents in London. There are
    about 170 murders a year in London; on most
    days there are no murders at all. Four on one day
    led to media headlines of the “London, Murder
    Capital” variety. But a similar analysis by David
    Spiegelhalter of the daily variation in murder
    figures (Significance, March 2009) found that
    four murders on one day could be expected about
    once every 3 years – so nothing terribly unusual
    had happened, and the newspaper headlines were
    merely announcing nothing very remarkable.

    gwaelod
    Free Member

    nobody has highlighted that the stats analysis concludes that

    “and so the chance of getting at least 6 deaths in any 2-week window over 8 years is estimated to be 2.4%. For whatever reason, this is therefore a surprising cluster.”

    MrAgreeable
    Full Member

    I can’t pretend to understand the maths in that article, but I’m guessing from the tone (and the chap’s other article quoted above) that the author thinks 6 cycling deaths over a short time is nothing to write home about.

    If so, it’s kind of missing the point. Lots more people die from diseases relating to inactivity than get killed cycling in traffic, but the underlying cause – a system of transport that’s completely geared around private motor vehicle traffic – is the same. Also ignores the high proportion of those deaths involving HGVs, and the concentration on certain locations, e.g. Bow roundabout.

    bails
    Full Member

    Deaths are probably not the right thing to look at in any case. The numbers are so small that you get big proportional spikes and dips all the time. What’s (probably) better to look at is the KSI rate. A 2 minute delay by an ambulance or an inch either way as a HGV sucks you into it’s wheels can make the difference between a ‘K’ and an ‘SI’, which then doubles the death rate for the whole month.

    But then you’ve got 50 times as many ‘serious injuries’ (14 vs 657 IIRC for 2012) so one more or less doesn’t affect the KSIs much but does affect the Ks. The Ks have dropped (faster ambulance response? Better trauma doctors?) but the KSIs haven’t so Boris and TfL just keep talking about deaths.

    gwaelod
    Free Member

    The author is actually saying this is unusual and unexpected.

    asterix
    Free Member

    unusual and unexpected

    maybe, but the stats don’t enable any further analysis of the reasons for the cluster. without more info on the particular circumstances, all we can do is ‘shrug our shoulders’

    molgrips
    Free Member

    If so, it’s kind of missing the point.

    No I think you’re missing his point actually.

    He’s not saying it’s acceptable. He’s just looking at the statistics of how likely a cluster of events is, and if that really indicates a raised level of danger or not.

    MrAgreeable
    Full Member

    There’s nothing in there about a raised level of danger, or any underlying causes. He just says the cluster is unusual. It’s not really looking at the bigger picture, is it?

    poly
    Free Member

    There’s nothing in there about a raised level of danger, or any underlying causes. He just says the cluster is unusual. It’s not really looking at the bigger picture, is it?

    I think you are misunderstanding the point of the question. The point of his question, is simply to ask if there is even an ‘event’ worthy* of investigation / explanation or if, in fact, there are no more accidents than ‘normal’ and thus anything you find or suggest is no different from any other 2 week window and chance just played a big part.

    I’ve generally found people struggle with understanding Poisson distributions. This example usually helps, with the simpler problem he describes: Lets imagine you have 208 piggy banks (each one represents a fortnight in the 8 yr period). Now give 108 people a penny and ask them one-at-a-time to put that penny in a piggy bank at random (each penny represents a cycling fatality). Do you have any piggy banks with six pennies in it? To understand how often that can happen you do that experiment lots of times and measure the results. Fortunately the clever mathematician Poisson worked out the maths which describe that so we don’t have to. You’d expect to do that experiment 35000 times to get one bank with 6 coins in it.

    The more complicated question arises because it is actually any consecutive fourteen day period, rather than consecutive fortnights. If you do that experiment (with 2922 piggy banks – looking for 6 coins in any 14 neighbouring banks) you expect to see that happen 24 times in 1000 experiments. That is not a particularly rare occurrence, but it is unusual.

    However, to understand it properly you need to consider, eg. are accidents more likely at certain times of year, or in certain weathers, or over that 8 yr period has the popularity of cycling increased (and how does that impact the likelihood of an fatality). When you include those elements (equivalent to making some of the piggy banks easier to reach than others) the chance of getting 6 in a 14 day period are changed – and potentially (because I haven’t done the analysis) no more than chance.

    * I am not implying that any road death is not worthy of investigation or explanation – but that this ‘cluster’ might have some bigger picture factors worth looking for, beyond each individual case, but you first need to know if a cluster of cases is odd.

    J-R
    Full Member

    without reading the link – no

    But actually reading what he says, the opposite is true!

    There is only a 2.4% chnace of us seeing this in 8years, so “For whatever reason, this is therefore a surprising cluster”

    It’s not really looking at the bigger picture, is it?

    But it’s useful to resort to having some facts before discussing the bigger picture.

    mattjg
    Free Member

    without reading the link – no

    But actually reading what he says, the opposite is true!

    guilty as charged

    mattjg
    Free Member

    and remember to always buy your lottery tickets on the day of the draw, otherwise you’re more likely to die before the draw than win

    true dat

    TiRed
    Full Member

    Statistics really isn’t concerned with causation, only the chance of an observed event (unobserved if you are a Frequentist). The murder example is the case in point; four murders on any one day is not surprising given the background rate. But six deaths in 14 days is unusual. But unusual events can occur by chance, and the 1/20 definition of chance is arbitrary.

    The most depressing fact is that the rate is 0.57/14 days. Implying an average waiting time of only 24 days between deaths – just sufficient for the media to lose interest.

    flap_jack
    Free Member

    and the 1/20 definition of chance is arbitrary.

    Quite so. Just because it happens to be close to 2 sigma, it’s been adopted. Appalling laziness.

    TiRed
    Full Member

    Only for normal distributions 😉

Viewing 19 posts - 1 through 19 (of 19 total)

The topic ‘London cycle deaths – stats analysis’ is closed to new replies.