-
• #52
My point is that 8 HGV fatalities is not the sample, it's a set of observations from a larger sample, or pair of populations, men and women cyclists, where the available observations are 'Died by HGV this year' or 'Did not die by HGV this year'. Clustering doesn't come into it.
Clustering does come into it. Even if we look at fatalities in general it's still a small sample. You mention people knowing nothing about probability theory but one of the main principles about extrapolating probability is that you can only predict future events from very large groups of numbers. It's the Monte Carlo fallacy.
-
• #53
As you appear to have the figures at your disposal, how about running the numbers.
Only remembering reading
http://www.tfl.gov.uk/assets/downloads/PC-Cas-Factsheet-Final-1986-2007.pdf
from
http://www.tfl.gov.uk/corporate/projectsandschemes/roadsandpublicspaces/2840.aspx -
• #54
I'm not sure that HGV-cyclist death patterns are best explained by the Poisson distribution, though. What are the clusters you are proposing?
Random clusters. The point is that 'random' doesn't mean 'evenly spread out', you get some clustering just by chance. Often people notice random clusters and think something special is making them happen when nothing is, it's just chance.
But other times, there is something special happening.
-
• #55
Clustering does come into it. Even if we look at fatalities in general it's still a small sample. You mention people knowing nothing about probability theory but one of the main principles about extrapolating probability is that you can only predict future events from very large groups of numbers. It's the Monte Carlo fallacy.
Clustering around what, though? The PD needs some dimension, such as time or distance, for observations to cluster in. But in this case we're not interested in time, or distance, or even sequence. We're interested in how many people died, from these populations, this year. And, sadly, this year's numbers aren't much different from last year's numbers.
Statistical significance does improve with a larger sample size, but as I keep saying, we have a large sample size which is all male and female cyclists this year. Not the 8 that we observed to die.
Your point about extrapolation and clustering would apply if, for example, out of the 50 or so cyclists that ride down the side road next to my flat each year, three of them had been killed by an HGV. 50 people is a small sample. Out of all of the people who had been killed by an HGV in London, one might expect to find some geographical clusters. That being so, you wouldn't want to extrapolate from the three deaths per 50 to a similar proportion of deaths in the wider cyclist population. But that's emphatically not what is going on here. The sample size is large to begin with.
-
• #56
Random clusters. The point is that 'random' doesn't mean 'evenly spread out', you get some clustering just by chance. Often people notice random clusters and think something special is making them happen when nothing is, it's just chance.
But other times, there is something special happening.
Like when you wait an hour for a bus and three turn up all at once?
-
• #57
Exactly, if busses arrive randomly.
-
• #58
Exactly, if busses arrive randomly.
I always wondered why that happened :-)
-
• #59
Clustering does come into it. Even if we look at fatalities in general it's still a small sample. You mention people knowing nothing about probability theory but one of the main principles about extrapolating probability is that you can only predict future events from very large groups of numbers. It's the Monte Carlo fallacy.
Either way isn't clustering or not rather missing the point? Any death on the road, cyclist or other wise, is a terrible event. but why are roads more dangerous than trains or planes for example? because any major incident of mass transit invloves a "cluster" and so a large amount of time and public money will be invested into ensuring it doesn't happen again. (I can't find it now but there was an Economist article a few years ago about how much more money was being spent on per life saved on the railways by introducing ATP compared with how many more lives could be saved on the roads for the same budget by improving bad junctions etc) Be it true or not if you can create a public perception of a lorry incident cluster won't it drive a action to work on the problem? Shit as it is, it's an arguement about politics as much as fact.
-
• #60
Except i'd guess buses are 'bunched' by more than randomness. The first bus after a big gap takes longer at the stops because there are more people waiting, so the buses behind catch up, exaggerating the effect.
-
• #61
Except i'd guess buses are 'bunched' by more than randomness. The first bus after a big gap takes longer at the stops because there are more people waiting, so the buses behind catch up, exaggerating the effect.
You've ruined it for me now.
-
• #62
Either way isn't clustering or not rather missing the point? Any death on the road, cyclist or other wise, is a terrible event. but why are roads more dangerous than trains or planes for example? because any major incident of mass transit invloves a "cluster" and so a large amount of time and public money will be invested into ensuring it doesn't happen again. (I can't find it now but there was an Economist article a few years ago about how much more money was being spent on per life saved on the railways by introducing ATP compared with how many more lives could be saved on the roads for the same budget by improving bad junctions etc) Be it true or not if you can create a public perception of a lorry incident cluster won't it drive a action to work on the problem? Shit as it is, it's an arguement about politics as much as fact.
By looking at what is spent on improving safety on the roads, you can calculate an implied value of life, and i seem to remember hearing that you get numbers in the range from tens to hundreds of thousands of pounds.
In the rail industry, they make explicit calculations of whether safety improvements are worth it, using 1.4 million per life (when i was working there, it has gone up since), but they also try to factor in 'societal concern' as extra multiplicative factors that apply in certain circumstances. Lots of people dying at once is one case were societal concern applies strongly, but risk falling disproportionally on a vulnerable group is another that might apply to the current discussion.
-
• #63
By looking at what is spent on improving safety on the roads, you can calculate an implied value of life, and i seem to remember hearing that you get numbers in the range from tens to hundreds of thousands of pounds.
In the rail industry, they make explicit calculations of whether safety improvements are worth it, using 1.4 million per life (when i was working there, it has gone up since), but they also try to factor in 'societal concern' as extra multiplicative factors that apply in certain circumstances. Lots of people dying at once is one case were societal concern applies strongly, but risk falling disproportionally on a vulnerable group is another that might apply to the current discussion.
yeah, what you did there was take my point and make it better :)
Societal cocern is a phrase i didn't have but it's exactly what i mean. One death barely gets a mention in the media, if as part of a campiagn you can group them it can become an "issue" and therefore news -
• #64
Originally Posted by wvm
If my maths is any good that makes female cyclists almost 30 times more at risk that males.......What sums did you do? If I just divide the fatality rates (7 and 1) by the proportions of female and male cyclists quoted in the article (.28 and .72) I get a ratio of about 18:1. A very long way away from 1:1.
Can't remember how I arrived at the figure but 7/.28 = 25 which is still a very high proportion.
Being slightly pedantic, I note the use of the word aggresive with regards to riding style. I prefer to think I ride assertively which is pretty much the same as aggresively but without looking back and flicking a vee ;)
-
• #65
Clustering around what, though? The PD needs some dimension, such as time or distance, for observations to cluster in. But in this case we're not interested in time, or distance, or even sequence. We're interested in how many people died, from these populations, this year. And, sadly, this year's numbers aren't much different from last year's numbers.
Statistical significance does improve with a larger sample size, but as I keep saying, we have a large sample size which is all male and female cyclists this year. Not the 8 that we observed to die.
Your point about extrapolation and clustering would apply if, for example, out of the 50 or so cyclists that ride down the side road next to my flat each year, three of them had been killed by an HGV. 50 people is a small sample. Out of all of the people who had been killed by an HGV in London, one might expect to find some geographical clusters. That being so, you wouldn't want to extrapolate from the three deaths per 50 to a similar proportion of deaths in the wider cyclist population. But that's emphatically not what is going on here. The sample size is large to begin with.
Either way isn't clustering or not rather missing the point? Any death on the road, cyclist or other wise, is a terrible event. but why are roads more dangerous than trains or planes for example? because any major incident of mass transit invloves a "cluster" and so a large amount of time and public money will be invested into ensuring it doesn't happen again. (I can't find it now but there was an Economist article a few years ago about how much more money was being spent on per life saved on the railways by introducing ATP compared with how many more lives could be saved on the roads for the same budget by improving bad junctions etc) Be it true or not if you can create a public perception of a lorry incident cluster won't it drive a action to work on the problem? Shit as it is, it's an arguement about politics as much as fact.
Yes, if we're talking about cycling deaths in general. I'm not, though. I'm talking about lazy assumptions (neither of yours) about female behaviour extrapolated from one statistic, which has a thankfully tiny sample size, which is the number of females killed in London compared to cyclists as a whole killed in London in that year.
-
• #66
in case i have posted wrong.
i firmly believe the problem is the motor vehicles (in this case lorries) not the cyclists.
-
• #67
And the information I found in the NY Times agrees with you, in 90% of the cases.
-
• #68
i firmly believe the problem is the motor vehicles (in this case lorries) not the cyclists.
Not really, I believe that the green approach between the road and pavement encourage more people to think it's okay to undertake;
-
• #69
That's moot if the driver fails to notice you/give you enough space due to:
- driving too many hours
- driving under the influence
- driving without due care and attention
- driving too fast
- not checking/fitting blindspot mirrors
- carrying an unsafe/poorly affixed load
- driving whilst using a phone
- driving to deliberately intimidate
- driving under pressure from employer to meet unrealistic deadlines
- driving on roads ill-suited to large vehicles
- blah
- driving too many hours
-
• #70
Yes, if we're talking about cycling deaths in general. I'm not, though. I'm talking about lazy assumptions (neither of yours) about female behaviour extrapolated from one statistic, which has a thankfully tiny sample size, which is the number of females killed in London compared to cyclists as a whole killed in London in that year.
Indeed, I'd be the last one to say that we could conclude that the difference in numbers was down to female cycling behaviour. What I am saying is that the difference between the numbers of men and women killed by HGVs in London is statistically significant. And that the sample size is not 8.
-
• #71
On the US site http://www.massbike.org/info/statistics.htm there is a huge collection of studies of vehicle on bicycle accident data and saw that there is no reference anywhere to gender related trends. So either there is no significant difference in the probability of a bicycle accident between male and female populations (given that there is a very large sample size being used) or alternatively they just didn't analyse the data for this, but you would thnk if there were such a pattern as the bbc article appears to be suggesting, it would have been highlighted long ago through these resources (there should be no reason why such a trend would be specific to London).
One site I saw in the past which had links to what I thought to be good articles on road safety for bikes was a US site, so you obviously have to adjust the behaviours that they are suggesting for the fact that they drive on the other side of the road to us...
most of it is common sense as you can imagine, but i find it does help to reaffirm the importance of things which I can sometimes get complacent about
http://www.oldskooltrack.com/files/home.frame.html
which links to:
[FONT=Verdana]How not to get hit by cars. Read it![/FONT]http://bicyclesafe.com/and [FONT=Verdana]Road safety skills[/FONT]
http://www.bicyclinglife.com/SafetySkills/RoadVogue.htm -
• #72
On the US site http://www.massbike.org/info/statistics.htm there is a huge collection of studies of vehicle on bicycle accident data and saw that there is no reference anywhere to gender related trends. So either there is no significant difference in the probability of a bicycle accident between male and female populations (given that there is a very large sample size being used) or alternatively they just didn't analyse the data for this, but **you would thnk if there were such a pattern as the bbc article appears to be suggesting, it would have been highlighted long ago through these resources (there should be no reason why such a trend would be specific to London**).
[URL="http://www.bicyclinglife.com/SafetySkills/RoadVogue.htm"][/URL]Did you even bother to try to read any of the articles on that site? The page hasn't been updated for 5 years and most of the links are broken.
The higher incidence of female cyclists dying under lorries in London has been discussed for years - here's one of the earlier papers that makes the link:
http://www.bmj.com/cgi/content/full/308/6943/1534
As to your claim that there should be no reason why such a trend would be specific to London, I can only assume that this is because it 'stands to reason'?
-
• #73
A complete trivial aside, but you can blame the HGVs on Thatcher. Olympics aside, the reason HGVs are so common on our roads is because she broke up the unions and British Rail (which previously dominated the heavy goods transit business with overnight freight trains). Not many of us get run over by trains in London.
-
• #74
I think wiganwill's mentioned before that she was throwing salt on the wound opened up by Dr Beeching.
-
• #75
Did you even bother to try to read any of the articles on that site? The page hasn't been updated for 5 years and most of the links are broken.
No - I did not see that there would be any value in me trawling through the source data to make my own conclusions, and was prepared to just read the conclusions/summaries on the website page i linked to. Have no reason to assume that they are in any way misrepresenting any of the original reports, but obviously cannot prove this.
The higher incidence of female cyclists dying under lorries in London has been discussed for years - here's one of the earlier papers that makes the link:
Thanks I had not seen that. a lot of the discussion on the thread has been around whether or not hypothesis implied by bbc is borne out by the data given the sample size. the truth would be better represented by studies of data over a greater time period, including this paper (and presumably data has continued to be collected since this paper).
As to your claim that there should be no reason why such a trend would be specific to London, I can only assume that this is because it 'stands to reason'?
Yes i am making this assumption myself, and of course could be wrong, apologies if it is ignorant.
My point is that 8 HGV fatalities is not the sample, it's a set of observations from a larger sample, or pair of populations, men and women cyclists, where the available observations are 'Died by HGV this year' or 'Did not die by HGV this year'. Clustering doesn't come into it.