A few posts ago, I wrote about having gotten the GB STATS19 data set to use in a lecture I'm giving in March. I also wrote how I'd forgotten why, in the time between getting and using the data. Well readers, you'll be pleased to know that when in bed last night, and because apparently this is the sort of thing I think about in bed, like a bolt of lightning, the reason came back to me:
How is the rate of cycle fatalities changing in Great Britain?
There are two sides to this question:
- Persons such as Boris Johnson claim that the rate of cycle deaths is falling. The sheer number of new cyclists on the road is increasing the exposure of cyclists to potential accidents and thus the per cyclist rate. This is a good thing.
- The DfT's figures show an increase in cycle deaths. This is a bad thing.
It is of course possible that both of these things is simultaneously the case. In fact, that's what I expect to see. But as a science person, I'm always suspicious of second/third/further-hand data; particularly when it is making a political point. Show me the data and let me make my own decision. Also, I want a graph for my PowerPoint.
Ok, so lets begin.
How many cyclists die in Great Britain each year?
Not such a simple case of searching the data. The STATS19 data set is split into three parts: Accidents (high level and general things like date, location etc.), Vehicles (note that pedestrians wouldn't be covered in this set and vehicles in which nobody was injured might be included if they were involved) and Casualties (not everyone involved in the crash is necessarily injured). Those parenthetical thoughts illustrate the problem here. Fortunately, all cyclists involved in an accident are on a bike (hence in the Vehicle set as such). However (and thankfully), not all cyclists involved in an accident are injured.
What we need to do is to combine the sets in a way we can then select out the records we want. Fortunately with some SQL-fu we can do that. An INNER JOIN between the Accident and Vehicle tables gives us records of all involved vehicles. We then INNER JOIN again with the Casualty table linking the casualties to vehicles. Finally, because we don't want 400-odd thousand records (yes, that many people really are injured each year on the roads and I promise you it's a massive underestimate), we can filter to accidents that involved a fatality and where that fatality was a cyclist.
For 2012 this gives us a figure of 120. To check we've SQL-ed it right, a quick Google, or asking the DfT, tells us that in fact 118 cyclists died. So where's the extra two come from? Well, some squinting at the data later and it turns out that two records show pedestrian fatalities ascribed to pedal cycle vehicles. I.e. two people died by being hit by cyclists. We will exclude them for the purpose of what follows and add this exclusion to out filter.
So 2012 was 118. What about the previous years? Well, now we have to do some substantial table linking and a lot of repetitive SQL.
— SOME TIME LATER —
4955 cyclists have died on Britain's roads between the start of Jan 1985 and end of Dec 2012.
Frequency of Cycle Deaths on Great Britain's Roads: 1985–2012
Plotted though, it looks like that. Not quite what I was expecting from all the hyperbole (on every side). In line with general trends, cycle fatalities have clearly fallen over the last 28 years to about half what they were in the late-80's. The uptick that everyone went nuts over last summer is hardly substantial in the light of the time series; though clearly an increase of any kind is undesirable.
How far do we cycle each year?
But the above is only half the story, we haven't adjusted for exposure. For lack of a less delicate way of putting things: if there are more cyclists, we expect there to be more cycle accidents. If there are more cycle accidents, we expect more cycle fatalities. Clearly, any and all deaths on the road are tragedies but if for example, there were half the number of cyclists on the road now as there were in 1985, then cycling is as risky as it was then. There are obviously some caveats, for example: cars are designed better and absorb impact better these days, cycling on roads and cycling off road carry distinctly different risk profiles (this balance isn't recorded), and this data is across the whole of Great Britain so may hide regional differences. Therefore as ever, statistics should only be used with those things in mind — not something you'll find mentioned in alarmist pro-cycling articles or anti-cycling rhetoric.
Anyway, we can get annual cycle figures for Great Britain back from 1949 onwards from the DfT. These carry some issues themselves but I won't go into that here. Also, there is a discontinuity (likely a change in surveying procedure) from 1992–1993 so the step here is not necessarily a drop in levels. However, it is the best we have so it is all we can use. Note also, that we're looking at the absolute level of cycling, not mode share. There is a case for considering this per journey for example, but the value of trip length vs frequency of exposure is another argument for another time. Here's the data:
Cycle travel in bn veh km in Great Britain 1985–2012 (DfT, 2013) — Data post-1993 not directly comparable with pre-1992.
We can see here that for all the progress we think we've made in getting more people cycling more often, we haven't even put 20% on the distance cycled in 1993. That's disappointing. Clearly we have more work to do. Though as ever, this may hide details such as a proportionately greater number of shorter journeys.
Bringing it together: Risk of cycle death
Finally, we need to divide one data series by the other to get a measure of risk. In this case that's: deaths per billion cycled kilometres:
Cycle deaths per billion cycled km in Great Britain 1985–2012 (DfT, 2013) — Data post-1993 not directly comparable with pre-1992.
Clearly, the risk of being killed in Great Britain has fallen, particularly since the late 90's. One thing this does show is the drama being generated last year, whilst useful in raising the profile of the issues of cycling fatalities, is clearly hinging on noise. This is hardly helpful.
As a sector, we've certainly made some improvements in cycle provision and this has obviously had some effect. However, conversely, this data hides known issues such as the massive over-representation of women in the cycle fatalities or the massive over-representation of HGV involvement in these fatalities relative to their representation in the fleet.
As I said above, every (and I mean every, there are many more than just cyclists) injury and death on the road is a tragedy; to the individual, their family and friends, and to society as a whole. An equivalent of the Swedish Vision Zero is something we should be adopting here, as ultimately, all deaths on the road are preventable. Look for comparison at the progress that's been made on the railways in the last decade. We can learn from that. Just because we have some of the safest roads in the world already, is no reason for complacency. We can, we should, we must do better.
Data sources are linked at point of appearance above. Additionally, STATS19 data is available from: