Shuffling accidents along the M11

Safety Camera no. 50, as it is known to Essex Police, is the unassuming name of one of the most controversial speed cameras in the UK.

When in 2009 the Information Commissioner ruled that the police should disclose individual speed camera revenues it emerged that this camera alone was responsible for over 38,000 speeding tickets in the previous five years, netting about £2.3m in fines. National headlines were inevitable.
'Britain's busiest speed camera', proclaimed the Daily Telegraph on 27 December 2009: 'England's most lucrative speed camera revealed' cried the Daily Mail the next day. Here's the Telegraph headline:
But the controversy intensified last year when it emerged that despite a record number of speeding fines, serious accidents near the camera had increased in the five years after it was installed in May 2001.
For the government, this threatened to undermine public confidence in safety camera policy. For those who believe that speed cameras are just another stealth tax it was a combustible data mix. The camera seemed to be implicated in the very accidents it was supposed to reduce. But does the evidence support such a conclusion?
Since its installation, Safety Camera no.50 has policed the 50mph limit imposed by the Highways Agency near the end of the southbound M11, about a mile and a half from London's North Circular Road. According to Essex Police, the camera was installed under Department for Transport National Safety Camera rules. At that time these specified, among other things, at least four killed or seriously injured within a three year period, and that collisions should be clustered close together on a section of road between 0.4 and 1.5km in length.
But was the camera actually preventing any accidents? In January 2009 the Daily Mail claimed that it was actually causing them, quoting a police spokesman as saying "People are slowing down for the cameras and then speeding up again. This in turn causes the collisions." 
The Mail quoted accident and casualty statistics for the camera location: "Figures released under Freedom of Information laws show that in the five years before the camera was put up, there were 13 accidents and 14 casualties. In the following five years, the number of accidents rose to 16 and casualties to 24."
Motoring campaigner Paul Pearson, who had obtained this information, blamed the camera. But a Highways Agency spokesman countered by saying that the pattern of accidents was not consistent with the camera itself being a factor. The agency published a press statement reinforcing the point, and regional director Jon Griffiths wrote in a letter to the Mail: “To say that the camera on the M11 between junctions 4 and 5 in Essex is causing accidents is untrue … The camera on the M11 has actually been successful in addressing the cluster of collisions occurring approximately 900 metres (1000 yards) south of the current camera location - which is what the camera was installed to do.
“The personal injury accident data (post installation of the safety camera) does not show a pattern of accidents which would be consistent with the camera itself being a factor. I am committed to improving road safety and to reducing the number of collisions on the trunk road network in the South East which includes this section of the M11.”
Apparently without irony, he concluded: "The best safety camera is the one which takes no fines at all but deters drivers from speeding."
By this definition,  Safety Camera no. 50 is not a success. The careful wording of his letter contains a clue to the continuing problem. While a data set this small can yield no conclusions about overall accident trends or causes, the Highways Agency accident plots are more suggestive.
At this point I should declare that I was one of over 3,000 drivers caught by this camera in 2007. In 2008, the Highways Agency sent me maps pinpointing the location of personal injury accidents (PIAs) for six years pre- and five years post-camera installation. The plots cover up to 500m north of the camera site, and about 900m south. The small differences between the accident numbers and those reported in the Daily Mail are probably due to their being drawn from slightly different sections of the motorway. But in both cases accidents increase after the camera's installation.
The maps show a dramatic shift in the location of accidents before and after the camera was installed. In the six years before the camera was installed (see map above) accidents clustered about 900m south of the location where the camera was subsequently placed .In the five years after the camera was installed, the southern accident cluster disappears, but accidents increase, most now occurring about 300-400m north of the camera, see map below:
Evidently, the HA attributes this shift to the camera. But surely there is an element of having one's cake and eating it when they then claim that the camera is not a factor in the accidents occurring after its installation?
No doubt the HA hoped that their statement and Mr Griffiths' letter would subdue the controversy. But it continues, even featuring on the website of an online accident compensation firm. Meanwhile, drivers caught by the camera continued not only to be fined, but also to receive a glossy information leaflet from the Essex Safety Camera Partnership. This includes a strong message about speeding and road safety: a photo-montage of a crashed car, a speed camera, a bouquet of flowers lying on a road, and a teddy bear. Beneath the caption "Why Use Safety Cameras?" was this key claim: 

"Research shows that safety cameras in Essex have helped to reduce the frequency of personal injury collisions by 23 per cent at safety camera locations".

This claim is repeated on the Essex Safety Camera Partnership website, albeit the claim there is for a 22 per cent  reduction.Clearly this reduction did not apply to Safety Camera no. 50. So, how is the 22 per cent figure estimated?
The source research, identified in the leaflet, was conducted by the PA Consulting Group and University College London and published in June 2004 by the Department for Transport: The National Safety Camera Programme, three year evaluation report.
Updated in 2005, the 164-page document includes the statistical model used to estimate accident reductions based on an analysis of vehicle speeds and accidents at 4,000 camera sites. It remains the evidential foundation of the government's safety camera policy.
A summary of key claims is made the DfT website and the full report is here. But there are at least two reasons why the 22 per cent accident reduction remains an over-estimate.The first must be that while the this estimate does take into account the well-documented downward trend in personal injury collisions, despite increased traffic, it does not include an adjustment for regression-to-the-mean.
The problem is that camera locations are selected on the basis of a series of accidents that may have been higher than usual purely by chance. When accidents subsequently reduce, this is mistakenly attributed to the effect of the camera.The report's authors openly acknowledge this potential effect. But of the 4,000 camera sites used in their study, only 216 proved suitable for further regression-to-the-mean analysis.
So their conclusions are something of a fudge: "Some proportion of the reduction observed in KSIs [killed or seriously injured] and a modest proportion of that in PICs [personal injury collisions] is attributable to regression-to-mean, though the reductions attributable to cameras would remain substantial after allowing for this."
The second reason must be that no account is taken in their report of another well-known problem: the under-reporting of serious road accident injuries by the police. There is some evidence that this problem is getting worse: if so, it could make a significant difference.

However, at least the end may be in sight for Safety Camera no.50. A source within the Highways Agency confirmed that it may soon be replaced by average speed cameras.