Road Safety News

Maths professor challenges GoSafe camera data

Monday 22nd February 2016

A professor who specialises in the mathematical and statistical modelling of transport problems is once again challenging data relating to the effectiveness of speed cameras in terms of reducing road casualties.

In December 2015, Mike Maher, an honorary professor at University College London (UCL), challenged camera data produced by TfL, and is now questioning that produced by GoSafe, the Welsh camera partnership.

An article in Local Transport Today (subscription only content) says that a report by the Local Government Data Unit Wales (LGDU) concluded that fixed cameras operated by GoSafe may have reduced fatal or serious collisions (FSCs) by around 46%. 

However, when professor Maher used data from 238 sites (61 fixed and 177 mobile) to assess camera effects on personal injury collisions (PICs) and FSCs he reached a different conclusion.

In his ‘Alternative Report on GoSafe Speed Camera Data’, professor Maher estimates that at fixed camera sites PICs fell 28.1% and FSCs fell 13.6%.

Meanwhile, for mobile cameras Maher estimates that PICs increased by 16.4% and FSCs rose 29.9%.

He then combined fixed and mobile data to produce an estimated 2.1% reduction in PICs and a 15.2% rise in FSCs at all camera sites. 

The contrasting results appear to be down to different methods used for the two sets of analysis.

The LTT report says the LGDU used a method deployed by professor Richard Allsop for an RAC Foundation report first published in June 2013 and revised in November 2013, while professor Maher used the ‘four time periods’ method to conduct his analysis.

In December 2013 both the DfT and RAC Foundation endorsed the four time period method, which excludes data from the site selection period (SSP) on the basis that it can contain abnormally high numbers of accidents.  

We have invited GoSafe to comment on this matter but have not as yet had a response. If we do hear from GoSafe we will add their comment in due course.


Comment on this story
Report a reader comment

What's your view - comment on this story:

I confirm that I have read and accept the moderation policy and house rules relating to comments posted on this website.
Your comment:
Your name and location:
Your email:

Isn’t the mean of a series of numbers the average of all the numbers, including high, medium, and low? It would be misleading to only use highs for before to compare with after, but sites weren’t all picked on the basis of a high point in their numbers. It is misleading to exclude the highs from the before figure, averaging only the low / medium numbers, and call that the mean.

Another fault in comparing accident numbers at sites is to assume that the cameras only have any net effect at those sites, and to assume that improvement in driver behaviour elsewhere is a trend that would have happened without the benefit of cameras.
David S, Scotland

Agree (3) | Disagree (4)

Shouldn't those with such a strong desire to establish the effect of detecting and prosecuting speeders on accidents, also have to include those offenders detected by police patrols and not just via cameras? I believe it is the same offence.

If they spot a speeder, police patrols don't check the accident history for the location, nor do they bother with RTM, SSP and FTPs (and any other acronyms) and probably don't expect it to necessarily have any effect on the accident rate for that particular road anyway - a speeder is a speeder wherever they are and need dealing with whether it's via automatic technology or by a police officer.
Hugh Jones, Cheshire

Agree (5) | Disagree (5)

It is excellent news that more road safety professionals and independents are considering, understanding and applying the FTP method. This is particularly important because the government has recently made it a requirement that speed camera operators must publish the effect of speed cameras on road safety. If this new rule is to be complied with, the operators must either run scientific trials, or use the FTP method.

The accuracy of the FTP method depends upon the correct identification of the SSP. This can be done by finding out what the SSP was from those who selected the sites (often not available), or it can be done from the actual crash data. The site-selection process will leave a "fingerprint" on the data. All it takes is to recognize that fingerprint and the FTP method can produce the most accurate evaluation of speed cameras (or any other intervention) that it is possible to achieve.
Dave Finney, Slough

Agree (4) | Disagree (4)

The FTP method developed by Dave can only be applied to sites that were created as a result of a specific collision analysis. This would probably cover the majority of fixed cameras installed during the ‘Hypothecation Years’ and the dates will have been specifically mentioned in the Operational Cased provided to the DfT by the camera partnerships. The analysis was usually carried out using financial years so for cameras installed in say 05-06, the analysis period would have been 02-04 or even earlier. I support this method of analysing camera sites where accurate information is available about the SSP, and don’t support it where the SSP is only estimated. The DfT were sent databases from the partnerships every single year but whether they remain on file is another matter.

It also can’t be used accurately where sites weren’t selected based on casualty analysis. This covers most of the cameras in Thames Valley which have been on the ground for over 20 years and weren’t subject to rigorous collision analysis back in the 1990s.
Richard Owen

Agree (6) | Disagree (6)

Paul Biggs comment that the mobile camera van will clock your speed before you notice it is exactly why they can be effective even when the van isn't there. Speeders wanting to avoid detection would have to slow down on approach to the site before they could see if the van was there or not.

The report that by LGDU Wales for GoSafe shows that average speeds and the amount of speeding has reduced at most mobile camera sites they compared.
David S, Scotland

Agree (9) | Disagree (9)

All of the above reports (Allsop, GoSafe, Maher) use the FTP (Four Time Periods) method in one way or another. The FTP method is the most accurate available because it is the only method that can completely exclude the effect of site-selection (also known as RTM). In order to do this, however, the method must be applied accurately.

Professor Maher has applied the FTP method accurately. This means that his reports (London and Wales) are the most accurate speed camera reports by anyone within authority.

There are now 4 reports that use the FTP method to it's full accuracy (latest first):
Wales speed cameras, M Maher
London speed cameras, M Maher
Fixed speed cameras, D Finney
Mobile speed cameras, D Finney

I wish to congratulate Professor Maher on his excellent work and look forward to the greater understanding of intervention analysis that will, as a result, genuinely save lives and prevent serious injuries.
Dave Finney, Slough

Agree (16) | Disagree (9)

How about the safety authorities hand over the data to a truly independent statistician for analysis, rather than process the figures in-house?
Kevin Williams / Survival Skills Rider Training

Agree (19) | Disagree (7)

Paul is right about the mobile camera vans being far more effective than the fixed ones in that they can detect the offenders from a long distance away, they have the ability to film other offences and because they are mobile, they can go anywhere but because of this versatility, their effectiveness can't be quantified which probably comes as a disappointment to some.

I worked with Go Safe for many years and the work they did in all of the three 'E's - not just enforcement - made and still does make a big contribution to 'keeping the Welsh roads safer' as it says on the tin or, in this case, on the side of the van.

I don't know why people like Professor Maher waste their time on things like this. Fixed cameras are limited in their site-specific effectiveness anyway.

It's like asking automotive brake manufacturers for proof that their products have prevented accidents and how many - impossible.
Hugh Jones, Cheshire

Agree (14) | Disagree (17)

I have a letter in the same issue illustrating the effect of regression to the mean using data from a speed camera site in Staffordshire where casualties fell 44% before the speed camera was installed. The original rules for speed installation required at least 3 KSIs in the 36 months prior to a camera proposal being submitted – so camera installation was likely to be the result of an unusual number of deaths or injuries, that would regress back to the mean regardless of whether cameras were installed or not. This isn't to say that if there is such a thing as a carefully placed, visible fixed speed camera, where there is a significant risk or hazard due to a speed limit being exceeded, wouldn't help to reduce casualties. We should also bear in mind that fixed and mobile cameras are very different animals - mobile cameras are an occasional part-time measure that have already clocked your speed when you notice the camera van, plus they can't credibly claim to reduce casualties when they aren't actually deployed.
Paul Biggs, Staffordshire

Agree (16) | Disagree (7)