DfT endorses new method to measure speed camera impacts

The DfT’s head of statistics has endorsed a new method for measuring the effect of speed cameras on road casualties, which aims to isolate the effect of the cameras from random variations.  

The ‘four-time period’ method has been used by Slough-based engineer Dave Finney and Professor Richard Allsop of University College London in reports examining camera effects. 

Finney explained his method in a report analysing data from mobile speed camera sites in the Thames Valley Safer Roads Partnership area (LTT 23 Nov 12). He separated the casualty data into four time periods:

• a pre-Site Selection Period when no cameras were present

• the Site Selection Period (SSP) – the period of time in which accident/casualty levels influenced the decision that a camera should be sited at the specific location

• ASBiC (After site selection but before installation of the camera)

• after camera installation 

Finney ignored the SSP period data because this would contain the influence of selection bias – i.e. an abnormally high numbers of accidents (the ‘regression to the mean’ – RTM – effect).

He therefore combined casualty rates in the pre-SSP and ASBiC periods to determine the mean collision rate before the start of speed camera operations. This was then compared with the rate after camera installation. 

Finney said the method proved “beyond reasonable doubt that the entire reduction in killed and serious injury casualties at mobile speed camera sites actually occurred due to ‘regression to the mean’, and not as a result of mobile camera operations”.

Allsop also used a four-time period method (though with a slightly different treatment of time periods) in a report advising citizens about how to interpret casualty data at speed camera sites, published by the RAC Foundation this summer (LTT 14 Jun). The report included analyses of data from fixed camera sites and concluded that cameras did have an observable effect on cutting personal injury collisions. 

Tim Stamp, the DfT’s head of statistics profession, endorses the four-time period method in a letter to retired engineer and speed camera critic Idris Francis.

“The basic conceptual framework for taking account of regression to the mean (RTM) in safety camera analysis, as used by both Professor Allsop and Mr Finney, strikes me as straightforward and logical,” says Stamp. 

Coincidentally, the RAC Foundation this week published a revised version of Allsop’s report. “The main change I have made is in the recommendation I make (to users of data who do not have knowledge of the SSPs for the cameras they are investigating) about the assumption they need to make about the SSP to apply my method,” Allsop told LTT.

“I recommended previously that they assume the SSP to be the last three calendar years before establishment of the camera. In the light of helpful comments and analysis by Professor Mike Maher [of the University of Leeds], fuller consideration of Dave Finney’s work and sets of recorded SSP for several partnership areas, I now recommend that they assume the SSP to be the first three of the last four calendar years before establishment of the camera.” 

Allsop has reworked his calculations with the revised assumption. His estimate of the reduction in personal injury collisions attributable to cameras across the nine partnership areas studied (previously 15%) is now 14%, and the estimate for fatal and serious collisions (previously 27%) is now 22%.

Explaining that the changes bring his method closer to Finney’s, Allsop told LTT: “I stand by my assessment that my method largely allows for regression to the mean.  I recognise that some of the reduction my method attributes to the effect of cameras may still stem from RTM, but I don’t believe that this can be more than a small proportion.”

In his letter to Idris Francis, Tim Stamp says a large-scale study of speed camera effects could be invaluable. “As site-level camera data continues to be made publicly available, it’s easy to see real value in a robust analysis that encompasses a wider range of camera sites and geographic areas, and that takes account of RTM and other relevant factors such as trend. You may well be interested in contributing to such an analysis.”

But Stamp adds: “It’s very clear that any results and conclusions will carry most weight and influence if they are presented in objective and dispassionate terms, with their methods, assumptions, calculations and underlying data all clearly described and made available for scrutiny. 

“Publication following independent peer review would be by far the best way to ensure this outcome.”

 Guidance on use of speed camera transparency data – updated November 2013 is available at http://tinyurl.com/nnllbuj

Discuss speed restrictions and camera enforcement at LTT's 'Time for 20mph' Conference in February 2014