The Effectiveness of Average Speed Cameras a report commissioned by the RAC Foundation Richard Owen

Background Road Safety Analysis Richard Owen Co-authors Not-for-profit company limited by guarantee registered in England Independent specialists in collision and casualty analysis, evaluation, online analysis systems, intervention design, training and more Richard Owen Former manager at Thames Valley Safer Roads Partnership Specialist in spatial analysis, GIS, and project management Co-authors Professor Richard Allsop - Emeritus Professor of Transport Studies at UCL Dr George Ursachi – RSA Analyst

History of Speed Cameras in GB 2000 – 2007 Focus on casualty reduction Government sets installation criteria 4 Collisions (KSI) per km in 3 years 8 Collisions (PIC) per km in 3 years Speed as a ‘causation factor’ 85th Percentile speeds > 10% + 2mph e.g. 35mph in 30mph limit 20% of drivers exceeding the speed limit The aim of this report is not to review this in any great detail but it important to understand that the programme was very much data-led with a focus on collision cluster sites

Evidence for Casualty Reduction 42% KSI 50% Fixed 35% Mobile 22% PIC 4100 Sites Analysed, results very impressive, although they didn’t analyse the handful of ASC separately

Evidence for Casualty Reduction Regression to Mean 36% at Fixed Sites 43% at Mobile Sites Regression to mean describes the reduction to ‘normal’ levels of collisions at sites that have been selected due to high collision frequencies The Four year evaluation report estimated the impact of RTM on collision reductions for fixed and mobile sites Time

RAC Foundation Objectives To create a national database/inventory of ASC sites of various kinds in Great Britain To establish a suitably large and appropriate control group of sites to enable an understanding of the difference in collision reduction between potential ASC sites with and without such enforcement To establish levels of occurrence of collisions before and after ASC installation (with consideration given to site-selection period, pre-installation and post-installation periods)

How we collected the data Support from manufacturers Support from authorities (Police, local authorities, camera partnerships) Installation dates Site selection periods Prior enforcement Other information Collision data independently sourced

Analytical problems We need to know if some sites are not suitable for analysis Input from authorities was crucial here It is possible that other changes could have occurred but weren’t recorded

Site Selection Periods In order to allow for RTM we need to know about the site selection periods so their collisions can be excluded from the analysis We used 25 years collision data for each site with a figure provided for each individual month which was then allocated to a specific period The month the camera was installed was excluded

Map sample Excluded

29% PIC Comparison sites GB Collisions 2005 - 2015 When reviewing the effectiveness of an intervention over the long term you need to consider what other trends on vehicle and user safety could have had on collision rates. In Great Britain between 2005 and 2015 there was a 29% reduction in all injury collisions for example so this needs to be adjusted for at our camera sites. We used information about similar roads (those with the same classification) in the same area and compared changes on collision rates there against our sites.

Control sites Cameras considered but never installed 9 sections, 25km of roads As we were designing the analysis methodology after the installation had taken place it was not possible to introduce randomised control trials. The use of this methodology on roads is a difficult exercise anyway as there are so many variables that can’t be controlled for. We found 9 stretches of road that were considered for treatment with ASC but ultimately not progressed. In our report we measured changes after the initial scoping phase was undertaken to see if anything happened.

Installation History Very few permanent systems were installed in the years of DfT control. There was some expansion in 2009 and 2010 but then no large increase again until 2015. According to Jenoptik there has been another big year in 2016. It will be possible to keep the database up-to-date in the future with support.

Standard “3 Before vs 3 Recent” Analysis 50% FSC Approach adopted by most authorities Doesn’t take into account trend Doesn’t allow for Regression to Mean 25% PIC The traditional analysis approach ignores RTM and SSP and just focussed on the three years before installation and the three years after If we apply this methodology to the data these are the results we get

Generalised Linear Model Monthly data for each site in each period Takes into account collisions on other similar roads Estimates the effect of the SSP Estimates the effect of installation The GLM was designed by Professor Richard Allsop and uses the tens of thousands of data points collected in the analysis. It isn’t important to understand the model (although you can read the report) but it is important to understand that it produces two sets of results; one for the SSP and one for the installation effect

16% PIC 36.4% FSC Results Site-selection period effect 24.9% FSC 36.4% FSC 16% PIC 16.7% PIC The site selection criteria appears to have resulted in site selection bias i.e. Collisions were unusually higher at camera sites during the site selection period Overall installation effect Site-selection period effect

Results No difference in collision reduction rates at sites installed pre-April 2007 versus after No significant difference in effectiveness on low speed (20 – 40 mph) and high speed (50 – 70 mph) sites Candidate Sites – No significant change in collisions post-consideration Groupings of sites were considered in the report: Those installed during DfT management vs those after – no significant difference Those installed at sites with different speed limits - no significant difference The candidate sites did not show a reduction in collisions post-consideration (only 35) and this could well have arisen by chance

What this means The presence of Average Speed Cameras reduces the frequency of injury collisions, even when other mitigating factors are taken into account When analysing the long-term impact of road safety interventions, consider the influence of general trend If you select sites for treatment based on high collision rates, not all of the subsequent reductions can be attributed to the intervention