Sign Out
Logged In:
 
 
 
 
 
Tab Image

Safety cameras and road accidents: effectiveness in local authority areas in England

This paper is a shortened version of the paper published as Safety cameras and road accidents: effectiveness in local authority areas in England, Journal of the Operational Research Society (2011) 62, 1181–1188, published online 2 June 2010. The authors were G A Hindle (University of Warwick, Coventry, UK), and Hindle (HS Ltd, Preston). Its purpose is to introduce the basic ideas of their paper. The full paper is available online on this site (www.theorsociety.com/Pages/Publications/JORS.aspx) to members of the OR Society.

Introduction

Safety cameras—previously called speed cameras—have been a central component of the road safety strategy within the UK (and many other countries) over recent years. Major investment in safety cameras in the UK was stimulated by central government when in 2000 a National Safety Camera Programme was instigated. This programme allowed the costs of prevention, detection and enforcement of speed and red-light offences to be reclaimed from the related fines. By 2007 Safety Camera Partnerships (SCPs) were in existence covering around 90% of local highway authorities in Great Britain. However, in March 2007, this programme as such was terminated by central government and safety camera programmes incorporated into a wider framework of Road Safety Partnerships (RSPs).

This change is not merely organisational but fundamental in the sense that national hypothecated funding of safety camera programmes has ended and decisions concerning investments in, or commitment to, safety cameras are now essentially local ones. In simple terms it is now the responsibility of local highway authorities (and their partners) to judge for themselves whether safety camera programmes should continue to expand, remain as they are or be rolled back in their areas of operation. Evidence on the effectiveness or otherwise of safety cameras will need to be evaluated by these authorities in comparison with other (competing) ways of spending road safety funding. It is clear that attitudes are already beginning to diverge between authorities with some remaining fully committed but many beginning to question the cost effectiveness of the programme.

In relation to the funding of safety camera programmes the changes in arrangements since the termination of the national SCP programme are considerable. In England, around £110 million per annum has been added to the overall road safety allocations to local highway authorities to compensate for the loss of income—estimated at around £93 million—from fines and penalties. However the extra funds are not directly linked to the current operational costs of safety cameras in local areas but are based mainly on the road casualties experienced in such areas during the years1994 to1998: the baseline years for government targeted casualty reductions to the year 2010. Road casualty data and analyses are also presented in an annual publication from the Department for Transport including reports on trends in relation to casualty reduction targets.

The association between this new funding and current camera programme costs is only a relatively weak one (R2=48%) and, hence inevitably many areas are worse off as a result with a clear incentive to consider rolling back the programme. On the other hand, other areas will be ‘over-funded’ and, in principle, could consider expanding the programme. However, in practice, because the link with income from fines and penalties has been broken and this incentive has disappeared, authorities are more actively considering using ‘extra’ funds in other ways. Furthermore, in relation to any expansion of safety camera programmes, central government is now focussing more on the potential for speed over distance cameras as opposed to the currently dominant fixed point and mobile cameras and such cameras have been trialled at around 200 sites in England over the past 4 years.

The research described in this paper aims to contribute to reflections on the effectiveness or otherwise of safety camera programmes and examines the relative success of existing programmes in local areas and explores local factors that may have influenced success or otherwise and, hence, may be relevant to strategic decision making by the newly formed RSPs.

Previous research into safety camera effectiveness

It should be noted that the Department for Transport's 2005/06 research programme did include an intention to commission research into understanding the wider mechanisms of change in accident occurrence and driver behaviour brought about by safety cameras with the stated aim of helping improve understanding of the wider effects of safety cameras. However, this intended project was abandoned.

The research reported here has focused on safety camera impacts in local highway authority areas in England. These areas are largely coterminous with the SCP areas that were formed during the (now) discontinued national programme and with the new RSP areas. However, the database that has been assembled for this research (accidents, casualties, camera sites, demographics, roads, traffic, geography and so on) has been mainly built up at a local authority (LA) level. There are 149 LA areas in England and the RSP areas are coherent clusters of these LA areas.

This research has focussed on PIC accidents and has not separately considered KSI casualties for reasons that have been discussed above: firstly, that the larger numbers of PIC accidents means that these are less subject to random variations and associated regression-to-mean effects than are KSI casualties and secondly, that the observed downward trend in the proportions of injuries that are being classified as ‘serious’ might reflect judgemental shifts (by the police) rather than a reducing severity of injury.

Thus the essential relationships that are explored in this research are between safety camera programmes and PIC accidents.

Variations in accident reductions (PIC) between local areas

The Figure shows the percentage reductions (ie improvements) in PIC performance over the past decade at LA level (baseline to recent). The baseline is defined as the average incidents between 1994 and 1998 (the official government baseline for targeting casualty reductions to 2010) and ‘recent’ is defined in this study as average values between 2004 and 2007. The chart illustrates the wide variability between areas and indicates clearly that there must be significant local factors influencing performance.

Figure
Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact help@nature.com or the author

Overall improvements in accident performance at local authority level.

Many of the major initiatives that have been shown to influence accident potential operate at a system-wide level: such as improvements in vehicle safety design, legislation on drink driving, seat belt wearing and mobile phones, road surfacing and road maintenance and so on. Also the most reported ‘contributing factor’ to road accidents appears almost ‘uncontrollable’ viz. road user judgement errors or inattention that is failure to see, recognise or deal appropriately with danger signals. Following this line of argument it might be expected, if the possible impact of the local safety camera programmes is excluded from consideration, that local areas would have followed national trends reasonably closely and that there would only be relatively small differences in accident improvement rates between local areas. However, the Figure shows that this has not been the case in practice. Also, in the PA/UCL study of camera site impacts in the various SCP areas the variability reported is remarkably high: from virtually zero to over 70% reductions in PIC accidents.

The most immediately identifiable difference between high and low performing local areas appears to reflect a rurality/urbanicity dimension. Areas that are classified using the Office for National Statistics (ONS) guidelines as ‘urban’ show a significantly higher improvement rate than other areas (ie more rural areas). The ONS guidelines are based on proportion of local wards (small areas) classified as ‘urban—less sparse’ where ‘urban’ is a settlement of greater than 10?000 populations and ‘less sparse’ refers to the wider environment of the settlement. For urban areas the average improvement recorded is around 23% whereas for rural areas this average performance is only around 12% over the past decade.

One factor that might have had an influence is speed management and particularly in this regards the safety camera programme. To an extent, safety cameras are less of a feature in more rural areas because of higher proportions of roadway that do not have posted speed limits. However, there are important exceptions to this and some roads in rural areas are recognised as especially dangerous and especially prone to fast driving and on such roads there are can be frequent camera sites both on the open road (normally limits of 50 miles per hour, although some speed camera operate on roads subject to the national rural speed limits of 60 and 70 miles per hour) and through the villages on route.

Before taking the analysis of this possible safety camera effect further it is useful to consider indicators of the extent of camera sites introductions in local areas and this analysis is described below.

Deriving an indicator of camera intensity on local roads

It is clear that local areas in England vary considerably in their reliance on a safety camera approach to speed management. The number of sites per head of population is over 50 times greater in the highest investment area than in the lowest. Other simplistic indicators reveal similar patterns across LA areas such as sites per SLR length, sites per vehicle kilometre and so on. Also speeding prosecution incomes vary widely: currently from a few pence through to over £7 per head of population.

In order to reveal more clearly the potential impact on road safety an indicator has been derived in this research of the relative camera site ‘intensity’ between areas and over time. The underlying hypothesis is that the more likely it is that drivers have come into ‘contact’ with sites during normal journeys (trips) the greater will have been the influence upon them. Thus, in the approach taken here camera intensity (for an area) is based on the number of sites likely to be encountered on a ‘typical’ local journey.

A typical journey in this sense will have different characteristics in different areas particularly average distances and types of road in terms of speed limits will vary. For example, in highly urbanised areas journeys will tend to be shorter (than in more rural areas) and a high proportion of roads will have posted speed limits (ie roads speed limited at 50 miles per hour or less where the majority of camera sites are located). Data on trip characteristics are available in the annual National Travel Surveys (an annual sample survey of personal travel carried out by the Department for Transport) at a regional level and also for areas defined on a rural urban basis (Department for Transport, 2007). The camera intensity indicator measure (for an area) derived here is

Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact help@nature.com or the author

where, s, the number of camera sites; p (SLR), proportion of travel on speed limited road; d, the average trip distance (kilometres); L (SLR), the length of speed limited road (kilometres). The indicator derived is in effect an approximate estimate of the average number of camera sites likely to be ‘contacted’ on a typical trip in an area. It is unlikely that these estimates will be especially accurate in an absolute sense but relative values should capture driver/rider perceivable camera intensity differences between areas.

ACI and PIC accident improvements

LA areas can be assigned to PIC performance categories on the basis of percentage reductions between the baseline years (average of 1994 to 1998) and recent years (average of 2004 to 2007). In the Table values for percentage reductions have been selected in order to assign an equal number of LA areas to five categories from very good to very poor over the decade. A number of index values associated with these categories are shown including an index of camera intensity derived from the indicator measure described above.

Table. Illustrative index values for areas categorised by PIC accident reductions (baseline to recent)

Measure Very good Good Average Poor Very poor

LA count

29

30

30

30

30

PIC reductions

1.98

1.53

1.10

0.68

0.03

PIC risk BL

1.71

1.19

0.89

0.91

0.82

PIC risk recent

1.32

1.02

0.88

0.98

0.99

Camera intensity

1.55

0.99

0.86

0.99

0.86

Net camera intensity

1.43

0.93

0.88

1.05

0.86

Percentage non-sparse pop

1.36

1.20

0.93

0.86

0.89

Percentage young pop

1.07

1.05

0.99

0.97

0.97

Deprivation

1.17

1.08

0.90

0.91

0.95


For both camera intensity and net camera intensity the differences between PIC performance catergories are statistically significant at better than 5% level, as judged by a chi-square test on the counts of Local Authorities above and below the all-England average values.

Notes on the Table :

  • The distinction between the two camera intensity indicators is that the net value uses the extra camera sites over and above camera sites active during the baseline years.
  • Non-sparse population is a percentage of an area's population resident in wards designated by ONS as ‘urban-less sparse’.
  • Young population refers to the proportion of an area's population that is between 17 and 25 years old.
  • The index of multiple deprivations is based on the work of the Social Disadvantage Research Centre, Oxford University.
  • Risk is calculated per vehicle kilometre on local roads, that is not motorway and major trunk roads.

All the index values in the Table are expressed as ratios to the all-England value of the measure. Table 1 gives a feeling for the types of area that have seen major improvements in PIC accident rates in contrast with areas showing very little improvement over the past decade. Good performance appears to be associated with highly urbanised areas with relatively younger driver age populations and with areas of high deprivation where the risk of PIC accidents was particularly high in the baseline years and remains high (but not as high) at the present time. Current (and net current) camera intensity does appear to be generally higher in such areas with an especially clear distinction between the ‘very good’ and the ‘very poor’ categories, as highlighted in bold in the table.

There is a hint of an association between ‘net’ camera intensity and PIC improvements (%) across all LA areas. However this overall relationship is relatively weak (R2=12%) and appears to be mainly reflecting differences between the ‘very good’ and ‘very poor’ performing authorities as shown in the Table. Also any such overall relationship could simply be a reflection of other associated features of LA areas that are linked with camera intensity. Rurality related indicators (population density, proportions of sparse population) have been used in the quantification of the area camera intensity indicator and there remain weak (negative) relationships across all LA areas between ‘rurality’ and camera intensity with R2 values of between 10% and 15%. Also there is a relatively strong relationship between the baseline risk assessment and ACI with an R2 value of around 29%.

Identifying camera intensity effects on PIC accident improvements

To control for risk and rurality/urbanicity assessments at LA level and identify any independent camera intensity effects a pair comparison approach has been adopted here. Each LA has been paired with another that is as close as possible to it on risk and urbanicity grounds and yet different in camera intensity.

The best single predictor of overall PIC savings at LA level between the baseline and the recent period is the incident frequency at the baseline with an R2 value of around 40%. However a multiple regression analysis also incorporating vehicle kilometres and urbanicity (based on the reciprocal of average nearest neighbour distances per census output areas (COAs) improves this R2 value by a highly significant 15%. These nearest neighbour distances (NNDs) are calculated from straight-line distances between COA centroids where these areas are the smallest census population unit containing around 125 households. The average NND for a target area is the sum of all the minimum distance links between COA centroids divided the number of COAs. The ‘expected’ savings generated from this regression model allows an approximate scoring of LA areas that reflect what is called Improvement Potential (IP) here and five (almost) equal LA number categories have been formed from the rankings of the LA on the basis of this scoring.

Camera site effects

In relation to wider area effects it appears that safety camera programmes have only had a statistically detectable positive influence on PIC incidents in mainly urban areas with high levels of prior (to site introduction) PIC risk. The question arises whether or not this is also the case at camera sites.

The PA/UCL study did not adequately explore this issue. No results are presented in relation to PIC risk and, although results are presented for urban/rural effects, the definition of ‘rural’ used is based only on the speed limit operating at sites viz., urban is defined as speed limits of 40 miles per hour or less and rural sites have limits higher than this. However many of the higher speed limit sites are located within areas that are mainly urban in character. For example, London has a higher proportion of ‘rural’ sites than the country as a whole: around 22% compared with 18% overall. A further difficulty in interpreting the PA/UCL findings with respect to urbanicity/rurality is that many SCP (and RSP) areas contain a mix of both urban and rural LA areas and only combined site impact results are presented.

Although, as mentioned above, no analysis of PIC risk effects is presented in the PA/UCL study, it seems likely that the relative inherent danger of sites in terms of the frequency of PIC incidents prior to camera introduction might influence the subsequent impact of such an introduction. In order to investigate this possibility in more detail further site-based information has been sought from the newly formed RSP areas. For eight of these partnerships it has been possible to obtain fully detailed individual site PIC accident information from the baseline years up to and including the year 2007.

Discussion and conclusions

The area based pair comparison study and the site-based interpretations reach very similar general conclusions concerning the success or otherwise of the safety camera programmes in England over the recent past, particularly that:

  • Overall, safety cameras over the recent past appear to have contributed to a limited extent to observed reductions in PIC road accidents.
  • The contribution of safety cameras shows up particularly clearly in highly urbanised and high PIC risk areas but there is little evidence of any statistically significant impact elsewhere.
  • The likely impact of cameras appears to be strongly influenced by prior PIC incident rates and in particular, in overall terms, at sites where less than 10 incidents per 3 year period are expected an impact has not been detected statistically.
  • Many areas (and especially more rural areas) already have a very high proportion of sites at which estimated impact has been very low or non-existent, where decommissioning should clearly be considered. It is recognised that there are other reasons for site introductions other than prior PIC incident rates, particularly that sites can be introduced because of community anxieties about speeding motorists.
  • As a corollary to the above conclusion many areas will inevitably find suitable qualifying sites more and more difficult to identify although not all suitable sites will necessarily have been identified to date and ‘new’ high-risk sites can emerge from traffic system changes.

It is difficult to escape the conclusion that the era of fixed point and mobile camera expansion has ended, that the peak has been reached and that the overall strategic direction is one of ‘rolling back’. In relation to removing existing sites, RSP partnerships have developed (or are developing) protocols based on monitoring speeds and accidents at current sites being considered for decommissioning. However there is, as yet, virtually no information in the public domain on the extent of decommissioning or any quantitative data on the effects. This issue interacts with a lack of information on the extent of camera removals from existing sites prior to full decommissioning. It is very clear that this issue is a politically highly charged one with a fear of litigation associated with accidents that might occur at decommissioned sites. For example, the first announcement of an intention to remove cameras in one area caused a furore of claim and counterclaim with one MP (Anne Snelgrove, parliamentary private secretary to Government Transport Secretary) immediately starting a ‘Save our Cameras’ campaign and a national road safety group (Brake) denouncing the move as ‘reckless’ (Jane Whitham, a spokesperson for Brake).

Full version first published to members of the Operational Research Society in JORS June 2010