category: literaturenote citekey: chapelonroadsafetyfrancehard2010 title: "Road safety in France: The hard path toward science-based policy" authors: "Chapelon, Jean; Lassarre, Sylvain" year: 2010 date: 2010-00-00 2010 doi: 10.1016/j.ssci.2010.04.015 publication: Safety Science url: "https://doi.org/10.1016/j.ssci.2010.04.015" zotero_key: X7GDB7CB zotero_storage: XNJZCH6R collections: transpordi_planeerimine folder: 001_artiklid/Road Safety Indicators firstAuthor: "Chapelon, Jean"

status: converted

Contents lists available at ScienceDirect

Safety Science

Road safety in France: The hard path toward science-based policy

Jean Chapelon a,1, Sylvain Lassarre b,*

^a Conseil général de l'Environnement et du Développement durable, Paris la Défense, France
^b Institut National de Recherche dans les Transports et leur Sécurité, Champs sur Marne, France

article info

Article history: Received 30 July 2009 Received in revised form 12 March 2010 Accepted 25 April 2010

Keywords: Road safety Policy Management Attributable risk Accident data

abstract

Very considerable advances have been made in road safety over the long term and especially in the recent years. This improvement was achieved in part due to the effectiveness of French decision-making system, even if it remained very perfectible. An analysis of the road safety management system is carried out in order to understand the strengths and weaknesses of the organisation of road safety in France.

The organisation of road safety in France is extremely centralised. The basis of road safety policy at the national level rests on an extensive information system (covering accidents, risk exposure, speed, utilisation of mobile phones) and on analyses of road risk (the risks attributable to alcohol, speed and the use of mobile phones).

This statistical information and these risk models are integrated in risk management tools such as monitoring, ranking and policing. Monitoring makes it possible to track the development of road safety, bench-marking to compare the performance of the country's different departments with each other, and policy making to refine the details of a policy.

The development of the governance of road risk is leading managers and decision-makers to perfect data-gathering procedures, standardise and simplify the analytical tools used, and broaden the range of risks covered.

-2010 Elsevier Ltd. All rights reserved.

1. Introduction

Very considerable advances have been made in road safety over the past 30 years. In France, these advances have resulted in a fall in the numbers killed on the roads by about 2% a year, whereas the volume of traffic grew by around 2.5% annually, signifying an improvement of about 4.5% in the effectiveness of road safety policies. This is in line with the long-term model of fatality risk developed on an annual basis, over 50 years, which shows a decreasing trend over time accompanied by sharp falls in risk due to national road safety measures such as speed cameras (a 48% drop between 2002 and 2008) (Chapelon, 2006; Lassarre and Hoyau, 2008).

This improvement was achieved despite the fact that the French decision-making system remained very perfectible.

This article aims to demonstrate the strengths and weaknesses of the organisation of road safety in France and to show how decision-makers are made aware of the findings of research reports and surveys and the extent to which decisions are taken and followed up on a scientific basis.

E-mail addresses: Jean.Chapelon@developpement-durable.gouv.fr (J. Chapelon), sylvain.lassarre@inrets.fr (S. Lassarre).

After outlining in Section 2 the organisation of road safety in France and the role played by the National Inter-Ministerial Road Safety Observatory (Observatoire National Interministériel de Sécurité Routière), we will start, in Section 3.1, by discussing the basis of a policy founded on scientific findings, including the role of essential information systems such as accident records, risk exposure data, speed measurements and measurements of other behavioural phenomena, including the use of mobile phones. The evaluation of ''objective" risk through the calculation of attributable risk will be taken in Section 3.2 as an example for three factors: alcohol, speeding and mobile phones. The risk models adopted are based on epidemiological methods and time series analysis, as recommended by the experts of the OECD (2008).

Secondly, the article will show in Section 3.3 how this information and these models are integrated in three basic risk management tools:

monitoring, which makes it possible to track the development of road safety,
bench-marking, which enables different departments to be compared with each other, and
policy making, which makes it possible to refine the details of a policy.

^* Corresponding author. Tel.: +33 1 4592 5764.

¹ Tel.: +33 1 4081 2364.

We will end in Section 4 by highlighting the inadequacies of the present structure, despite the progress made, and pointing to the avenues for improvement in the future as part of a discussion on the governance of road risk.

2. French organisation of road safety and the role played by the Road Safety Observatory

The Inter-Ministerial Road Safety delegate is in charge of road safety in France. He prepares the Inter-Ministerial Road Safety Committee (Comité Interministériel de Sécurité Routière – CISR) which takes decision, and implements these decisions. The interministerial character of the delegation is more paraded than real. Of course, there are five or six advisors detached from different ministries, such as Interior, Justice, Education and Health, but their ties with their ministries of origin are loose and they have very little influence over their ministry's actions in the area of road safety, which remains in the hands of the new Ministry of Sustainable Development, assisted by a Secretariat of Transport.

On the other hand, the strong point in the French organisation is its hierarchical nature, with the important role played by the Prefect of each department (Fig. 1) who, together with the head of road safety projects, one of his close collaborators, coordinates the local actions of the different ministries and encourages the activities of local community groups.

At the same time, the majority of road safety decisions are taken at meetings of officials on the basis of administrative and legal rather than technical and scientific factors. Scientific arguments hardly penetrate the decision-making body (CISR) or the adviser body (National Road Safety Council: Conseil National de Sécurité Routière – CNSR). The cost and effectiveness of measures are only two criteria among others. Comprehensive studies of the costeffectiveness or cost–benefit outcomes are rare, whereas this was a common practice in the 1970s.

As these measures form part of the government's action programme, the way they are presented to the press and the public is extremely important. Since this tends to be determined by the political agenda, the attempt to create an effect of surprise takes precedence over in-depth studies and discussions with all the parties involved.

Our knowledge of the risks and the effectiveness of measures is evolving. The effectiveness of vehicle testing, which was introduced in France in 1985 and is now in use throughout Europe, has been brought into question (Christensen and Elvik, 2007).

Decision-making is not a regular and linear process and may be subject to comings and goings. The principle of regular medical checks decided in March 2003 was dropped a few years later, when its systematic nature raised questions about its cost and the ensuing debate highlighted the fact that the risk of error (a false negative) represented a major shortcoming of the measure, since it

Fig. 1. Organisation of road safety in France.

prevented perfectly fit drivers from driving, including some elderly people, with disastrous consequences for their daily lives (Leproust et al., 2008a,b).

The mission assigned to the National Inter-Ministerial Road Safety Observatory, which is under the authority of the road safety delegate, is to gather scientific information for the benefit of decision-maker in order to improve the quality of decisions, as well as to guide the activities of departmental road safety observatories (Observatoires Départementaux de Sécurité Routière – ODSR).

To do that, it relies in particular on a committee of experts (Fig. 1) composed of a small number of specialists, who are not necessarily the most qualified experts themselves but are able to call on the most competent specialists in every area they have to deal with.

Matters may be referred to this committee of experts by the Chairman of the National Road Safety Council or the minister responsible for road safety, or the committee may act on its own initiative.

Among the most recent reports produced by the committee experts, one can cite two examples:

An in-depth study on drinking and driving, which required more than 6 months' work and for which the committee heard evidence from a score of outside personalities (Chapelon et al., 2007). Produced on the committee's own initiative, the report contained some 10 recommendations, a good number of which have since been acted on (see Section 3.2.1.4).
A more immediate study on the advisability of introducing a specific regulation limiting young people's access to powerful vehicles, which was carried out at the request of the minister. It gave rise to a recommendation (no prohibition) that was approved by the National Road Safety Council and was adopted by the government.

The last element in the organisation of road safety in France is the National Road Safety Council (Fig. 1), which is a sort of Road Safety Parliament. It enables all the parties involved in road safety (Members of Parliament, highway administrators, insurance companies, motor manufacturers, accident victim defence associations, associations representing particular categories of road user) to meet and debate all road safety issues. As far as our area of interest is concerned, it has the enormous advantage of disseminating the culture of road safety in such a way that all the parties involved have the same level of information.

3. What is a science-based policy?

Road safety policy is founded above all on the input of a system of reliable information composed of, on the one hand, quick indicators, accident records, safety performance indicators (SPI: risk exposure data, and measurements of speed and of the use of mobile phones obtained from roadside surveys) and on the other hand, estimates of the risks attributable to major factors, such as alcohol consumption, speeding, and the use of mobile phones. Lastly, it is based on the management of road risk using three preferred instruments: monitoring, bench-marking, and policy making.

The following section presents the information system (Section 3.1), the attributable risk (Section 3.2), and the risk management (Section 3.3). All these analysis and tools are enforced in routine by the ONISR. The results are presented on the web-site of ONISR and CNSR.2

² www.securiteroutiere.gouv.fr/observatoire and www.securiteroutiere.gouv.fr/ cnsr.

3.1. The information system

The rule that there is no management without measures applies to the management of road safety and necessitates very broadbased information gathering (OECD, 2008), with some national specificities.

3.1.1. Quick indicators (accidents, deaths, injuries and hospitalised victims)

Due to the complexity of accident records (which comprise some 70 items) and of our organisation, the data took, until now before the improvement that are made (see end of Section 3.1.2), around 5 months to arrive. That is why we utilised a very simple system of quick indicators that was confined to the monthly numbers of accidents, fatalities, and injured and hospitalised victims. This system enables departmental police forces to return information to the national level within 3 or 4 days of the end of the month, making it possible for the results to be published around the seventh day of the following month.

We were obliged to adjust corrector coefficients because of a number of errors, but the margin of error is less than 2%.

The advantage of this organisation is that the monthly publication of road accident data by means of a press release is a big event that interests the media. As a result, the coverage given in the media is equivalent to roughly three times of the media impact of our public information campaign.

3.1.2. Accident records

Every personal injury traffic accident must be recorded in a BAAC (analytical bulletin of a personal injury road accident), which has to be filled in by the relevant police force and comprises very full information broken down into four main parts: the characteristics and location of the accident, and the vehicles and users involved, knowing that the users are attached to the vehicles that are attached to the location.

The importance of accident records for the study of accidents requires one to be concerned about their quality. That is why a project was launched in 2000 for modernising accident records, which comprised four elements: a quality plan, the simplification and updating of the form (a new BAAC-2002 standard was introduced on 1 January 2004), the harmonisation of the data (including a definition of the injury severity), and the decentralisation of the architecture (database shared between local and national levels).

The first problem in quality terms relates to under-reporting: too many accidents are not recorded on a BAAC form. The reason for that may be that police forces have not been informed of the accident: that is the case in particular of minor accidents that do not disrupt the traffic, accidents involving a single vehicle, and accidents involving bicycles (Amoros et al., 2008).

To improve the degree of under-reporting, the definition of an injured person was restated and clarified: it is a person requiring medical attention. The definition of a seriously injured person was also simplified: before, it was defined as someone who was hospitalised for at least 6 days; now, it is defined as someone who is hospitalised for at least 24 h, making it easier for the police to record him. Lastly, to reduce losses in transmission, the BAAC's indicators are now compared with the quick indicators for the numbers of killed and injured produced each day or each week by the different police forces.

The decentralisation of the information-gathering architecture: in the current architecture, the record is validated at the central level and the corrections made at the local level3 are not used at the national level. In the new computer-based architecture, the mode of transmission is to remain centralised, but the creation of a shared database will enable the departmental level (prefecture, departmental development directorate, local authorities) to dispose of directly exploitable information at the earliest. The idea is to bring the producers (police forces) and consumers (prefecture, DDE, local authorities) closer together in order to improve quality. This BAAC portal has been in use since March 2009.

3.1.3. Risk exposure

The number of vehicles times the kilometres driven is the most commonly used indicator of exposure to risk, but its accuracy varies according to road network categories. It can be measured very accurately on motorways and main roads using automatic counting devices, while on departmental roads it can be extrapolated using an abacus on the basis of the number of kilometres driven on the busiest network.4 On the other hand, there is rarely a figure available for traffic on urban roads.

In terms of user category, a reliable risk exposure indicator is obtained from surveys of private motor cars, commercial vehicles and heavy goods vehicles. On the other hand, only a very poor indicator is available for motor cycles, and none at all for moped, bicycles and pedestrians.

3.1.4. Speed measurements

For the past few years, the Observatory has been using investigators from a private firm to measure speed and seatbelt or crash helmet use, on the basis of a sampling plan made by INRETS.5 The data represent around 200,000 observations a year and are regularly published, in particular in the Observatory's annual report (ONISR, 2008).

The survey protocol was defined as follows: the only roads retained were straight, level roads without any disruptions to traffic, with no crossroads or traffic lights over a distance of at least one kilometre before and after the measuring point, or a few hundred meters in urban areas. The idea is to choose observation sites where the drivers have no natural limit of speed and can choose their speeds. To measure speed, 362 observation points representative of the French road network were retained (285 during the day and 77 at night). The investigators – around 50 of them – proceeded during half an hour (motorways) or 1 hour (others roads) to all the points in turn to record observations following a predefined breakdown of days of the month and hours of the day, then every 4 months at the same points at the same hour and on the same day of the week. The observations were distributed so they were spread uniformly over the 4-month period and covered every day and every period of the day (between 9.30 am and 4.30 pm in the daytime and between 10.00 pm and 2.00 am at night). Almost all the measurements during the day and all the measurements at night were thus taken outside peak hours. Furthermore, given that each observation point necessitated a sequence of measurements in each direction and in each lane, more than 2000 observation sessions took place each year and more than 200,000 speed measurements were recorded and processed (the year 2008 was incomplete, with 137,507 observations having been recorded during the first 8 months).

3.1.5. Measuring mobile phone use

Measurements were taken by investigators positioned on the roadside at 81 different locations that represented the different types of network: connecting motorways (15 locations), relief

³ Recovering of some lost accidents, and improving localisation, or some variables' contents (type of vehicle, date of driving licence...).

⁴ The statistical department of the ministry of ecology, energy and sustainable development is in charge of statistics for motorways and main roads, ONISR gathers data for secondary roads.

⁵ National Institute for Transport and Safety Research.

motorways (8 locations), interurban roads and country roads (23 locations), and urban roads (35 locations). These measurements were taken between 1 December 2007 and 31 March 2008, on weekdays between 10.00 am and 5.00 pm, and the observation time was 25 min per location. The vehicles observed were motor vehicles with four wheels or more: light vehicles, vans, heavy goods vehicles and buses.

The investigators gathered the following information:

the driver is holding the telephone to his ear,
the driver is holding the telephone in his hand but not to his ear, and
the driver is not holding a telephone.

A total of 15,335 vehicles were observed in the whole survey. It was not possible to measure the use of hands-free telephone kits reliably, however, despite the use of an electronic gadget that can detect the utilisation of a mobile phone. Additional tests will be conducted to try and develop a protocol for measuring the use of hands-free mobile phone kits.

3.2. Evaluating the risks

Evaluating risk at a national level entails estimating the risk attributable to a road safety threat in the broadest sense of the term: either a cause, an exacerbating factor or an attenuating factor. It relies on epidemiological methods and covers general factors that are usually taken into account in matters to do with public health, but also factors that are specific to road transport in terms of infrastructure and vehicles. We then turn to a sequential accident model with temporal divisions determined by events (a sequential model), which are more suitable for formulating action plans at local level.

Three factors are taken as examples: two major ones, which are drinking and driving and speeding, and a new one, the use of mobile phones, knowing that other factors could have been retained as safety belt, although the rate of use is very high in France, and in the area of infrastructure, such as the effect of passing from a section of the network in the countryside to a dual-carriageway road network.

3.2.1. Risk attributed to alcohol

Starting from 2008 data on the relationship between the amount of alcohol a driver has consumed and the risk of a road accident, we look for estimating the number of human lives that could have been saved if no one had driven with a blood–alcohol level higher than the authorised limit (0.5 g/l) which is equal by definition to the attributable risk.

To do that, one has to know how much alcohol drivers have consumed.

3.2.1.1. Alcohol level of drivers on the road. Little is known about the blood–alcohol level of drivers on the road (which is to say not involved in accidents). The prevalence among drivers stopped at preventive police checks in 2007 was 2.8%, but these drivers cannot be regarded as a representative sample, since the police generally target their checks.

In the BAAC form, the ''person presumed responsible", information filled by the policeman,6 enables us to know the alcohol levels of drivers who are not responsible. One may consider that the characteristics of the drivers who are not responsible are representative

Table 1 Percentage of drivers not responsible who are over the blood–alcohol limit (>0.5 g/l) during the week and at weekends in 2008.

	Injury accidents (%)	Fatal accidents (%)
In the week	0.6	1.6
At weekends	1.9	8.3
All accidents	0.9	3.25

Source: ONISR, accident records.

of the characteristics of all drivers on the road, since these are drivers drawn by chance while driving. Using this method, the prevalence in 2008 was 0.9% in personal injury accidents and 3.25% in fatal accidents, the difference being due to the fact that these accidents did not happen at the same time, with the fatal accidents having tended to take place at night and during the weekend. It can be seen that the rate observed in preventive checks (2.8%) is intermediate between the two, which signifies a high degree of targeting in the checks carried out by the police.

One relies that much more easily on these data insofar as a large number of pointers show that data which only interest the police, such as responsibility or alcohol levels, are very well furnished by police forces.

The percentage of drivers involved in personal injury accidents who are over the blood–alcohol limit is three times greater at weekends than during the week. Furthermore, the percentage of drivers who are over the blood–alcohol limit at the time of fatal accidents is still higher (Table 1). This is attributable to the fact that these accidents happen at certain times and especially at night.

3.2.1.2. Evaluating the number of fatal accidents that could be avoided. To calculate the number of fatal accidents that could be avoided, one starts from accidents in which blood–alcohol levels are known7 thanks to the use of an estimate of the relative risk. When the risk is low, the odds-ratio method is a statistical method for estimating relative risk that is based on the yield of crossover products.

The percentages used are, on the one hand, the previously calculated proportion of drivers who are presumed not to be responsible and, on the other hand, the share of fatal accidents in which alcohol is involved. This is equivalent to the proportion of drivers presumed responsible who are over the blood–alcohol limit, if one makes the assumption that, where a driver presumed responsible is over the blood–alcohol limit, it is the alcohol that is mainly responsible for the accident.

The proportion of drivers who are over the blood–alcohol limit is almost 10 times greater among those presumed responsible than among the reference drivers who are not responsible (Table 2). We select as reference drivers those who are not responsible at the time of a fatal accident, since we have seen that the day and the hour play a major role in the distribution of injury and fatal accidents.

The relative risk of being responsible for a fatal accident when one has drunk, as estimated by the odds-ratio calculated from accidents with known blood–alcohol levels, is equal to 11.5. One can calculate the ''efficiency" of a driver with a blood–alcohol level below the legal limit in order to ensure that the accident in which he was involved was not fatal.

His ''efficiency" would be:

1 -1=odds-ratio ¼ 91%:

⁶ ONISR considers that this variable is correctly filled. In 20% of accidents, the policeman does not indicate any responsible user.

⁷ The proportion of fatal accidents with alcohol status known (either no alcohol drivers or at least one positive) is very high: 77%.

Table 2 Breakdown of drivers presumed not responsible and drivers presumed responsible in fatal accidents according to whether alcohol is present or not.

Fatal accidents	Alcohol (%)	No alcohol (%)
Drivers presumed not responsible	3.25	96.75
Drivers presumed responsible	27.9	72.1

Hence, the maximum number of fatal accidents that could have been avoided if no driver had driven with a blood–alcohol above the legal limit would be:

878 91.3% = 802, or 25.4% (802/3151) of the fatal accidents where blood–alcohol levels are known.

3.2.1.3. Evaluating the number of human lives saved. The 878 fatal accidents involving drivers with blood–alcohol levels above the legal limit and who are presumed responsible resulted in the deaths of 972 people. In this case, one thus has an average of 1.11 people killed per accident. Applying that multiple to the 802 fatal accidents that could have been avoided, one obtains a potential minimum reduction in deaths caused by drinking and driving of:

802 1.11 = 890, or 26.08% (890/3412) of those killed in fatal accidents where alcohol levels are known.

Extrapolating from all accidents and assuming that accidents with indeterminate or unknown alcohol levels break down into accidents with and without alcohol in exactly the same way as those where alcohol levels are known, one could avoid around 1000 fatal accidents a year and save the lives of 1200 people.

3.2.1.4. Analysing the risk attributed to alcohol. Beyond the calculation of risk, the analysis of this risk can be made: the alcohol impaired drivers represent a very small part of drivers (1%), a different situation from speed impairment (30% of drivers over speed limit and about 10% 10 km/h above speed limit). The average alcohol rate is very high: 1.6 g/l for all accidents and 1.8 g/l for fatal accidents. During the last ten years, despite all the measures taken, there has been no progress in alcohol risk reduction. This leads the expert committee to propose a new set of safety actions: confiscation of the vehicle (confiscation of the driving licence is not enough: alcoholic driver uses to drive without driving licence), improving relationship between medical driving licence committee and the doctors expert on alcohology.

3.2.2. Risk attributed to speeding

In a certain number of cases, it is not known whether a factor of risk existed or not at the time of the accident and the odds-ratio method cannot be used to estimate the relative risk. That is the case, for example, of speeding and the use of mobile phones.

In effect, very little is known about the involvement of speed in individual accidents. What is know about the effect of speed on accidents does not result directly from our knowledge of the role of speed, therefore, but from a statistical analysis of the evolution of accidents on road networks as a result of the actual speeds practised, which may rise or fall over time as a result of a change in regulations.

In such a case, one method for evaluating what is at stake is to use data derived from international scientific literature, obtaining a relative risk factor from analyses of experiments on speed limits around the world.

To the extent that the relationship between average speed and accidents is well known (Nilsson, 1974; Cohen et al., 1998; Aarts and van Schagen, 2006) and that it consists of a power of 4 between the variation in average speed and the variation in the rate of fatal accidents, it is possible, by calculating the relative theoretical reduction in average speed if everyone abided by the speed limit, to multiply it by a factor of 4 in order to arrive at the relative fall in fatal accidents.

One could object that this distribution of speed is very theoretical and very far removed from the normal curve that is usually observed. This is what is shown by the chart below, which depicts the development of the distribution of speed on French motorways (Fig. 2). One sees that, even after the installation of speed cameras, there remains a non-negligible proportion of drivers who exceed speed limits, while at the same time a significant proportion of drivers who already abided by speed limits before 2002 reduced their speed in the following years.

However, even if one knows that, whatever measures are taken, the distribution of speeds will remain very Gaussian, a definition of gains in average speed that limits the reductions in speed only to those who exceeded the speed limit seems to us to be an indicator which, although theoretical, is useful for road safety policy. In any event, road safety decision-makers would not understand an ideal objective indicator starting from the principle that there would remain an intransigent fringe when it comes to abiding by speed limits.

Moreover, the same is true of the risk attributable to alcohol: there is no chance that all drivers will abide by legal blood–alcohol limits. In addition, this indicator is not that far removed from reality, since even if it is certain that one will never get all drivers to abide by speed limits, the impact on road safety will be amplified by the reductions in speed of those who already abided by the limits, as we saw in the 2002–2006 period.

Retrospectively, the 10% decrease between 2001 and 2008 of the average speed contributes to a 34.5% decrease of the number of fatalities, which represents nearly three quarters (exactly 72%) of the total decrease of fatalities (48.2%).

Fig. 2. Share of drivers below a given speed on motorways in 2001 and 2006.

Prospectively, the average speed would decrease if every driver respects speed limits by 4.8%, which would lead to a decrease of 18% of the number of fatalities, which represents the attributable risk to speed.

3.2.3. Risk attributed to the use of telephones while driving

3.2.3.1. General outline. As far as mobile phones are concerned, we use a method that is similar to the one used for speed, by multiplying the added risk by the risk exposure.

In 2005 an Australian study (McEvoy et al., 2005) sounded the alarm about the importance of the extra risk of telephoning while driving: 3.8 for hands-free sets and 4.9 for hand-held phones, thereby corroborating other studies such as the Canadian study (Redelmeier and Tibshirani, 1997).

To evaluate the risk attributable to mobile phones in France, there was a tendency not to go back to this evaluation of the added risk, taking the view that it had been well covered by McEvoy, but rather to look at the prevalence of this risk in France by observing drivers' use of telephones at the wheel.

3.2.3.2. Frequency of use of mobile phones while driving. In terms of types of network, the proportion of drivers observed holding a telephone to their ear varies from 1.8% on country roads and 1.9% on urban streets to 2.6% on connecting motorway network, on which it seems most difficult to monitor this practice. Taking account of the distribution of traffic between the different networks, the average is estimated to 1.9%.

When the driver is holding a telephone in his hand but not to his ear (0.5%), one may assume that it is in one of the following situations: he is reading or writing an SMS, he is dialling a number, or he is using a hands-free telephone or kit.

These findings confirm that the use of handheld mobile phones while driving is not a rare event. Among motoring offences for which points are deducted from driving licences, it is the next most commonly observed one after speeding offences. The fact that a specific measure is going to be available for road traffic will make it possible to improve the monitoring of this practice and better estimate what is at stake for road safety.

It should also be said that these observations were made during the day, between 10.00 am and 5.00 pm. For that reason, the use of handheld telephones may possibly have been overestimated, since one might think that drivers telephone less while driving at night, making the assumption that it is harder to find someone to talk to at that time of day. Nevertheless, as the proportion of night traffic is low (10% of the total), the results would only be marginally altered.

3.2.3.3. Estimate of the potential consequences of using mobile phones while driving. Drawing on the findings of the McEvoy study, the potential consequences of the use of handheld telephones alone represent more than 7% of additional accidents, equal to 1.9% (the proportion of users) multiplied by 4.9 (relative risk) minus 1 (normal risk).

On the other hand, there are more uncertainties when it comes to estimating the implications of using hands-free phone kits, given that it was not possible to measure this practice from the roadside. The results of an opinion survey carried out by the ONISR in March 2008 show that people who use mobile phones the most while driving are generally equipped with hands-free kits. One might thus think that what is at stake is certainly not negligible, and the ONISR is thinking about the best way to proceed in order to gain more reliable knowledge about the proportion of drivers who telephone using hands-free kits.

Lastly, the March 2008 survey showed how the use of SMSs while driving was developing. It found that an SMS was read every 350 km and an SMS sent every 670 km by drivers who said they telephoned while driving.

3.3. Risk management

Risk management is the way to drive the road safety policy from a sound risk assessment process which identifies the main characteristics of the road risks and their degree of acceptability. Different strategies are available to reduce the frequency and the severity of the traffic accidents: prevention, mitigation, risk avoidance, ... Risk management is based on three components:

monitoring, which makes it possible to track what is happening, once safety options have been implemented, and understand, and even anticipate, the development of the situation;
bench-marking, which enables the performance of each department to be compared with that of the others, when the implementation of safety measures is delegated to them, and then to benchmark the results; and
policy making which enables the details of a policy and of the safety options to be refined so as to make it more effective.

3.3.1. Monitoring: monthly analysis of the situation taking account of the effects of weather and seasonal and time-of-year factors

In view of the random nature of the numbers of accidents and victims, an analysis of time series is recommended to extract the trend and the seasonal effects from the irregular part.

In effect, monthly data are characterised by substantial seasonal variations: traditionally, there are fewer accidents during the winter months, but they are followed by a steady increase during the spring, with the number of accidents rising to a peak during the three summer months and remaining at a high level in the autumn.

These seasonal variations do not make it possible to make a direct comparison between some month and the previous month. As a result, a month is always compared with the same month the year before.

The disadvantage of comparing monthly results at 12-month intervals is that there is no reference level to determine whether it was a ''good" or a ''bad" month. In particular, weather conditions have a considerable influence on the results, mainly as a result of their effect on traffic: generally speaking, good weather conditions bring an increase in traffic and hence a rise in the number of accidents. Furthermore, we do not know whether behaviour during the month in question was good or bad, whereas it would be more logical to compare it directly with the previous month, when behaviour patterns can be supposed to be similar.

To facilitate comparisons between monthly results, the Observatory uses the Giboulée model developed by the SETRA8 and INRETS (Bergel et al., 1994) which calculates risk indicators corrected for the variations year-of-effects of weather and seasonal and time.

With this correction, the Observatory calculates a monthly risk indicator in the form of an ''annual equivalent", which is interpreted in the following way: for example, given that the annual equivalent for May 2008 is 4520 and assuming that behaviour patterns remain the same during the year and that weather is neutral, the result in May 2010 will be 4520 fatalities, which makes it possible to evaluate the results for each month in comparison with an annual target.

Fig. 3 shows that, whereas the results in early 2008 appeared to be positive when compared to 2007, it would be wrong to be optimistic since the outcome for 2008 reflected a bad year-end perfor-

⁸ Road and Motorway Research Agency.

Fig. 3. Annual equivalent of the number of deaths per month.

mance, and if the same trend were to continue, the number of deaths will have increased by some 3% or 4% by the end of 2009.

To sum up, we used a very sophisticated model but applied it in a very simple and pragmatic manner so as to produce clear conclusions for decision-makers.

3.3.2. Bench marking: a method for comparing the performance of each department by standardising the risk

To judge the road safety performance of a department, one traditionally has a choice between two methods: either one tracks developments from year to year, or one compares it with the performance of other departments.

The drawback of the first method is that, not knowing whether the results of the previous year were good or bad, one cannot interpret developments during the current year.

The drawback of the second method is that one cannot be content with a comparison between raw figures (there are large and small departments) and that one has to take account of risk exposure. Population is a totally inadequate indicator of risk exposure, since the number of kilometres driven per inhabitant is known to be very different from one department to another, ranging from 1230 vehicle-kilometres per inhabitant in Paris to 29,380 vehiclekilometres per inhabitant in the Landes (ONISR, 2009).

3.3.2.1. Basic principle of local accident indicators (LAIs). The purpose of LAIs is to provide a departmental indicator that takes account of the volume of traffic and its distribution between the different types of road network (motorways, country roads, urban areas). An LAI compares the level of risk of a department with what it would be if the department had the same levels of risk on its different networks as the whole of France.

From that standpoint, the LAI is based on a classic method of standardisation, since it relates an observed degree of risk to an expected degree of risk that is estimated on the basis of the number of road deaths over five years.

We distinguish between five categories of network: motorways, interurban roads, secondary roads, other country roads, and urban areas.

The risk exposure indicator is the vehicles kilometres travelled or, in the case of urban areas, the population of urban districts of over 5000 inhabitants that have a central town with 5000 inhabitants or more.

An LAI is defined for each department using the following formula, with j the department (from 1 to 96) and i the network (from 1 to 5), such that:

t(i, j) is the number of deaths on network i in department j;
e(i, j) is the risk exposure on network i in department j; and
r(i) is the average risk on network i throughout France.

Thus

$$r(i) = \frac{\sum{j=1}^{96} t(i,j)}{\sum{j=1}^{96} e(i,j)}$$

$$LAI(j) = \frac{\sum{j=1}^{5} t(i,j)}{\sum{j=1}^{5} t(i,j)}$$

One recognises in the LAI the classic relationship between the observed number of deaths and the expected number of deaths. The LAI makes it possible to go from an analysis in terms of an accident rate related to the population to an analysis in terms of an accident rate related to distance, applying the following simplified formula:

LAI ¼ deaths=vehicles kilometres travelled

¼ ðdeaths=inhabitantsÞ=ðvehicles
- kilometres travelled=inhabitantsÞ

where the vehicles kilometres travelled are a weighting of the vehicles kilometres travelled observed by the average risks by network in France as a whole.

Standardisation by vehicles kilometres travelled (and hence the LAI approach) provides a lot of ''explanations", since the maximum difference falls from 1 to 10 to 1 to 3 and the variance narrows from 0.36 to 0.25.

Interpreting LAIs is more difficult since, to the extent that the distance factor is already taken into account, one has to turn to other explanatory factors. As a first approach, some factors grab the attention immediately. One can see, for example, that among the highest-ranked departments, the majority are departments with very marked urban characteristics and that the best results for the trunk road network are often scored by departments that have the largest networks of fast four-lane dual carriageways (Fig. 4).

However, one may also notice that, among departments sharing similar characteristics, LAI values differ considerably, whether it is the overall LAI or LAIs by networks. With the distance factor neutralised, other factors have to be sought to interpret the differences. These variables have to be selected for their accidentinducing potential and have to be documented at the departmental level. In the current state of knowledge, the variables on which researchers are working are alcohol, seatbelts, roadside trees and shrubs, motorcycles, and the proportion of traffic on country roads.

3.3.3. Policy making

All speed measurements showed the extent to which drivers exceed speed limits in France. At the end of 2001, about 35% of motorists were found to drive at 10 km/h more than the speed limit. This highlighted what was at stake as regards speeding in France and enabled us to understand why the country was so far behind other comparable European countries (Delorme and Lassarre, 2005). In the light of these findings, traffic victims' associations called for tougher speed controls, which led the government to start testing automatic speed cameras in 2001–2002 and then to introduce them from 2003 onwards.

The installation of 1400 fixed cameras and the deployment of 700 mobile cameras brought a substantial reduction in speed: the average speed declined by 10 km/h in 2008, with 10% of drivers driving at 10 km/h more than the speed limit. As a result, we observed a sharp fall in the number of deaths on the road – by 48% between 2001 and 2008. This example shows that knowledge had a particularly powerful impact on policy.

According to the calculation of attributable risks, alcohol has now become the first factor of accidents in France, a fact that impressed the policy-makers. The analysis also underlies the main characteristics of drink driving (very small number of drivers but

Fig. 4. Map of the standardised rates of fatalities by department.

very high rate of alcohol) what oriented the propositions of the expert committee.

It's the same for the telephone when driving whose effect was before totally under-evaluated.

4. What governance?

In terms of governance, we view the role played by the Observatory as an interface between experts and decision-makers in the transmission of information of a scientific nature for the purpose of formulating road safety policy. Information is disseminated to the media through press conferences that are programmed as a function of the political agenda. The results of expert studies are debated in more broadly-based organisations, such as the National Scientific Research Centre (CNSR), where other interested parties participate, such as motor manufacturers, insurance companies, victim defence associations and particular categories of user. This led to the decision taken in 2000 to seek to improve the quality of the information gathered and of the risk analyses carried out to guide decision-makers. Two action plans are currently being implemented to improve this process of risk evaluation: data from the road accident victims register of the department of the Rhône are being used to provide more rigorous estimators of the numbers of injured according to seriousness; and research in road safety economics is being encouraged in order to reintroduce monetary profitability criteria into decision-making, criteria that could thus be integrated into a broader, long-term sustainable approach to the road system.

Additionally, the Observatory has developed a programme for evaluating the policies pursued with teams of researchers and technicians from the Ministry, and the resulting reports have been published – one on automatic speed cameras (ONISR, 2008) and one on drinking and driving (ONISR, 2007).

The experience of the past few years has shown the importance of developing simple but rigorously constructed methods and tools, providing telling results for decision-makers and the public.

Nevertheless, the situation is far from perfect. Firstly, these tools are perfectible, and bench-marking the methods promoted by the European Commission as part of projects like SafetyNet (Thomas et al., 2005) will help get them shared more widely.

In terms of risk selection, the Observatory does not cover all the issues, far from it. We are starting to invest in risk factors such as falling asleep at the wheel and taking medicines. The construction of risks is determined in part by networks of experts belonging to the different parties concerned, but will also depend on events, such as serious accidents, and on political and social considerations.

Furthermore, experience shows that, over and beyond the results obtained and the tools used, a permanent concern is getting decision-makers to assimilate and validate them. Road safety managers need to watch out increasingly for the legalisation of the organisational process of risk management in order to preserve their own legitimacy and reputation.

Acknowledgements

We thank Nicole Muhlrad (INRETS) for assisting us in improving the english expression.

References

Aarts, L., van Schagen, I., 2006. Driving speed and the risk of road crashes: a review. Accident Analysis and Prevention 38 (2), 215–224.
Amoros, E. et al., 2008. Actual Incidences of Road Casualties, and Their Injury Severity, Modelled from Police and Hospital Data. EJPH, France.
Bergel, R., Girard, B., Lassarre, S., Le Breton, P., 1994. Un modèle de désaisonnalisation des indicateurs de sécurité routière. In: Carré, J.R., Lassarre, S., Ramos, M. (Eds.), Modélisation de l'insécurité routière. Tome II. Actes du séminaire 1993. Paradigme, Caen.
Chapelon, J., 2006. Impact du contrôle sanction automatisé sur la sécurité routière (2003–2005), ONISR. <http://www.securiteroutiere.gouv.fr/cnsr/travaux\_cnsr. htm> (séance du 15 mars 2006).
Chapelon, J., Biecheler, M.B., Facy, F., Fontaine, H., Got, C., Perez-Diaz, C., Sibi, P., 2007. L'alcool sur la route: état des lieux et propositions, Comité des experts de la sécurité routière. <http://www.securiteroutiere.gouv.fr/cnsr/travaux\_cnsr.
htm> (séance du 9 juillet 2007). Cohen, S., Duval, H., Lassarre, S., Orfeuil, J.-P., 1998. Limitations de vitesse – Les décisions publiques et leurs effets. Hermès, Paris.
Christensen, C., Elvik, R., 2007. Effects on accidents of periodic motor vehicle inspection in Norway. Accident Analysis and Prevention 39 (1), 47–52.
Delorme, R., Lassarre, S. (Eds.), 2005. L'insécurité routière en France dans le miroir de la comparaison internationale. La comparaison entre la France et la Grande-Bretagne. Rapport no. 216, Les collections de l' INRETS, Bron, France.
Lassarre, S., Hoyau, P.-A., 2008. Evolution récente du risque routier en France et en Grande-Bretagne. In: Delorme, R., Lassarre, S. (Eds.), Les régimes français et
britannique de régulation du risque routier: la vitesse d'abord. Synthèse no. 57, Les collections de l' INRETS, Bron, France.
Leproust, S., Lagarde, E., Salmi, L.-R., 2008a. Systematic screening for unsafe driving due to medical conditions. BMC Public Health 8, 27.
Leproust, S., Lagarde, E., Salmi, L.-R., 2008b. Risks and advantages of detecting individuals unfit to drive: a Markov decision analysis. Journal of General Internal Medicine 23 (11), 1796–1803.
McEvoy, S.P., Stevenson, M.R., McCartt, A.T., Woodward, M., Haworth, C., Palamara, P., Cercarelli, R., 2005. Role of mobile phones in motor vehicle crashes resulting in hospital attendance: a case-cross over study. British Medical Journal 331, 428–430.
Nilsson, G., 1974. The effects of speed limits on traffic crashes in Sweden. In: Proceedings of the International Symposium on the Effects of Speed Limits on Traffic Crashes and Fuel Consumption, Dublin, Organisation for Economy Cooperation and Development (OECD), Paris.
OECD, 2008. Towards Zero: Ambitious Road Safety Targets and the Safe System Approach. Paris.
ONISR, 2009. La sécurité routière en France: bilan de l'année 2008. La Documentation Française, Paris.
Redelmeier, D., Tibshirani, R., 1997. Associations between cellular–telephone calls and motor vehicle collisions. The New England Journal of Medicine 336 (7), 453–458.
Thomas, P. et al., 2005. Designing the European Road Safety Observatory. International Journal of Injury Control and Safety Promotion 12 (4), 251–253.

chapelonroadsafetyfrancehard2010.md 51 KB Historie Surový