category: literaturenote citekey: ekslermeasuringunderstandingroadsafety2010 title: Measuring and understanding road safety performance at local territorial level authors: "Eksler, Vojtech" year: 2010 date: 2010-00-00 2010 doi: 10.1016/j.ssci.2009.12.010 publication: Safety Science url: "https://doi.org/10.1016/j.ssci.2009.12.010" zotero_key: T7F962GA zotero_storage: 3A2M2QKK collections: transpordi_planeerimine folder: 001_artiklid/Road Safety Indicators firstAuthor: "Eksler, Vojtech"

status: converted

Contents lists available at ScienceDirect

Safety Science

Measuring and understanding road safety performance at local territorial level Vojtech Eksler *

European Transport Safety Council (ETSC), 20 Avenue des Celtes, 1040 Bruxelles, Belgium

article info

Article history: Received 12 August 2009 Received in revised form 18 November 2009 Accepted 12 December 2009

Keywords: Local territorial level Bayes risk Serious injuries Safety performance

abstract

Road safety performance has traditionally been measured at national level, because the national authorities have assumed main responsibility for managing road safety of citizens. With the rise of EU governance, the role of national government has slightly decreased while the role of local authorities has been reinforced in many countries, creating a new playfield for policy actions. Assessing road safety performance at local territorial level may provide new inputs needed to trigger further improvements in road safety, as it brings about a higher accountability of policy makers and brings relevant issues closer to citizens. The measures and methods for such evaluation have now become broadly available and their application may bring a difference in current pace of road safety improvements. An example of the application of full Bayes spatio-temporal model on local road risk data is provided, illustrating the potential of local areal analysis for a better road safety management.

-2009 Elsevier Ltd. All rights reserved.

1. Introduction

Historically, cities and regions have played an essential role in the formation of national-states and national societies. Their inhabitants gave the sense to the notion of national identity, cultural norms and habits. The sense of social identity, citizen responsibility and respect of authorities which are acquired through active community life have then a great impact on the way how people obey the law. Since behaviour and compliance with traffic rules are the key determinants for the occurrence of road crashes, it is straightforward to seek how to improve compliance with law at local level, at which many behaviour-shaping mechanisms operate.

The European integration has contributed to a certain relaxation of the 'constraints of state' and cities and regions have emerged as one possible level for regulation of interests, groups and institutions. Several European countries have recently seen their governance systems decentralized. Decentralized governance systems tend to have stronger accountability mechanisms and lower corruption (Nas et al., 1986; Rose-Ackerman, 1999), which are the factors presumably related to compliance with road traffic law. Lederman et al. (2005) argue that decentralization leads to a higher accountability, since easier monitoring means less severe informational problems at local level. In countries with decentralized administrative power, employment of the least amount of special legal powers of the state compatible with the task to be performed, are sought (Hood and Schuppert, 1988) to bring power as close to people as possible. Moreover, local governments are directly responsive to their citizens needs but may be subject to capture by elites (Bardhan and Mookherjee, 2006).

A common feature of policy making in all so called SUN countries (Sweden, United Kingdom and the Netherlands, three countries with the best road safety record in the world) is a vital participation of civil society organized in associations and interest groups (Koornstra et al., 2002) and the proximity of policy making to citizens. These have a huge implication contributing to a higher accountability and responsiveness. It is indeed a common feature of road safety improvement strategies in countries with best road safety records not to rely on traditional topdown governance, in which actions at local level are governed exclusively from the national level to which they are subordinated. Specific arrangements and responsibility distributions could be tracked in SUN countries. While in the UK, the key areas of application were infrastructure improvements and traffic law enforcement including contracts between local police forces and local authorities on the number of checks of specific offences, in the Netherlands, the accountability was enhanced through the local transport plans, being regularly put under scrutiny by the central government. In Sweden, the local initiatives by civil society, companies and authorities were likely contributing to the improvements leading to its exceptional road safety record.

Alongside with the natural interest in local territories as areas of policy actions in road safety, there are several other motifs to put local territorial level at the centre of attention of contemporary road safety research. First, it can be claimed that European and national authorities have already used a great part of the potential for interventions. The EU has already passed legislation on driver licensing, properties of vehicles, and safety of infrastructure.

^* Tel.: +32 (0) 22 30 41 06. E-mail address: vojtech.eksler@etsc.eu

National authorities have brought extensive legislation regarding road user behaviour, including tools for enhancing compliance with law, such as penalty point system. Fields of actions targeting citizens at local level have, at the same time, been left unexploited until now.

Beside, territories have recently become the trial fields of numerous advances in road safety research, such as intelligent transport system applications (e.g. intelligent speed adaptation, alcohol interlocks, or the shared-space urban road design concept).

To allow assessment to be made on the impact of these innovative concepts on road safety, there is a need for evaluation tools and techniques, which are reliable and robust, but at the same time simple and understandable. This contradiction is most likely the reason why assessment of road safety level is so rarely done at local level. However, exceptions exist and several observation and evidence based studies could be identified in the literature. Their authors realized that at specific cases, it is indeed more feasible to somewhat restrain the geographical area for running statistical analysis. For example Servadei et al. (2003) assessed the effect of Italy's motorcycle helmet law on brain injuries performed at Romagna region and Farchi et al. (2007) assessed the effectiveness of the new penalty point system in Italy in Lazio region. Similarly, there are studies showing the evidence of differences in road users' behaviour at sub-national level. For example, there are significant intra-national differences in terms of the usage rate of restraint systems between Italian (Taggi et al., 2006) or Swiss regions (Siegrist et al., 2006). Similarly, the prevalence of driving while intoxicated varies between regions of France, Italy or Spain.

To summarize, there is a growing interest in local territorial level as the area of various policy actions. There is also a great potential for road safety improvements through local policy actions, which should not be ignored. The thorough understanding of mechanisms and the availability of tools for the statistical analysis are the main prerequisites for allowing road safety practitioners to evaluate the effects of various policy measures applied at local level and give appropriate feedback to policy makers on the effectiveness of their actions.

2. Framework for the analysis

There are numerous consideration issues when it comes to assessing road safety performance at local territorial level. They include among else the choice of spatial unit, time frame, indicators and relevant statistical method, but all of these are driven by the general framework for the analysis.

The assessment of road safety at local territorial level is of a particular interest where the main areas of policy actions are likely corresponding. The administrative areas of a country are to be considered first, as the areas of general actions, spending powers and targeted policy actions. As second, the administrative area of police forces is to be considered, as police enforcement is one of the most efficient road safety interventions. Last but not least, the invisible borders of cultural and normative background should be taken into account (Wegman et al., 2008).

Ideally, these would correspond, allowing a non-distorted picture about the effectiveness of policies to be made on the background of road safety outcomes analysis. Recently, benchmarking as an evaluation tool has gained popularity also in road safety. The benchmarking as an integrated part of a continuous improvement plan should be viewed as a set of activities designed to bring gradual, but continual improvement to a process through constant review. It can be understood as an implementation tool of the Shewhart's cycle consisting of plan, do, check, act (PDCA), which is perhaps the best known theoretical framework for improvements (Shewhart, 1939). Thus the analysis is the core of the improvement process.

The quantitative analysis of crash outcomes as the measure of road safety performance is thus the prerequisite for managerial approach in road safety. As lord Kelvin put it: ''I often say that when you can measure what you are speaking about, and express it in numbers, you know something about it; but when you cannot express it in numbers, your knowledge is of a meagre and unsatisfactory kind; it may be the beginning of knowledge, but you have scarcely, in your thoughts, advanced to the stage of science, whatever the matter may be" (Stellman, 1998).

The evaluation of road safety performance could be accomplished by analyzing the level and trend in road crash outcomes standardized per exposure. Crash outcomes traditionally considered by road safety practitioners are road fatalities. Road injuries are often missing, as the statistics suffer from incompleteness, due to underreporting and data handling deficiencies. The use of health system based data combined with police reports may overcome this weakness. While at national level, health care systems could provide data necessary for the estimation of indicators regarding injury; in some countries the emergency department is the most common source especially for slight injuries, while in others, there are general practitioners or primary care centres (Gio rgi Rossi et al., 2005 or MacKay, 2003). While in case of fatal and serious injuries, the local registrations are forwarded to central registry, this is not always the case for other types of injury data. But road injury data are available at registry of all emergency medical care centres operating under given jurisdiction in a given territory. To that end, it can be argued that at local level, road injury data are easier to collect and link with road crashes. This is also the main reason why studies on underreporting of road crash injuries are usually carried out at local territorial level, such as regions (Amoros et al., 2003). Thus ideally, the use will be made of both fatal and serious injury counts at local level. As regards exposure indicator, vehicular travel is preferred over a static use of population of vehicle registry data.

It is a common practice to benchmark road safety performance of countries, despite obvious difficulties with the interpretation of differences identified. While a substantial part of the gap between any two countries can be explained by differences attributable to the application of different road safety policies, there is always a part of the gap caused by structural differences, such as road network composition, climate, urbanization, climate, culture and many others. Similarly, differences in the development of risk indicators at national level can be partly attributable to changes in mobility patterns, traffic performance and modal split.

At local territorial level, many of these factors do not have to be taken anymore in consideration when searching for the explanation of various road safety performances (Wegman et al., 2008). Factors such as climate, culture or norms could be often considered the same for sub-national entities. At the same time, however, some physical structural factors could come to play a more important role, but they can be easily accounted for by a simple standardization. For example, Eksler and Lassarre (2008) identified the population density and urbanization as the single most important synthetic factor suitable for standardizing risk rates at sub-national level, when a comparison of road safety levels is made. The trend in road safety indicators is likely less sensitive to certain structural factors, despite some evidence on the effect of urbanization on amenability to road safety intervention. Benchmark on the trend of the risk thus has a higher explanatory value compared to the benchmark made at national level, where the availability of effective interventions can make a crucial difference in the performance over the time.

3. Analytical methods

Injuries from road crashes had originally been understood as a part of public health problem. But due to their relatively limited weight on health system at beginning of 20th century, they were often marginalized and eventually became forgotten part of public health research and initiatives (Porter, 1999). As a consequence, modern statistical techniques developed in epidemiology were rarely applied in road safety research. More particularly, the methods for statistical analysis in epidemiology have developed considerably during the past century, including changes in basic concepts, methods of data analysis, and methods of exposure measurement (Pearce, 1996). For example, since early 1990s, we have been witnessing a broad application of Bayesian analysis in descriptive epidemiology contributing to major advances in ecological research.

Such a gradual development in analytical methods has not always been a case in road safety research, where the basic concepts and methods have undergone a rather limited development. The use of modern statistical methods such as the use of Bayesian statistical techniques have been limited to the analysis of road crash data on road stretches rather than to general data on territorial level (see e.g. Heydecker and Wu, 2001 or Qin et al., 2005).

When analyzing various road safety outcomes in terms of counts or rates, different statistical approaches must be considered for different spatial levels of analysis. More particularly, when analyzing scare events at local level, the rough rates are not more appropriate as the counts suffer from the randomness. For the difference in road deaths counts to be statistically significant, a large difference is needed. For example, if there were 20 deaths in one year, there must be maximum 12 deaths in another for the observed reduction to be statistically significant. This only demonstrates an obvious difficulty in working with crash outcomes of lower frequency such as road fatalities and trying to figure out the differences in their occurrence in time and space. This is particularly true for the analysis at sub-national level. Several spatial levels are usually considered in road safety analysis, from continent to the tiny communities. While the numbers analysed at the local level can be too small and subjected to random variations to produce reliable rates, the numbers at aggregated spatial level are usually enough stable and representative to be analysed directly as crude rates. Analysis of road risk rates at different levels of spatial aggregation may therefore require different statistical approaches and methods.

While at the level of EU and its member states the use of crude rates appears satisfactory, at the level of small areal units, the use of relative rates estimated via Bayesian methods reveals to be utterly indispensable. Similarly, where the traditional methods encompassing crude rates are appropriate for the analysis done at continental and country level, the analysis performed at regional or community level necessitates special statistical methods to be applied.

Similarly, different road safety outcomes should be considered at local territorial level, to overcome the problem with the scarcity of crashes and their outcomes. Counts of serious injuries on roads are the obvious additional measures beside traditionally used counts of road deaths (see Table 1).

Table 1 Type of rates and outcomes for the assessment of road safety at local level.

Level	Rates	Outcomes
EU	Crude	Deaths (KSI)
Countries	Crude	Deaths (KSI)
Regions	Bayes	KSI
Municipalities	Bayes	KSI

Applying absolute figures of fatalities or other road safety outcomes and dividing them by a chosen exposure is the simplest and most commonly used method for comparing road safety performance of countries. Epidemiologists often insist in the necessity to account for the demography by applying Standardized Mortality Ratio (SMR), but this approach is rarely used in road safety analysis. Epidemiologists insists that standardization is necessary when comparing several populations that differ with respect to age (and/ or gender), because age and gender have such a powerful influence on health risk. Despite the recent evidence of some epidemiological studies that such indirect standardization is inappropriate for small area comparisons (not necessarily in terms of area, or population size), since SMRs for different geographic areas have different denominators (Julious et al., 2001), this is not a major obstacle for an analysis that includes regions, because fairly similar geographic units are usually subject to assessment.

The risk faced by people in road traffic is not equal for men and women, and the degree of risk also varies significantly with age (Massie et al., 1994; Evans, 2001). Given the demographical differences between sub-national units, it is straightforward to apply the sex–age standardization to road injury rates. The following formula is then used to calculate the SMR of a particular geographical unit (country, region):

$$SMR_i = \frac{Y_i}{Y_i^*}$$

where Yi is the observed number of road injuries in the unit i; Yⁱ is the expected number of road injuries in the same unit i.

$$Yi^* = \sum{jk} \tau{jk} \cdot N{ijk}$$

where sjk, is the standardized injury rate of the age–sex group j, k; Nijk is the population count of this group in the geographical unit i.

Comprehensive set of analyses unveiled that the demography accounts for up to 10% of the variance in road mortality registered within regions of some EU countries. The effect of population structure is usually insignificant and seems to be more relevant in larger countries such as Germany or Spain (Eksler, 2007).

The concept of standardized ratios could be applied for all types of injury and indeed combined with different measures of exposure. However it has some limitations as the events under consideration are scarce and subjected to randomness. Given that the variance in rates is proportional to the square root of the mean, there will be high sampling variability for the small area units. To overcome this variability one can carry out smoothing of the raw rates via hierarchical modelling. This approach can be justified in a likelihood random effects context, or via a minimum mean squared error argument and it is now exclusively applied from a fully Bayesian perspective. The approach is more flexible in adapting to each unique situation, more efficient in using all available evidence and more useful in providing relevant quantitative arguments, than traditional methods (Spiegelhalter et al., 2004).

Crash injuries, where the crash injury count is denoted Yi, expected number of injuries kⁱ and the exposure (number of inhabitants or traffic volume in MVKMs) Ei are modelled assuming the data to be Poisson distributed:

$$Y_i \sim Poisson(\lambda_i)$$

This model can be rewritten in terms of rates rather than counts as:

$$Y_i \sim Poisson(\theta_i E_i)$$

where hⁱ is the crash injury rate in area i and Ei is the exposure in the area i, here either the number of registered population of area i, or the estimated number of kilometers travelled by motorized vehicles within the area i. The kⁱ = hiEi and the log(Ei) are treated as known offset terms. The maximum likelihood estimates hⁱ = Yi/Ei of this classical model assumed in many disease mapping studies correspond to standardized ratios. These ratios would be highly unstable due to the use of sparse character of the counts modelled. In the generalized linear model the hⁱ for each area is a function of covariates and parameters indicating the association between the covariates and the rate. This model can be extended to form a generalized mixed linear model incorporating a random effect within hi. The modelling is implemented for counts (Yi), assuming that the response Yi's given a vector m of random effects are conditionally independent.

$$\lambda_i | v \sim \text{Poisson}(\theta_i^v E_i)$$

hm ⁱ represents the expected rate of Yi/Ei given all the random effects. The log risk is modelled as:

$$\log(\theta_i^{\nu}) = \alpha + \boldsymbol{X}_i \boldsymbol{\beta} + \nu_i$$

where a is a fixed effect, Xⁱ represents a vector of explanatory variables (if considering an ecological model), b a vector of fixed effect parameters, mⁱ is the unstructured heterogeneity (error) known also as random effects, therefore, this kind of model is commonly known as a mixture model where fixed and random effects are combined (mixed) together. The exp(mi)'s represent local area crash (injury) ratios adjusted for the explanatory variables. If mⁱ is drawn from the zero mean normal distribution m = N(0, r² ), where r² is the variance parameter, the model takes on a pretty standard mixed form, with only r having to be estimated during the model fitting procedure. A prior gamma distribution Gamma (0.5, 0.0005) is assumed for the precision equal to the inverse variance.

The application of the full Bayes modelling can be demonstrated by the example of a Bayesian analysis of road fatality risk at municipality level in Belgium over the period 2001–2006 (Eksler and Lassarre, 2008). Fatalities per vehicular travel in terms of motor vehicle kilometres were used here as the measure of road safety performance. The full Bayes spatio-temporal model has been run, allowing to account for the structure in underlying crash mechanisms distribution, resulting in a smoothed representation of fatal road injury risk distribution across the country. The model had a following form:

$$\log(\theta{ij}^{v}) = \alpha + \beta \cdot X{ij} + v_i + (\varphi + \delta_i)t_j$$

where Yij is the observed number of road fatalities for the ith area, (i = 1, ..., N), and the jth year, (j = 1, ..., T), a the constant term, xij the covariate for the ith area, in the jth year, b the regression coefficient, mⁱ the total heterogeneity of the intercept (consisting of Vi the unstructured heterogeneity and Ui the spatially structured heterogeneity), u the mean linear time trend over all areas, tj the year j, and dⁱ spatial structured heterogeneity of the time trend.

It is possible to map the Bayes relative risk as the exponential of various random effects. The temporal trend estimate expressed as the 1 exp(di) could be mapped to provide an idea about the distribution of the trend in fatality risk across the country. The area specific time effect consists here again of the unstructured and structured component. In Fig. 1, the structured component is mapped only, as to highlight the underlying structure in road risk trend. When using the estimated vehicular travel as the exposure of risk, the effect of safe agglomeration when measured by road mortality disappears and the single spots are no more visible on the map. The nests of municipalities with similar trends appear, in most cases following the border of hierarchically superior regions. The latter allow a judgement to be made about the effectiveness of regional road safety policies in the country. The gradient appearing in the map of the trend in fatality risk points to the

Fig. 1. Structured part of Bayes Relative Risks (BRR) and of the relative temporal trend estimates for Belgian municipalities (2000–2005).

north–south division of country in terms of reduction, which was in general very high in the Northern part of the country (Flanders) and much smaller in the Southern part of the country (Wallonia). This may point to different effectiveness of road safety policies in the two regions.

When aiming at the comparison of risk levels, or providing full understanding on the distribution of risk and its trends, a direct standardization of risk rates could be useful. It is possible to standardize by a large number of variables depending on the exposure considered and explanation sought. For example, when analyzing mortality rates, population density could be used as suggested by numerous authors (e.g. Orselli, 2003; Baker et al., 1987 and Chapelon 2002). These authors identified an almost log linear relationship between the road mortality and population density at the level of countries and regions. The direct standardization by population density reduces the variance of mortality rates significantly. Eksler (2009) estimated that for different EU countries, on average 60% of the variance of road mortality comes from the sub-national differences in population density. The remaining difficulty is to correctly interpret rates standardized by explanatory factors such as population density and provide relevant recommendations for policy making, since some factors remains abstract and not directly linked with measures at the scope of policy makers.

4. Discussion and conclusions

There are two basic reasons for comparing safety performance of sub-national jurisdictions. In the first place, a ranking of relative performance of each are very useful for comparison within countries. To ensure a meaningful benchmarking process, factors that could enable safety practitioners to achieve similar safety improvements need to be included within the ranking process. This might be done by developing a model of safety outcomes that includes these factors. The second reason is to provide better understanding of the factors affecting safety improvement, so that safety practitioners can design more effective programmes. This requires greater focus on understanding how the effects of programmes are modified by the nature of the safety problems faced by each area. Lessons can not only be learned from comparison between areas within countries, but also from comparison between similar areas in different countries. This can probably be more effectively achieved by case studies comparing small groups of areas, than by attempting to model a wide range of areas. Evaluation of road safety performance at local territorial level requires an application of good measures of safety performance and of appropriate analytical tools. Most notably, the analysis should be based on fatal and serious injury figures and the application of robust statistical techniques shall be assured.

The methods described in this paper are purely descriptive and cannot fully explain distribution of risk and of its trend at local territorial level. A direct standardization by explanatory factors may be necessary, but the choice of appropriate factors remains a difficult task. Analyses making use of highly sophisticated statistical techniques could bring new impetus to the work of road safety practitioners and policymakers, but their impact could be limited by the fact that the results of statistical modelling could be difficult to communicate to general public.

The analysis of local area road safety data should be viewed as a complementary approach to local road crash data analysis. Recent advances and increased popularity of modern geographic information system applications made it possible to regularly analyse crash data in respect to road network leading to the identification of high risk sections. Such analysis is a common part of network safety management and a crucial tool for road administrators. Its benefits are concentrated around infrastructure improvements, while the benefits of local area analysis could potentially point to other target areas for safety improvements.

When benchmarking the performance of local entities, the trend is preferred above the level, which is likely more sensitive to various structural factors attributable to local territories. The benchmark of the trend (progress in improving road safety) is also more appropriate measure in respect to policymakers' accountability. Benchmarking road safety performance at local territorial level has a great potential for road safety improvements. The methods and tools for such benchmarking have now become available and road safety authorities shall be encouraged to apply them alongside with traditional approaches.

References

Amoros, E., Martin, J.L., Laumon, B., 2003. Comparison of road crashes incidence and severity between some French counties. Accident Anal. Prev. 35 (4), 537–547. Baker, S., Whitfield, R.A., O'Neill, B., 1987. Geographic variations in mortality from

motor vehicle crashes. New Engl. J. Med. 316, 1384–1387. Bardhan, Pranab, Mookherjee, Dilip, 2006. Decentralisation and accountability in

infrastructure delivery in developing countries. Econ. J. 116 (508), 101–127 (January).

Chapelon, J., 2002. Un nouvel indicateur d'accidentologie locale. In: Courier des statistiques, vol. 103. September 2002.

Eksler, V., 2007. Road mortality in Europe: how sensitive is it to demographic structure and population dynamics? IATSS Res. 31 (1), 80–88.

Eksler, V., 2009. Road mortality in Europe: a regional approach, PhD thesis, University of Versailles St-Quentin, Versailles, France.

Eksler, V., Lassarre, S., 2008. Evolution of road risk disparities at small-scale level: example of Belgium. J. Safety Res. 39 (4), 417–427.

Evans, L., 2001. Gender and age influence on fatality risk from the same physical impact determined using two-car crashes. Soc. Automot. Eng.

Farchi, S., Chini, F., Rossi, P.G., Camilloni, L., Borgia, P., Guasticchi, G., 2007. Evaluation of the health effects of the new driving penalty point system in the Lazio Region, Italy, 2001–2004. Injury Prevent. 13, 60–64. doi:10.1136/ ip.2006.012260.

Giorgi Rossi, P., Farchi, S., Chini, F., Camilloni, L., Borgia, P., Guasticchi, G., 2005. Road Traffic Injuries in Lazio, Italy: a descriptive analysis from an emergency roombased surveillance system. Ann. Emerg. Med. 42, 152–157.

Heydecker, B.G., Wu, J., 2001. Identification of sites for road accident remedial work by Bayesian statistical methods: an example of uncertain inference. Adv. Eng. Soft. 32 (10–11), 859–869.

Hood, C., Schuppert, G.F. (Eds.), 1988. Delivering Public Services in Western Europe, Beverly Hills and London. SAGE Publications, London. 268 p.

Julious, S.A., Nicholl, J., George, S., 2001. Why do we continue to use standardized mortality ratios for small area comparisons? J. Public Health Med. 23, 40–46. Koornstra, M., Lynam, D., Nilsson, G., Noordzij, P., Petterson, H.E., Wegman, F.M., Wouters, P., 2002. SUNflower: a Comparative Study of the Development of Road Safety in Sweden, the United Kingdom, and the Netherlands. SWOV,

Liedschendam. Lederman, D., Loayza, N.V., Soares, R., 2005. Accountability and corruption: political institutions matters. Econo. Polit. 17 (1).

MacKay, M., 2003. National differences in European mass accident data bases. In: Gennarelli, T.A., Wodzin, E. (Eds.), Contemporary Injury Severity and Outcome Issues. a Joint Programme of the AAAM and IRCOBI 24 September. IRCOBI, Born France. pp. 51–55.

Massie, D.L., Campbell, K.L., Williams, A.F., 1994. Traffic accident involvement rates by driver age and gender. Accident Anal. Prev. 27, 73–87.

Nas, T., Price, A., Weber, C., 1986. A policy-oriented theory of corruption. Am. Polit. Sci. Rev. 80, 107–119.

Orselli, J., 2003. L'analyse statistique des variations spatio-temporelles des accidents de la route, Conseil générale des Pont et Chaussées, No. 2002-0153- 01.

Pearce, N., 1996. Traditional epidemiology, modern epidemiology, and public health. Am. J. Public Health 86 (5), 678–683.

Porter, D., 1999. Health, Civilization, and the State: a History of Public Health from Ancient to Modern Times. Routledge, New York. vi + 376 pp.

Qin, X., Ivan, J.N., Ravishanken, N., Liu, J., 2005. Hierarchical bayesian estimation of safety performance functions for two-lane highways using Markov Chain Monte Carlo modeling. J. Transp. Eng. 131 (5), 345–351.

Rose-Ackerman, S., 1999. Corruption and Government: Causes, Consequences, and Reform. Cambridge University Press, Cambridge. 290 p.

Servadei, F., Begliomini, C., Gardini, E., Giustini, M., Taggi, F., Kraus, J., 2003. Effect of Italy's motorcycle helmet law on traumatic brain injuries. Inj. Prev. 9, 257–260. doi:10.1136/ip.9.3.257.

Shewhart, W.A., 1939. Statistical Method from the Viewpoint of Quality Control. Dover, New York.

Siegrist, S., Allenbach, R., Cavegn, M., Nieman, S. And Ackermann, Y., 2006. SINUS report 2006, Sicherheitsniveau und Unfallgeschehen im Strassenverkehr, 2005. Schweizerische Berungsstell für Unfallverhütun BfU, Bern.

Spiegelhalter, D., Thomas, A., Best, N., Lunn, D., 2004. WinBUGS User Manual Version 2.0. MRC Biostatistics Unit, Cambridge, UK.
Stellman, J.M., 1998. Lecture on ''Electrical Units of Measurement" (3 May 1883), published in Popular Lectures vol. 1, p. 73, quoted in Encyclopaedia of Occupational Health and Safety (1998) by Jeanne Mager Stellman, p. 1973.
Taggi, F., Dosi, G., Giustini, M., Crenca, A., Cedri, C., Fondi, G., Iascone, P., Marturano, P. 2006. The Ulisse System for the monitoring and the use of the safety belts and
the helmets in Italy (2000–2005). Rapporti ISTIAN 06/39. Istituto Superiore di Sanita, In Italian.
Wegman, F., Commandeur, J., Doveh, E., Eksler, V., Gitelman, V., Hakker, S., Lynam, D., Oppe, S., 2008. SUNflowerNext: towards a Composite Road Safety Performance Index. SWOV Institute for Road Safety Research, Leidschendam, the Netherlands.

ekslermeasuringunderstandingroadsafety2010.md 35 KB 文件历史 原始文件