Impact of infrastructure and local environment on road unsafety: Logistic modeling with spatial autocorrelation
Introduction
Road safety constitutes an increasingly significant concern in most countries. Economic losses from accidents can reach up to 2.5% of GNP (Elvik, 2000). According to the World Health Organization, 600,000 people die and 15 million are injured in road accidents each year. In the European Union, a total of 1.2 million accidents occur every year leading to more than 40,000 fatalities. The figure is similar in the United States, though slightly less. This high number of deaths and casualties is increasingly considered unacceptable. Many countries have implemented policies aiming to drastically reduce the number of road fatalities. While the situation has improved in the past twenty years in most countries (Page, 2001), many efforts remain to be made.
The study applies to Belgium1 where some 1500 people are killed each year on the roads. This statistic corresponds to one of the highest risks in Europe. Hence, the objective of the federal government is to reduce this figure by 50% for the year 2010. Four sets of measures are considered: (i) legislation, (ii) education, (iii) enforcement and (iv) vehicle and infrastructure engineering.
This paper is focused on the last aspect: it examines the influence of infrastructure on road (un)safety. In a broader sense, the objective is to model the impact of characteristics of roads and local environmental conditions on road (un)safety. As spatial data are considered, spatial autocorrelation is integrated in the modeling process. This is important to avoid a model for which the significance of the explanatory variables is overestimated. However, there are not many explicit spatial models in the road accident analysis literature despite the actual spatial component of accidents occurrence; to our knowledge, spatial logistic models are even non-existent. We propose here a synthesis of the different approaches that exist to take into account spatial autocorrelation within logistic modeling.
The structure of the paper is as follows. The Section 2 consists of a contextual description of the objectives of the study, followed by a brief review of the literature on the impact of environmental structures on road safety. The study area is presented in Section 3. The explanatory variables are specified in Section 4. This is followed in the fourth section by a short description of the type of modeling used: logistic regression. The steps of the model construction are developed in Section 6 for regional roads and in Section 7 for freeways. Finally, we conclude in Section 8.
Section snippets
Background and objectives
A significant amount of research in international literature is devoted to describing and explaining the occurrence of road accidents. Centers of interest can extend from the legal and juridical point of view to the technical aspects of vehicles and infrastructures; other viewpoints are the psychological, behavioral and socio-economic components of road users. This paper is centered on the geographical aspects: with a view to sustainable development and regional planning, the aim here is to
The study area
Road unsafety is expressed here as belonging to a black zone: a binary variable takes the value 1 for the spatial units of black zones, 0 otherwise. The definition of spatial units depends to a large extent on the finest spatial aggregation level for which accident data are available. In Belgium, the basic spatial unit (BSU) is the hectometer of road. Black zones are localized for the numbered road network of the Belgian province of Walloon Brabant (Fig. 1), 460 km long. They are identified by
The explanatory variables
The aim is to identify environmental structures related to road safety and unsafety. As pointed out in Section 2, roadway geometrics and roadside features have already been discussed by several authors. Here, explanatory variables are organized in four groups: (i) road use characteristics, (ii) physical features of roads, (iii) land use characteristics and (iv) natural environmental features. They are summarized in Table 2.
Traffic exposure is expressed by the average daily number of vehicles
Logistic modeling
As specified in Section 3, the response variable takes the value 1 if a BSU belongs to a black zone, 0 otherwise. As it is a categorical variable, logistic regressions are used. Linear regression analysis is indeed not appropriate for a discrete outcome. The main reason is that this kind of variable can take only discrete values (0–1 in this case) while linear regression allows for a variation between −∞ and +∞. However, the basis of both form of modeling is the same: the aim is to evaluate an
Modeling for regional roads
This section is divided into three subsections. First, the procedure leading to the construction of a logistic model is described. However, this model does not take into account the spatial specificity of data analyzed. Next, spatial autocorrelation is integrated in the modeling process (Section 6.2); the purpose is to avoid an overestimation of the significance of explanatory variables. We propose a synthesis of the different approaches that exist to take into account spatial autocorrelation
Modeling for freeways
The same series of modeling steps is applied to freeways. The definition of several covariates is however slightly different. First, distances up to 300 m are considered for the transition variables (200 m for the roads) because of the longer range of interactions between places on expressways. Furthermore, the location of major junctions corresponds here to the location of freeway accesses (entries/exits). Finally, the daily traffic exposure is categorized into three classes (four classes for
Conclusions
The objective of this paper is to model one aspect of road unsafety: it is expressed as the belonging of BSUs to a black zone. Logistic regressions are used to evaluate the impact of road features and local environmental characteristics on the spatial concentration of road accidents, in other words on their spatial co-occurrence. Since only spatial variables are considered, this leads to a spatially oriented modeling. This is related to the approach privileged in this study: explanatory
Acknowledgements
This research was funded by the Belgian National Fund for Scientific Research. The author also wants to thank D. Antoine from the Walloon Ministry of Equipment and Transport (MET) for having provided data, the Professor I. Thomas (UCL) for her guidance through this work, and the anonymous referees for their useful comments.
References (52)
- et al.
An investigation into the relationships between area social characteristics and road accident casualties
Accid. Anal. Prev.
(1997) Using logistic regression to estimate the influence of accident factors on accident severity
Accid. Anal. Prev.
(2002)- et al.
Safety rounds in public environments: experience of a new tool for prevention of accidental injuries
Accid. Anal. Prev.
(1990) How much do road accidents cost the national economy?
Accid. Anal. Prev.
(2000)- et al.
Identifying black zones with a local spatial autocorrelation index and a kernel method. A comparative approach
Accid. Anal. Prev.
(2003) Urban development and road safety
Accid. Anal. Prev.
(1986)- et al.
Differences in causality factors for single and multi-vehicles crashes on two-lane roads
Accid. Anal. Prev.
(1999) - et al.
Explaining two-lane highway crash rates using land use and hourly exposure
Accid. Anal. Prev.
(2000) - et al.
Demographic and environmental correlates of pedestrian injury collisions: a spatial analysis
Accid. Anal. Prev.
(2000) - et al.
Impact of roadside features on the frequency and severity of run-off-roadway accidents: an empirical analysis
Accid. Anal. Prev.
(2002)