Evolution of road risk disparities at small-scale level: Example of Belgium Vojtech Eksler ⁎ , Sylvain Lassarre INRETS - GARIG (French National Institute for Transport and Safety Research, Group for the Analysis of Road Risk and its Governance), 23, rue Alfred Nobel - Cité Descartes, 77420 Champs sur Marne, France Available online 9 August 2008 Abstract Problem: Road accident outcomes are traditionally analyzed at state or road network level due to a lack of aggregated data and suitable analytical methods. The aim of this paper is to demonstrate usefulness of a simple spatiotemporal modeling of road accident outcomes at small-scale geographical level. Method: Small-area spatiotemporal Bayesian models commonly used in epidemiological studies reveal the existence of spatial correlation in accident data and provide a mechanism to quantify its effect. The models were run for Belgium data for the period 2000-2005. Two different scale levels and two different exposure variables were considered under Bayesian hierarchical models of annual accident and fatal injury counts. The use of the conditional autoregressive (CAR) formulation of area specific relative risk and trend terms leads to more distinctive patterns of risk and its evolution. The Pearson correlation tests for relative risk rates and temporal trends allows researchers to determine the development of risk disparities in time. Results: Analysis of spatial effects allowed the identification of clusters with similar risk outcomes pointing toward spatial structure in road accident outcomes and their background mechanisms. From the analysis of temporal trends, different developments in road accident and fatality rates in the three federated regions of Belgium came into light. Increasing spatial disparities in terms of fatal injury risk and decreasing spatial disparities in terms of accident risk with time were further identified. Impact on industry: The application of a space-time model to accident and fatal injury counts at a small-scale level in Belgium allowed identification of several areas with outstandingly high accident (injury) records. This could allow more efficient redistribution of resources and more efficient road safety management in Belgium. © 2008 National Safety Council and Elsevier Ltd. All rights reserved. Keywords: Road risk; Risk exposure; Bayes hierarchical model; Spatial correlation; Convolution model 1. Introduction The common practice of monitoring road accidents and their consequences is normally assessed by analyzing either national or road network-related accident risk data (see Flahaut, 2004; Noland & Quddus, 2004; Page, 2001). However, both scales have certain limitations. State-level analysis cannot unveil regional differences, or more generally, an existence of high-risk areas or road sections, which may result in rather inefficient road safety management. The analysis of network-related accident outcomes, on the other hand, allows identification of roads and those sections with outstandingly high numbers of accidents or accident injuries. Given the complexity and costs of such analysis, it is often limited to a certain number of main roads with a large proportion of road traffic; for these roads, the traffic volume is monitored, which is a prerequisite for performing any reliable risk analysis. An analysis performed at a small-scale geographical level has a number of advantages compared to the state and road network data analyses mentioned above. It allows identification of the geographical areas having outstandingly high/low accident occurrence and outcomes. It also points toward possible problems in road network settings and traffic management. Ideally, the analysis should provide a bridge between the two methods, allowing suggestions to be made on the identification of high-risk roads within the road network. The choice of spatial units for small-scale analyses has traditionally been dominated by national territorial disaggrega- tion and availability of statistical methods. Little effort has been made to identify the ideal disaggregation level for a spatial analysis of road accident outcomes, despite a common under- standing that lowering the geographical level of analysis is an important tool for better insight in regional risk differences Journal of Safety Research 39 (2008) 417 – 427 www.nsc.org ⁎ Corresponding author. E-mail addresses: eksler@inrets.fr (V. Eksler), lassarre@inrets.fr (S. Lassarre). www.elsevier.com/locate/jsr 0022-4375/$ - see front matter © 2008 National Safety Council and Elsevier Ltd. All rights reserved. doi:10.1016/j.jsr.2008.05.008