Quantifying the road network structure and its impact on road traffic crashes: A Bayesian CAR modelling approach

Abstract

Road traffic crashes (RTCs) are a major cause of fatalities worldwide. However, the influence of the road network structure on RTCs has not been adequately explored. Furthermore, methodologies employed in earlier studies to quantify road networks have often relied on visual inspection, which is both subjective and impractical. Therefore, this study aimed to address these gaps by (1) utilizing graph theory metrics to quantify the road network structure and (2) developing a statistical model to determine how various characteristics of the network structure—such as connectivity, density, complexity and centrality—are correlated with RTCs while accounting for over-dispersion and spatial auto-correlation. Using a Bayesian conditional auto-regressive model, a spatial analysis of fatal RTCs was conducted at the ward level in Delhi, India. The findings demonstrated a significant positive association between road network connectivity and fatal crash risk. Areas with a higher density of intersections involving major roads were linked to a greater number of fatal crashes. Furthermore, areas with a higher number of intersections deviating from the typical 90-degree angle (higher skewness) were associated with a higher incidence of fatal RTCs. Conversely, an efficient network structure (lower circuitry) and higher network centrality were negatively correlated with fatal RTCs. In addition, wards with a mix of higher-category and lower-category roads (increased entropy) faced an increased risk of fatal crashes. In summary, this study underscores the significant impact of network structure on road safety outcomes. Based on the findings, the study offers policy recommendations for developing targeted road safety measures to address the issues identified via network analysis.