An evaluation of pedestrian crash risk factors at urban intersections in a developing country: Comparing the classification accuracy of methods accounting for unobserved heterogeneity

Abstract

Pedestrian safety has always been a concern at urban intersections, especially in low-income developing countries with higher casualty rates. As one of the cities with the highest pedestrian fatality rates in Iran, Mashhad lacks studies that pinpoint the causes of these crashes. The choice of appropriate methodology was guided by the two-fold objective of the study: first, disaggregating crashes into homogeneous clusters; and second, examining the effects of risk factors on pedestrian crashes while accounting for the inherent unobserved heterogeneity in crash data. The study compared the classification accuracy of modeling approaches using receiver operating characteristic analysis. By analyzing three years (2015–2017) of pedestrian crashes in Mashhad, this study identified risk factors associated with higher severity of vehicle–pedestrian crashes at intersections. The results show that models incorporating the heterogeneity effect, such as the cluster-aggregated model and the random parameter model, have higher classification accuracy for crashes than models that do not consider heterogeneity. Based on the risk factors associated with increasing fatal crashes, several low-budget and immediate countermeasures are suggested in the hope of improving pedestrian safety at intersections.

Keywords

• Pedestrian safety
• Crash severity
• Unobserved heterogeneity
• Latent class cluster analysis
• Mixed logit model