A hierarchical approach to analyzing pedestrian crashes in an urban city

Abstract

Introduction: Despite the well-documented benefits of walking, pedestrians are among the most vulnerable road users. This problem requires urgent attention, particularly in urban areas where walking is increasingly promoted as a safe and attractive mode of transportation. To address this problem, a hierarchical approach was developed to analyze pedestrian crashes in an urban environment at the macro-, meso-, and micro-levels, based on a crowdsourced dataset from Hong Kong. Methods: To quantify the safety challenges faced by pedestrians, a Bayesian Poisson state-space model was developed first to evaluate the relationship between longitude and pedestrian traffic injuries. The quasi-induced exposure method was then used to measure the annual relative risk of crash involvement when walking. Based on an officially published household travel-diary survey, the danger of walking was further measured as the rates of fatality and injury per minutes walked. The data were then aggregated to 209 tertiary planning units, and a neighborhood-level model using a novel Bayesian spatially varying coefficients approach was developed to quantify the effects of various area-wide factors on the frequency of pedestrian–motor vehicle (PMV) crashes. Models with population, walking trips, walking time, and walking distance as the measure of pedestrian exposure were estimated and compared. Finally, a micro-level analysis was conducted to identify the factors that contributed to the frequency of PMV crashes at 262 sampled traffic-signalized intersections, based on a novel Bayesian measurement-errors model that allowed adjustment of uncertainties in the vehicle and pedestrian volumes. Results: Hong Kong has witnessed considerable decreases in the numbers of pedestrian fatalities and injuries over the past two decades, with an average annual decrease rate of 2.28% (95% CI: 0.35% to 4.90%) and 2.19 (95% CI: 0.66% to 3.74%), respectively. The crash risk for pedestrians initially increased during 1998–2008 and then decreased sharply. Although pedestrians were more likely to be fatally injured than motorists per minutes traveled, their injury rate was significantly lower than those of cyclists, motorcyclists, private car drivers and passengers, and taxi passengers. Eight variables were found to have significant associations with the frequency of PMV crashes at the meso-level. Pedestrian activities, vehicle kilometers traveled, road density, intersection density, bus stop density, and the number of parking lots showed positive associations with the PMV crash frequency, whereas the percentage of motorways and median monthly income had negative effects. Likewise, after controlling for pedestrian and traffic volumes, the presence of curbside parking and the presence of ground-floor shops showed positive associations with the risk of PMV crashes at traffic-signalized intersections, whereas the presence of playgrounds and the presence of pedestrian signals had negative effects. Our empirical analysis demonstrates the presence of measurement errors in pedestrian and vehicle volumes. The reduced goodness-of-fit and biased parameter-estimations due to the absence of proper exposure measures from PMV crash-frequency models was also verified. Conclusion: This thesis provides a deeper understanding of the environmental determinants of PMV crashes. Based on the proposed hierarchical approach, multifaceted countermeasures are suggested to improve the mobility and safety of pedestrians.