A cross-comparison of different techniques for modeling macro-level cyclist crashes. (April 2018)

• Record Type: Journal Article
• Authors: Guo, Yanyong; Osama, Ahmed; Sayed, Tarek
• Is Part Of: Accident analysis and prevention. Volume 113(2018)
• Page Start: 38
• Page End: 46
• Publication Date: 2018-04
• Subjects: Cyclist crashes -- Macro-level crash models -- Spatial effects -- Random parameters model -- Full Bayesian estimation; Accidents -- Prevention -- Periodicals; Accident Prevention -- Periodicals; Accidents -- Prévention -- Périodiques; 363.106
• DOI: 10.1016/j.aap.2018.01.015
• Languages: English
• ISSNs: 0001-4575

Abstract

Highlights: Approaches of modeling macro-level cyclist safety are compared and factors contributing to cyclist crashes are investigated. Poisson lognormal (PLN), random intercepts PLN, random parameters PLN, and spatial PLN models were developed. The SPLN model had the best goodness of fit, followed by the RPPLN, RIPLN, and PLN models, respectively. Cyclist crashes were positively associated with bike and vehicle exposure measures, households, and signal density. Cyclist crashes were negatively associated with average edge length, average zonal slope, and off-street bike links. Abstract: Despite the recognized benefits of cycling as a sustainable mode of transportation, cyclists are considered vulnerable road users and there are concerns about their safety. Therefore, it is essential to investigate the factors affecting cyclist safety. The goal of this study is to evaluate and compare different approaches of modeling macro-level cyclist safety as well as investigating factors that contribute to cyclist crashes using a comprehensive list of covariates. Data from 134 traffic analysis zones (TAZs) in the City of Vancouver were used to develop macro-level crash models (CM) incorporating variables related to actual traffic exposure, socio-economics, land use, built environment, and bike network. Four types of CMs were developed under a full Bayesian framework: Poisson lognormal model (PLN), random intercepts PLN model (RIPLN), random parameters PLN model (RPPLN), and spatial PLN model (SPLN). The SPLN model had the best goodness of fit, and the results highlighted the significant effects of spatial correlation. The models showed that the cyclist crashes were positively associated with bike and vehicle exposure measures, households, commercial area density, and signal density. On the other hand, negative associations were found between cyclist crashes and some bike network indicators such as average edge length, average zonal slope, and off-street bike links.