Bicyclists injury severities: An empirical assessment of temporal stability.

Cyclists are among the most vulnerable participants in road traffic, making their safety a top priority. Riding behavior of bicyclists could shift over time, affecting the level of injuries sustained in bicyclist-involved crashes. Many studies have been done to identify the factors influencing bicyclist injury severity, but the temporal stability of these variables over time needs further study. The temporal instability of components that affect the cyclist injury levels in bicycle collisions is explored in this paper. To obtain potential unobserved heterogeneity, yearly models of cyclist-injury levels (including potential consequences of no, minor, and severe injury) were measured separately applying a random parameters logit model that allows for potential heterogeneity in estimated parameters' means and variances. Employing a data source on bicycle collisions in Los Angeles, California, over the course of six years (January 1, 2012 to December 31, 2017), several variables which may impact the injury level of cyclists were explored. This paper has also employed a set of likelihood ratio tests assessing the temporal instability of the models. The temporal instability of the explanatory parameters has been evaluated with marginal effects. The results of the model assessment indicate that several factors may raise the chances of severe bicyclist injuries in collisions, including cyclists older than 55 years old, cyclists who were identified to be at-fault in crashes, rear-end collisions, cyclists who crossed into opposing lane before the collision, crashes occurring early mornings (i.e., 00:00 to 06:00) and so on. The results also showed that the details and estimated parameters of the model do not remain stable over the years, however the source of this instability is unclear. In addition, the findings of model estimation demonstrate that considering the heterogeneity in the random parameter means and variances will enhance the overall model fit. This study also emphasizes the significance of accounting for the transferability of estimated models and the temporal instability of parameters influencing the injury severity outcomes in order to dynamically examine the collected data and adjust safety regulations according to new observations.