Severity analysis of red-light-running-related crashes using structural equation modeling

Abstract

The study investigated factors affecting the severity of red-light-running (RLR) -related crashes by means of crash records from the State of Florida (US). Previous studies have attempted to incorporate road, driver, and environmental factors into ordinal regression models to predict the severity of crashes. However, investigating the relationship among variables can become a difficult task with traditional statistical techniques. Some explanatory variables, in fact, may impact crash severity indirectly, through one or more mediating variables (measured or unobserved). This study employed structural equation modeling (SEM) to explore the indirect relationship between crash severity and its contributing factors. SEM is a statistical technique able to account for unobserved variables, whereas traditional severity models have focused on measured variables only. Three unobserved variables were proposed in this study to better describe the dynamic of RLR-related crashes: precrash travel speed (TS) of the bullet vehicle (at fault), the kinetic energy (KEs) applied from the bullet vehicle to the subject vehicle(s), and crash severity (overall energy resulting from a crash). A SEM model was developed to estimate the hypothesized relationships among speed, kinetic energy, and crash severity. Measured (observed) variables obtained from crash records were included in the SEM model to define latent dimensions. The results showed that crash data supported the model hypothesis and measured/unobserved variables adequately predicted crash severity. Overall, speed and kinetic energy were demonstrated to positively affect crash severity with kinetic energy being the more influential factor. Moreover, it was demonstrated that the most influential factor with respect to TS was alcohol-impaired driving conditions and the least influential was age of the at-fault driver. Regarding KEs, the number of vehicles involved in the crash showed the most influence and vehicle year was the least influential factor.