Mixed logit model and classification tree to investigate cyclists crash severity

Growing concerns about emissions, urban traffic congestion, and the promotion of an active lifestyle are inducing more people to choose bike for their daily commute. The increase in bike usage underscores the need for improving the cyclist’s safety. Our study examined the 72 363 cyclist crashes that occurred in Great Britain in the period 2016-2019 with the objective of (1) examining how various factors influence cyclist crash severity, (2) identifying complex interactions among these crash patterns, and (3) proposing countermeasures aimed at solving the identified risk factors. To achieve these goals, a Classification Tree (CT) model was used as an exploratory tool to detect patterns and interactions that may not have been hypothesized a priori and an econometric approach, such as Mixed Logit Model (MLM), was used to quantify global effects and test the interactions identified by the CT and all the explanatory variables within a statistically rigorous framework. Specifically, six interaction variables were identified from the CT terminal nodes with the highest probability of fatal crashes by tracing back their pathways to the root node. These interactions were then included as additional explanatory variables in the MLM to guarantee that all risk factors were tested within a unified statistical framework. Interestingly, all the interactions were statistically significant. Thus, the CT model is explicitly used as a supporting tool to identify potential interactions, while conclusions are extracted from the MLM results. Based on the identified risk factors, a set of targeted safety countermeasures has been proposed to minimize cyclist crash severity and improve overall road safety.