Modeling crashes involving children, finite mixture cumulative link mixed model

Despite the efforts in the literature review on the traffic safety of children, the majority of past studies mainly focused only on the child's seatbelt status, or its position while ignoring other underlying factors that might contribute to the severity of those crashes. Inclusion of ther factors is especially important for a mountainous state like Wyoming with one of the highest rates of children's traffic fatality in the country. Thus, this study is conducted to fill the gap by identifying important factors contributing to the severity of crashes involving children. Here child is defined as any passengers under 9 years old. A first step in identifying factors to the severity of crashes involving children is implementing a reliable statistical method that could account for heterogeneity across various observations. So, in this study, to account for the heterogeneity in the dataset, the standard cumulative link model (CLM) was extended to the random effect model, while instead of assigning the subjective attribute for random effect, an objective hierarchy through the finite mixture modeling (FMM) was used. The FMM was employed in the context of the CLM to prevent the loss of information due to disaggregation of the dataset into the homogeneous datasets. The comparison results highlighted that the random effect model by the objective hierarchy would result in a significant improvement in the model fit compared with the standard cumulative link model. The results highlighted factors such as safety equipment in use, type of collision, and various drivers' characteristics and actions such as belting condition, alcohol and drug involvement are some of the factors contributing to the severity of child crashes. As expected, the main findings of our results highlighted that various drivers' actions and behaviors are the main causes that children would undergo a higher severity level in crashes. An extensive discussion regarding the implications of the results and the implemented statistical method were given in the context of the manuscript.

Automated summary

Despite previous studies mainly focusing on child seatbelt status, this research in Wyoming identified various factors contributing to the severity of crashes involving children. By extending the CLM to a random effect model with objective hierarchy through FMM, the study found significant improvements in model fit and highlighted the impact of safety equipment, collision type, and drivers' characteristics and actions on the severity of child crashes. Results emphasized that drivers' behaviors are the main causes of higher severity levels in child crashes. Extensive discussion on implications and statistical methods were provided in the manuscript.