Random effects selection in Mixed GLM models: applications to the modelling of some road accident severities
The main objective of our work is to contribute to the construction of statistical methods (modelling and estimation) of the parameters involved in the evaluation and understanding of the severity of road accidents. Many works on the subject concern mainly the logistic or multinomial regression model with fixed parameters, however the occurrence of an accident being a multi-factorial phenomenon and the impact of certain factors on the degree of severity of the injuries (age, sex, physiological factors, etc.) can lead to a certain unobservable heterogeneity between the different individuals of the same vehicle.
This heterogeneity, when not taken into account, can create biases and distort the interpretation of the results. We propose to model the severity of injuries during a crash using new methods to better understand the problem which is now the leading cause of death for young people (WHO 2013). Road traffic injuries are a social, economic and political issue. However, different factors play a determining role in many accidents, such as the type of roads used by users, the different types of vehicles involved in the accident, the type of collision, etc. All these factors are complex to study, and the results of the study should be used as a basis for further research.
All these factors are complex to study, so the use of GLMM (Generalized Linear Mixed Models) and its applications, efficient numerical methods are indispensable if we want to be able to better face this problem, while taking into account the unobservable heterogeneity between different individuals of the same vehicle and to efficiently predict the severity of the injuries.