Dynamic Statistical Model Selection: Application to Traffic Accident Analysis in Spain

Four time series: number and severity of both van accidents and victims in Spanish roads have been estimated using two main types of methodologies which are applied in the study of traffic accidents: demand for road use, accidents and their gravity (DRAG) and unobserved components model (UCM). Each model was estimated using explanatory and intervention variables that were divided into 9 groups. The explanatory variables were included in the models depending on how significant their t-values were. The final models were chosen using fit statistics such as AIC and R2. Finally, predictions were obtained by both methodologies and the prediction errors were compared. None of the two methods was clearly better; it depended on which response variable was being predicted.