Explanatory and prediction power of two macro models. An application to van-involved accidents in Spain

The figures representing road safety in Spain have substantially improved during the last decade. However, the severity indicators concerning vans have not improved as favorably as those of other types of vehicles, such as passenger cars and heavy freight transport vehicles. This study is intended to analyze the main factors explaining van accident behavior and to get a further insight into dynamic macro models for road accidents. For this purpose the authors are using four time series related to the frequency and severity of van accidents on Spanish roads and two types of methodologies applied in the study of traffic accidents: linear regression with Box–Cox transformed variables and autoregressive errors (DRAG), and an unobserved components model (UCM). The four response time series modeled are the number of fatal accidents, the number of accidents with seriously injured victims, the number of fatalities and the number of seriously injured victims. Since the choice of the appropriate macro model for the analysis of road traffic accidents is not a trivial matter, they are considering multiple factors such as goodness of fit and interpretation, as well as the prediction accuracy in order to choose the best model. Overall, the final results make sense and agree with the literature as far as the elasticities and coefficient signs are concerned. It was found that the DRAG-type model yields slightly better predictions for all four models compared to UCM. With these macroeconomic models, the effect of some influential factors (fleet, drivers, exposure variables, economic factors, as well as legislative actions) can be addressed. Estimating the effect of the vigilance and surveillance actions can help safety authorities in their policy evaluation and in the allocation of resources.

Language

  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 01521918
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Apr 2 2014 1:53PM