Investigating the application of beta-binomial models in highway safety

Statistical regression models are widely used in highway safety for modeling motor vehicle crash data. They include the Poisson model, the Negative Binomial (NB) (Poisson-gamma) model, as well as their modified forms or extensions. Recently, a new type of statistical model, based on Beta-Binomial (BB) distribution, has been proposed by researchers as a substitute to the Poisson-gamma model. The objectives of this study are to describe the characteristics and development of BB and NB models for modeling motor vehicle crashes and to compare their statistical performance. In order to accomplish the objectives of this study, the models are first analyzed as a function of the crash data generating process. Then, given the theoretical formulation, the characteristics, including the probability density function (PDF), of each model are described accordingly. Using crash data collected in Toronto, Ontario, the statistical performance of both models is compared and evaluated. The study shows that BB models do not outperform Poisson-gamma models, and the difficulty in handling the PDF during the modeling process and incorporating the exposure make them less attractive. Finally, further research on the development and application of BB models are discussed, including their applicability when this model type is used in cases where the data are subjected to small sample size and low sample mean values.

Language

  • English

Media Info

  • Pagination: 14p
  • Monograph Title: Canadian Multidisciplinary Road Safety Conference XVII, June 3-6, 2007, Montreal, Quebec

Subject/Index Terms

Filing Info

  • Accession Number: 01386964
  • Record Type: Publication
  • Source Agency: ARRB
  • Files: ATRI
  • Created Date: Aug 22 2012 9:56PM