HIERACHICAL BAYESIAN MODELS FOR ROAD ACCIDENT DATA

This paper is concerned with models for accident numbers at sites within a road network. It highlights the need to analyse data in a disaggregated form, rather than considering total accidents at a site. Obviously it is easier to analyse the data in a more aggregated form, but this approach can overlook important aspects of the underlying phenomena and misleading conclusions may be drawn from an aggregated analysis. The data analysed in this paper concern 156 single-carriageway link sites in Kent. The models proposed are multiple response models, where different types of accident (according to severity and the number of vehicles involved) are modelled simultaneously. Two quite similar classes of model are investigated. Both classes can be described as mixed generalised linear models, or hierarchical models, with the mean of each response variable equal to the product of a random effect and a regression term based on important explanatory variables such as speed limit, link length and estimated traffic. For one class the random effect is gamma distributed and for the other class it is a log-Normal random effect. The models are hierachical Bayesian, structuring the unknown parameters in different layers. In this way a better stochastic approximation is proposed of the causal machinery producing the phenomenon analysed. Inference is based on empirical quantities describing the posterior distributions of the parameters of interest. The multi-dimensional integrals involved in the analysis are approximated using Markov Chain Monte Carlo methods. (A)

• Availability:
  • Find a library where document is available. Order URL: http://worldcat.org/issn/00410683
• Corporate Authors:

  PRINTERHALL LIMITED

  32 VAUXHALL BRIDGE ROAD
  LONDON,   United Kingdom  SW1V 2SS
• Authors:
  • TUNARU, R
• Publication Date: 1999-6

Language

• English

Media Info

• Features: References;
• Pagination: p. 318-24
• Serial:
  • Traffic Engineering & Control
  • Volume: 40
  • Issue Number: 6
  • Publisher: Hemming Group, Limited
  • ISSN: 0041-0683
  • Serial URL: http://www.tecmagazine.com/

Subject/Index Terms

• TRT Terms: Approximation (Mathematics); Crash rates; Distributions (Statistics); Location; Mathematical analysis; Mathematical models; Methodology; Networks; Regression analysis; Roads; Significance (Statistics); Simulation; Statistics; Stochastic processes
• Uncontrolled Terms: Road networks
• ITRD Terms: 1612: Accident rate; 6471: Analysis (math); 6442: Approximation; 6572: Distribution (stat); 9061: Location; 6473: Mathematical model; 9102: Method; 6588: Regression analysis; 6532: Significance (stat); 9103: Simulation; 6555: Statistics; 6574: Stochastic process
• Subject Areas: Data and Information Technology; Highways;

Filing Info

• Accession Number: 00769885
• Record Type: Publication
• Source Agency: Transport Research Laboratory
• Files: ITRD
• Created Date: Oct 7 1999 12:00AM