Road Accident Prediction Models Developed from a National Database: Poisson and Negative Binomial Regressions

The purpose of this paper is to develop road accident prediction model and to identify the important variables for the occurrence of accidents so that new insights can be obtained for safety intervention programs in the Great Britain. Generalized regression models with Poisson and negative binomial probability functions were fitted by using GenStat software. The daily accident data were extracted from official STATS 19 records from 1991 to 2002. Two data sets were developed, one for whole of Great Britain and other for 51 groups of counties separately. Each group represents a Police force that covers one or more counties. The variables used were day, month, year, holidays, Christmas, time, and New Year for the data set of whole Great Britain. The variables of population, length of roads, population density and the police force were used in addition to all other variables for the data set of 51 groups of counties. An incremental approach was used for adding variables during the modeling process. The negative binomial regression model was selected because the data were found to be over dispersed relative to a Poisson process. The most dangerous day and month were found to be respectively Friday and November. Christmas, New Year, and holidays had fewer accidents than other days probably because of low traffic. The day of week, population, population density, length of roads and police force variables were found to explain most of the variation in daily accident occurrence. Keywords: Negative binomial regression, Poisson regression, STATS 19 data, standardized deviance residual.

• Corporate Authors:

  Transportation Research Board

  500 Fifth Street, NW
  Washington, DC  United States  20001
• Authors:
  • Memon, A Q
• Conference:
  • Transportation Research Board 85th Annual Meeting
  • Location: Washington DC, United States
  • Date: 2006-1-22 to 2006-1-26
• Date: 2006

Language

• English

Media Info

• Media Type: CD-ROM
• Features: Figures; References; Tables;
• Pagination: 21p
• Monograph Title: TRB 85th Annual Meeting Compendium of Papers CD-ROM

Subject/Index Terms

• TRT Terms: Crash rates; Days; Linear regression analysis; Mathematical models; Mathematical prediction; Poisson distributions; Regression analysis; Statistical analysis; Time of crashes
• Subject Areas: Data and Information Technology; Highways; Safety and Human Factors; I80: Accident Studies;

Filing Info

• Accession Number: 01029968
• Record Type: Publication
• Report/Paper Numbers: 06-1643
• Files: BTRIS, TRIS, TRB
• Created Date: Jul 31 2006 7:48AM