DEVELOPMENT AND APPLICATION OF A RAILROAD-HIGHWAY ACCIDENT-PREDICTION EQUATION

This paper reports the development of an accident-prediction equation for train-vehicle collisions at railroad-highway grade crossings that can be used as the basis for the establishment of a priority order for signal improvements. Most of the quantitative and physical factors in the grade-crossing environment were included. Of the 6000 public grade crossings in Florida, 1140 on state roads were used as the study base. The accident-prediction model was developed by the use of a stepwise regression analysis and three unconventional statistical techniques: (a) the analysis of the plots of the residuals, which indicated that a transformation was required (with the transformation of the dependent variable to a logarithmic form, the plot of the residuals was reasonably symmetric); (b) the observed interaction between the independent variables, which resulted in the use of dummy variables, particularly those for active (warning devices) times daily traffic and number of trains; and (c) a bias in the accident prediction that was introduced by the use of logarithms and eliminated by use of a nonlinear least squares adjustment. The accident-prediction model had a multiple correlation of 0.43. The independent variables in the model were the traffic, number of trains, vehicle speeds, train speeds, number of lanes, and presence of warning devices. The accuracy of the accident-prediction equation was demonstrated by comparisons of actual accidents to predicted accidents. The actual number of train-vehicle accidents in 1975 was 70 percent of the number predicted by the model. In 1975, the total number of accidents remained unchanged from that in 1974, but the number of train-vehicle accidents decreased 22 percent. /Author/

Media Info

  • Media Type: Print
  • Features: References; Tables;
  • Pagination: pp 12-19
  • Monograph Title: Lighting, visibility and railroad-highway grade crossings
  • Serial:

Subject/Index Terms

Filing Info

  • Accession Number: 00168077
  • Record Type: Publication
  • ISBN: 0309026539
  • Report/Paper Numbers: HS-022 157
  • Files: TRIS, TRB
  • Created Date: Jan 30 1978 12:00AM