A Poisson regression model is proposed to establish empirical relationships between truck accidents and key highway geometric design variables. For a particular road section, the number of trucks involved in accidents over 1 year was assumed to be Poisson-distributed. The Poisson rate was related to the road section's geometric, traffic, and other explanatory variables (or covariates) by a loglinear function, which ensures that the rate is always nonnegative. The primary data source used was the Highway Safety Information System (HSIS), administered by FHWA. Highway geometric and traffic data for rural Interstate highways and the associated truck accidents in one HSIS state from 1985 to 1987 were used to illustrate the proposed model. The maximum likelihood method was used to estimate the model coefficients. The final model suggested that annual average daily traffic per lane, horizontal curvature, and vertical grade were significantly correlated with truck accident involvement rate but that shoulder width had comparably less correlation. Goodness-of-fit test statistics indicated that extra variation (or overdispersion) existed in the developed Poisson model, which was most likely due to the uncertainties in truck exposure data and omitted variables in the model. This suggests that better quality in truck exposure data and additional covariates could probably improve the current model. Subsequent analyses suggested, however, that this overdispersion did not change the conclusions about the relationships between truck accidents and the examined geometric and traffic variables.


  • English

Media Info

  • Features: References; Tables;
  • Pagination: p. 10-18
  • Monograph Title: Safety research: heavy vehicles, information systems, and crash studies and methods
  • Serial:

Subject/Index Terms

Filing Info

  • Accession Number: 00627216
  • Record Type: Publication
  • ISBN: 0309054176
  • Files: TRIS, TRB, ATRI
  • Created Date: Apr 2 1993 12:00AM