Estimating Rear-End Crash Frequencies in the Presence of Overdispersion: an Application of Heterogeneous Negative Binomial Regression Model

Objectives: Rear–end crashes have rarely been addressed in the literature. Thisstudy aimed to evaluate the safety effects of road and roadside characteristics onthe frequency of rear-end crashes.Methods: A heterogeneous negative binomial (HTNB) regression model wasdeveloped using crash data collected on 388 segments of Malaysia federal roads.The data used in this study includes a total of 2,215 records of rear-end crashesoccurred over a 3-year period from 2007 through 2009.Results: The result implies a strong evidence of overdispersion in the crash data,and hence the HTNB model was used to handle the issue. The findings also showthat variables of average daily traffic, horizontal curvature, speed limit, outsideshoulder width, access points, and area type significantly influence the number ofrear-end crashes.Conclusions: The HTNB model, as an extended form of standard negativebinomial regression model, is an effective technique to model crash datacharacterized by unobserved heterogeneity. The modelling results presented inthis study appeared to be practical tools for developing appropriatecountermeasures for remedial treatment. The findings of this study suggest thatrear-end crashes could be potentially reduced through some specific efforts suchas proper control of minor access to the roadway, posting lower speed limits andwarning signs in areas with higher tendency to rear-end crashes, and wideningshoulder width.

Keywords:

Rear-end crashes , Crash prediction models , Dispersion parameter , Negative binomial model