Crash Prediction Models for Rural Motorways

In the paper, crash prediction models for estimating the safety of the rural motorways are presented. Separate models were developed for total crashes and severe (fatal plus all injury) crashes. Generalized linear modeling techniques were used to fit the models, and a negative binomial distribution error structure was assumed. The study made use of a sample of 2,245 crashes (728 severe crashes) that occurred in the period 2001–2005 in the Motorway A16 Naples-Canosa in Italy. Many characteristics of the motorway are sub-standard. It allowed to investigate a wide spectrum of geometric configurations. The models were developed by the stepwise forward procedure using explanatory variables related to: traffic volume and composition, horizontal alignment, vertical alignment, design consistency, sight distance, roadside context, cross-section, speed limits, and interchange ramps. The decision on whether or not to keep a variable in the model was based on two criteria. The first was whether the t-ratio of the variable’s estimated coefficient is significant at the 5% level. The second criterion is based on the improvement of the goodness of fit measures of the model that includes that variable. Goodness of fit measures were theparameter R2α and the Akaike’s Information Criterion. All the parameters have a logical and expected sign. Most important result is that design consistency measures significantly affect road safety not only on two-lane rural highways but also on motorways.