macro-level crash prediction model

In order to achieve the objective of a safe transportation system and reducing the irreparable consequences of the road crashes, it is necessary to consider the traffic safety issues along with other transport planning factors such as the air pollution, socio-economic. Recently, statistical models to quantify the impact of transport planning factors on the traffic safety and, subsequently, to establish a link between the transportation planning and traffic safety have been considered by planners. The purpose of the current study is to develop the comprehensive macro-level crash prediction models by which the effect of a range of the characteristics of traffic analysis zones on the crash frequency could be considered. To do so, in addition to the Poisson model, which is the most common and basic crash prediction model, Poisson-Lognormal and conditional autoregressive models are also used to consider the unobserved heterogeneity and spatial correlations. The Deviance Information Criterion (DIC) has been used to compare the proposed models. The results of the comparison of the models indicate that considering the two factors of overdispersion and spatial correlations by the Poisson-Lognormal and conditional autoregressive models, respectively, significantly improves the performance of the models and reduces the value of DIC from 4623.41 in Poisson model to 2066.82 and 2055.28 in aforementioned models. It is also worth noting that the conditional autoregressive model has the best performance, which highlights the importance of considering the spatial dependence in the crash prediction models with rectification of the estimation and replacement of the omitted variables.