Proposta metodológica de avaliação dos efeitos espaciais de dependência e heterogeneidade dos acidentes de trânsito em diferentes configurações de áreas com a incorporação da superdispersão

Abstract

Since crashes are spatial phenomena, problems such as dependency and spatial heterogeneity emerge and are not considered in the traditional modeling process of the Road Safety Performance. Another aspect that has an impact in the modeling process is the use of arbitrary spatial aggregation units in relation to crashes, which may cause statistical inconsistencies, known as Modifiable Areal Unit Problem (MAUP). In this context, this research has the main objective of proposing a method to evaluate the spatial effects of dependence and heterogeneity of crashes in different configurations of areas with the incorporation of superdispersion. A georeferenced database was compiled at the traffic zone level in Fortaleza, which included injury crashes, exposure factors, road network characteristics, land use, socioeconomic and demographic factors. The application of the systematized spatial data exploratory analysis proposal led to the necessity of jointly considering the effects of dependency and spatial heterogeneity in the modeling of road safety performance at the traffic zone level in Fortaleza. The evaluation of the geographically weighted models that consider the overdispersion showed that the Geographically Weighted Negative Binomial Regression (GWNBR), among the tested ones, was the only approach capable of producing unbiased estimates. The influence of different aggregation scales on variables and macroscopic models of crashes in the city of Fortaleza was investigated. It was verified that the variations of the descriptive statistics depend not only on the scale, but also on the format of the involved variables and the aggregation criterion of the areas. A sensitivity analysis of the coefficients of non-spatial regression models, which were calibrated for various area configurations, showed that considering the crashes in the delimitation of the analysis areas improves the precision of the models. Finally, the variation of the coefficients of the spatial model GWNBR was evaluated and it was verified that, for the specific case of the city of Fortaleza, it was not possible to confirm the hypothesis that these models are less sensitive in relation to the scale change. The application of the proposed method contributed to the improvement of existing techniques that consider the spatial aspects in the macroscopic modeling of road safety performance.