Análise de variações temporais na estimação de acidentes em segmentos arteriais urbanos

Abstract

Accident Prediction Models (APM) have emerged as a statistical tool with the potential to improve road safety studies. The use of time series of traffic accidents in the calibration of APM introduces a temporal correlation between observations of the same entity representing an additional source of bias to the modeling process. This study evaluates the effect of temporal trends in the development of APM for urban arterial segments in Fortaleza. APM were developed from generalized linear models (GLM) and generalized estimating equations (GEE) for a sample of 283 segments with the total number of accidents for the years 2007 to 2011 assuming various structures of temporal correlation. The results do not support the hypothesis of temporal trends in the sample, however, the GLM developed with individual coefficients for each year outperformed the temporally aggregated models.