Crash Estimation At Signalized Intersections Significant Factors And Temporal Effect

Longitudinal intersection crash data are observations on a cross section of intersections that are observed over several time periods. Such cross-section and time series data structures have positive temporal correlation within each intersection. Using the basic negative binomial regression leads to invalid statistical inference due to incorrect test statistics and standard errors that are based on the misspecified variance. Generalized estimating equations (GEEs) provide an extension of generalized linear models to the analysis of longitudinal data, which account for the correlation in the repeated observations for a given intersection. In this study, the GEE procedure was used to model the temporal correlation for longitudinal intersection crash data. This analysis was based on 3-year data for 208 four-legged signalized intersections in central Florida. Intersection crash frequencies were fitted through use of GEEs with a negative binomial link function for four correlation structures. Both the functional form and the link function of GEE models were assessed with the cumulative residuals method. This method showed that the total average daily traffic per lane was the best representation of traffic volume, and the GEE model with autoregression structure had the best model performance. Variable relative effect analysis identified the relative effect for the variables in the models. Intersections with heavy traffic, a larger total number of lanes, a large number of phases per cycle, and high speed limits and those in high population areas were correlated with high crash frequencies. The intersections with more exclusive right-turn lanes and with a partial left-turn protection phase had lower crash risks