[automatically translated] As is known, the forecast models of accidents (an area known as the International Safety Performance Functions) are long since used in safety assessments; to them it is given the task to condense prior knowledge on the extent of safety (road sections, intersections, etc.) similar to those into account. When integrated into an empirical Bayesian estimation procedure, the forecast models of accidents contribute to the correction of the distortion due to regression to the mean and to enhancing the accuracy of the estimate. In the present paper presents an example of determination of these models for the case of urban roundabouts. Through the same example, it intends to show the opportunity to take into account the time correlation of the data, in consequence of which the hypothesis of independence authorizing the use of generalized linear models (GLM) is at fault. At the same time, we intend to show how additional effort that takes into account the actual data correlation structure can be offset by improved accuracy in estimating model parameters.
|Publication status||Published - 2008|