Fixed and Random Effect Models to Predict Operating Speed Percentiles on Rural Two-Lane Highways

This paper presents the results obtained from the estimation of free-flow speed on two-lane rural highways. The data used for the analysis were collected in the Northwest of Italy using video cameras and a laser speed gun. The model structure adopted separates the estimate of the central tendency of speeds from the typical deviations of individual speeds. Hence, in the model the same set of variables can be used to determine the mean value and the standard deviation of the speed distribution; the desired speed percentile is then calculated by considering the associated standard normal random variable (Z). Random effects (RE) were included in the model to account for the variability in time and space of the data that contains repeated measurements for the same road/section/direction and to remove the dependency between any estimation errors from individual observations. Fixed effect (FE) models are also calibrated for comparison purposes and the BIC criterion is used for variable selection and applied to both the FE and RE models. Estimates reveal many geometric variables in the FE model to be significant, while the RE model only selects the section curvature and the pedestrian density for the estimation of the mean, and nearly all the geometric variables in the estimation of the standard deviation. The random effects relative to the road are not significant, which means that, once the effects related to the random selection of cross sections and direction have been observed, the speed measurements are relatively stable.