Transferability of Models That Estimate Crashes as a Function of Access Management

Abstract

Access management can improve safety or operations but may reduce accessibility for adjacent residences or businesses, causing conflict between the needs of local and through traffic and resulting in access management decisions being challenged by the public. Accordingly, transportation administrators require assurance regarding the accuracy of projected safety impacts of access management decisions. Several existing mathematical models quantify these impacts for selected access management techniques. Since new models require substantial resources to construct, it is prudent to investigate the extent to which existing models can be applied in other locations. A study of mathematical models that predict the number of crashes that will result as a function of signal spacing, median treatments, unsignalized driveways, and other key variables is reported. The study compared five models using a 10-year set of geometric, operational, and crash data for three case study corridors. The crash-prediction ability of these models was assessed using traffic flow data other than the data set used to construct the models. Because some of the models were not developed for application elsewhere or were intended to be used only with a site-specific adjustment, the error percentages indicate the extent to which the models are transferable, but they do not constitute a critique of the previous research. With no site-specific adjustment, the accuracy of the models ranges from an average error of 34 to a few hundred percent. With simple site-specific adjustments, errors are in the range of 27 to 29 percent. Substantially less effort is required for these site-specific adjustments than would be required to create a new mathematical model. Recommendations were developed for using these models in practice, understanding their limitations, and interpreting their sensitivity to key inputs, data needs, and computational requirements.