Book Description

Several research studies have produced mathematical models that predict the safety impacts of selected access management techniques. Since new models require substantial resources to construct, this study evaluated five existing models with regard to their applicability to locations other than the one for which they were designed. The predictive power of the models was assessed using three sites in Virginia. The study also considered the practical aspects of applying the models in Virginia to estimate the likelihood that necessary data are available, the number of computations required to apply the models, the simplicity of the rationale underlying the models, and the sensitivity of the models to inputs. The applicability of the models was wide ranging. Without site-specific adjustments, the average percent error of the models ranged from 34 to a few hundred percent. With simple site-specific adjustments, the error ranged from 27 to 29 percent. Because some of the models were developed for a specific site or were intended to be used only with site-specific adjustments, these error percentages indicate only the extent to which the models are transferable with respect to estimating crashes, not the performance of the models themselves. The wide variation in applicability was due to discrepancies in data definitions, the availability of data, the structure of the model, and the assumptions used. Recommendations were developed for using the models in practice and for understanding their limitations. Two principal conclusions were as follows: (1) existing models (with minor adjustments for some) can predict crashes as a function of access within 34 percent of the actual number, and (2) some of the models are simple enough to be used in practice.