This paper describes a Bayesian statistical technique for using crash records to estimate the frequency and rate of median-crossing crashes (MCC) on each set of highway sections, when MCCs are not explicitly identified in computerized crash records. The technique requires an analyst to review only a subset of hard-copy accident reports to produce a training sample, which is then used to identify computerized data associated (possibly imperfectly) with whether or not a crash was an MCC. This association can then be exploited to use larger sets of computerized records to increase statistical power over that provided by the training sample alone.

This technique was applied to data from Minnesotas freeways and rural expressways. Estimates which allowed highway sections to be ranked with respect to estimated frequency of MCCs, estimated density of MCCs, or estimated MCC rate were computed, and rankings with respect to estimated frequency are reported.