Assessing the importance of functional form selection in developing calibration functions for the highway safety...

The Highway Safety Manual (HSM) offers crash prediction models for estimating the number and severity of crashes for several facility types. However, since these models were developed using data from selected states within the United States, they should be calibrated when applied to a new jurisdiction. The HSM recommended using a scalar calibration factor to calibrate the prediction models. Recently, researchers proposed using a calibration function, instead of a factor, to calibrate the models. The calibration function can also be derived using explanatory variables, such as traffic flow or segment length. However, it is not clear what functional forms (or parameterizations) should be considered in developing calibration functions, and if the selection of the functional form is important when the sample size is small.

We considered 12 functional forms to develop calibration functions for datasets collected on multi-lane freeways in Texas and compared their performance using the goodness-of-fit metrics.

We found that with a large and complete sample, a more complex function (with flow, and segment length) could produce a better fit. As the sample size reduces, simpler functions such as those developed with only the segment length as a variable produce better results.

The application and assessment of most of the functional forms evaluated in this study proved to be useful and should be considered depending on data or available sample size.

Calibration is a crucial step to ensure HSM models are adopted to the local conditions of the jurisdiction under consideration. Our results show that the choice of functional form is important and should be considered by practitioners when calibrating the HSM models.