Sensitivity Analysis of Data Set Sizes for Highway Safety Manual Calibration Factors

Abstract

The Highway Safety Manual (HSM), Part C, presents crash prediction models for three types of facilities. Each of these models was developed with data from a few states. As noted in the HSM, “The general level of crash frequencies may vary substantially from one jurisdiction to another for a variety of reasons… . Therefore, for the Part C predictive models to provide results that are meaningful and accurate for each jurisdiction, it is important that they be calibrated for application in each jurisdiction.” However, states may find the data requirements for calibration to be challenging. The HSM offers general recommendations for the size of the calibration data sets, but not much research is available to provide guidance to the states on how the quality of the calibration factor relates to the size of the calibration data set. This research uses data from Washington State for rural two-lane, rural multilane, and urban or suburban arterial highway segments and evaluates the quality of the calibration factors generated from data sets of different sizes, that is, from various percentages of the complete data set. Roadway and crash data are for the 3-year period from 2006 to 2008. The calibration factors generated from the whole data set for each highway type are considered the ideal calibration factors, and factors generated from different data set sizes are compared with these. The probability that the generated calibration factors fall within 5% and 10% of the ideal calibration factor is calculated. The results of this sensitivity analysis are reviewed, and recommendations are derived and presented.