The first edition of the Highway Safety Manual (HSM) includes a number of safety performance functions (SPF), which can be used to identify collision-prone locations on a roadway network. The HSM recommends that these SPFs be calibrated in order to more accurately reflect a specific jurisdiction’s unique roadway characteristics, driver behavior, etc. Another alternative is the creation of jurisdiction-specific SPFs. For this study, negative binomial regression was used to develop a set of models using five years of collision data (2005-2009) from the city of Regina, Saskatchewan. Three intersection categories were investigated: 3-leg unsignalized, 4-leg unsignalized, and 3 and 4-leg signalized. The SPFs provided in the HSM were also calibrated using this data, and a set of calibration factors were produced. Statistical goodness of fit (GOF) tests were performed in order to determine the best-fitting SPFs for the study region. In addition to the statistical tests, CURE (cumulative residual) plots were utilized to perform two comparisons: between candidate jurisdiction-specific model forms, and between the jurisdiction specific SPFs and the HSM’s SPFs (both calibrated and un-calibrated). It was found that the jurisdiction-specific SPFs provided the best fit to the data used in this study, and would be the best SPFs for predicting collisions at 3 and 4-leg intersections in the City of Regina.
