Transferred versus local Safety Performance Functions: A geographical analysis considering two European case studies

Two main approaches can be used to predict road accidents: transferring existing Safety Performance Functions (SPFs) from other areas (transferred SPFs), and developing local SPFs. Both approaches have advantages and disadvantages, and are affected by the difficult choice of predictors. Regional variables or terrain factors may lead prediction improvements. However, results from previous relevant research are contradictory and transferability assessments are mainly based on North-American experiences. Because of these inconsistencies, this study is an attempt of providing new insights on the choice between alternative accident prediction methods by taking into account the geographic variability in the European context. In particular, it addresses three main issues: (1) it compares the prediction accuracy of transferred and local SPFs; (2) it determines the significance of regional factors in explaining safety performances, (3) it assesses the variability of results among the different contexts considered. Research questions are addressed as based on two-lane rural road sites in Italy and Scotland. The analysis shows differences between the two countries, due to the different nature of the networks, but not within each country. Both advantages and disadvantages were highlighted in the evaluation of transferred and local SPFs. Calibration of transferred SPFs may be less demanding than their local estimation, even if they may lead to unreliable estimates when compared to comprehensive SPFs. However, locally developed SPFs may not provide more significantly reliable estimates than transferred SPFs. Segment curvature and shoulder types are statistically significant predictors in both the Italian and Scottish models, even having different importance.