Safety performance functions (SPFs) are used for predicting road accident frequencies and are critical to the safety evaluation of road intersection and its design. Despite their importance, the SPFs specific for pedestrian-vehicle accident evaluation are relatively few. Moreover, most established pedestrian SPFs are intersection-based and thus difficult for examining the safety effect of
pedestrian safety countermeasures, such as refuge islands, which are usually leg-based. Moreover, the traffic rules and environments in Taiwan are different from other countries; developing local pedestrian SPFs is therefore desired. The study develops both intersection- and leg-based pedestrian SPFs based on a comprehensive dataset collected from the Taipei City Government; the
2012—2017 dataset consists of data including road accidents, signal controls, road geometry, and pedestrian facilities. The analysis results of zero-inflated negative binomial models show that the intersection- and leg-based SPFs have a similar performance in cross validation. In contrast to the intersection-based SPFs, the leg-based SPFs could directly capture the impacts of pedestrian
facilities and road geometry on pedestrian accidents. The study also discusses the data issues for establishing quality leg-based SPFs as a reference for the government to plan the establishment of road safety databases in the future.