Intersections and non-intersections: A protocol for identifying pedestrian crash risk locations in GIS

Intersection and non-intersection locations are commonly used as spatial units of analysis for modeling pedestrian crashes. While both location types have been previously studied, comparing results is difficult given the different data and methods used to identify crash-risk locations. In this study, a systematic and replicable protocol was developed in GIS (Geographic Information System) to create a consistent spatial unit of analysis for use in pedestrian crash modelling. Four publicly accessible datasets were used to identify unique intersection and non-intersection locations: Roadway intersection points, roadway lanes, legal speed limits, and pedestrian crash records. Two algorithms were developed and tested using five search radii (ranging from 20 to 100 m) to assess the protocol reliability. The algorithms, which were designed to identify crash-risk locations at intersection and non-intersection areas detected 87.2% of the pedestrian crash locations (r: 20 m). Agreement rates between algorithm results and the crash data were 94.1% for intersection and 98.0% for non-intersection locations, respectively. The buffer size of 20 m generally showed the highest performance in the analyses. The present protocol offered an efficient and reliable method to create spatial analysis units for pedestrian crash modeling. It provided researchers a cost-effective method to identify unique intersection and non-intersection locations. Additional search radii should be tested in future studies to refine the capture of crash-risk locations.