Understanding e-scooter rider crash severity using a built environment typology: A two-stage clustering and random parameter model analysis

Abstract

E-scooters are an emerging transport mode that is transforming urban mobility; however, their proliferation has raised concerns about safety. This study combines UK e-scooter crash data with built environment characteristics from the crash locations. A two-stage framework was followed: first, a typology of built environments was developed using K-means++; second, crash severity within each cluster was analysed using a random parameter binary logit model. Four built environment clusters were identified: (1) car-centric and mixed-use zones, (2) commercial and industrial zones, (3) intersection-dense areas, and (4) residential and central areas. Collisions with motor vehicles, younger e-scooter riders, and higher speed limits were the most common risk factors across the clusters, with the first two clusters showing a higher impact of these factors on the likelihood of severe crashes. In the first and second clusters, riding on the carriageway significantly increased injury severity. In the second cluster, three collision types were significant, more than in other clusters where only side-impact collisions were significant. This indicates high e-scooter–motor vehicle friction in the second cluster. Among all collision types, head-on collisions increased the likelihood of severe outcomes more than others. In the third and fourth clusters, peak hours were associated with a lower likelihood of severe crashes, while this variable showed the opposite impact in the first cluster. The results highlight that consideration of the surrounding built environment is paramount when analysing e-scooter crash severity, as unique contributing factors were identified specific to each built environment type, along with varying magnitudes or directions of marginal effects.