MODELLING THE TEMPORAL DISTRIBUTION AND SEVERITY OF E-SCOOTER CRASHES IN SERBIA

The increasing use of electric scooters has created new road-safety challenges, particularly because crash occurrence and injury severity may follow different temporal patterns. This study examined the temporal distribution and severity of police-reported e-scooter crashes in Serbia. Data were included 282 crashes recorded between 22 July 2023 and 26 December 2025. The principal analysis was restricted to 243 crashes recorded during the complete calendar years 2024 and 2025. Hourly crash counts were analysed using Poisson regression with cyclic harmonic terms for daily and annual variation, while killed-or-seriously-injured (KSI) outcomes were examined using Firth penalised logistic regression. The severity model included year, weekend status, morning and night periods, site type and certainty of motor-vehicle involvement. Of the analysed crashes, 54 (22.2%) were classified as KSI. Adjusted crash intensity was higher in 2025 than in 2024 (IRR=1.32, 95% CI 1.03–1.69) and lower during weekend hours than weekdays (IRR=0.63, 95% CI 0.46–0.87). The fitted temporal distribution reached its daily maximum in the mid-afternoon and its annual maximum in late August. Morning crashes between 06:00 and 09:59 had significantly higher adjusted odds of a KSI outcome (aOR=2.63, 95% CI 1.20–5.76). Night-time, weekend, street-segment and motor-vehicle linkage estimates were imprecise and did not provide clear statistical evidence. The findings demonstrate that periods with the highest crash frequency differ from those associated with the most serious consequences. Prevention should combine time-targeted enforcement, speed management, continuous micromobility infrastructure and improved linkage of police, health and exposure data.