Uncovering crash narratives: enhancing safety reporting for underprotected road users in transit bus collisions

Crashes involving transit buses and under-protected road users (URUs), including pedestrians and cyclists, present persistent safety concerns in multimodal transportation systems. Structured crash datasets often lack detail on contextual factors such as unsafe URU actions, unusual crash scenarios, and contributory behaviors. This study addressed these limitations by leveraging the crash narrative data from the National Transit Database’s (NTD) Major Safety Events dataset. A corpus of 5,629 URU-involved bus crashes (2014–2023) was analyzed using Latent Dirichlet Allocation to derive 20 topics representing latent behavioral and situational factors (e.g. URU distraction, boarding and alighting conflicts). Topic indicators were integrated with 10 structured variables to create an enhanced URU crash database. URU crash injury severity was modeled using fixed and random-parameters ordered logit specifications; the random-parameters model outperformed the fixed-parameters specification, capturing unobserved heterogeneity. Results indicate that bus speeds >25 mph, dark conditions, and locations without traffic control were associated with increase in fatal injury probabilities of 0.108, 0.089, and 0.030, respectively. Five narrative-derived topics, including URU distraction and boarding/alighting conflicts, were significantly associated with severe crash outcomes. Findings indicate that integrating narrative text mining with structured data strengthens crash documentation and supports data-informed strategies to prevent URU crashes with transit buses.