Analysis of injury severity of single-vehicle and two-vehicle crashes involving lightweight vehicles (K-car) in Japan: A random parameters approach with heterogeneity in means

Objectives: This study investigates the factors influencing injury severity in crashes involving lightweight vehicles (K-cars) in Japan, addressing safety concerns arising from their structural vulnerability. It aims to overcome limitations in prior research by developing a disaggregated analytical framework that separately examines single-vehicle and multiple two-vehicle crash scenarios (K-car vs K-car, K-car vs regular car, K-car vs truck) and explicitly accounts for unobserved heterogeneity. A further objective is to test the transferability of findings across these distinct crash types. Methods: The study utilized nationwide crash data (2020–2022) from Japan’s National Police Agency involving at least 1 K-car. Injury severity was consolidated into a binary variable (casualty vs no-injury). Five separate random parameters probit models with heterogeneity in means were developed to analyze single-vehicle crashes, 3 types of two-vehicle crashes, and an aggregated dataset of all crashes. These models assessed the influence of temporal, environmental, roadway, vehicle, and driver characteristics on injury outcomes. To validate the analytical approach, an out-of- sample prediction method was employed to quantify the prediction bias and assess the transferability of models across crash types. Results: The model accounting for heterogeneity demonstrated a superior goodness-of-fit. Key findings included the following: Older drivers exhibit a dichotomous risk profile, facing higher injury risk in single-vehicle crashes but lower risk in two-vehicle crashes. In addition, medians with only painted lines are associated with more severe outcomes than those with physical barriers. Out-of-sample prediction analysis confirmed that the models are not transferable across crash types, underscoring the necessity of type-specific modeling. Conclusions: Safety analysis for lightweight vehicles must be disaggregated by crash type while accounting for unobserved heterogeneity. The study reveals complex factors, such as the dichotomous risk profile of older drivers, which demand targeted policy interventions. These research findings provide an evidence-based framework for developing effective countermeasures, including enhanced driver assistance systems and improved roadway designs, to comprehensively improve the safety of lightweight vehicles.