Safety contributing factors analysis of older vulnerable road users: General and local perspectives

Increasing attention to older people’s traffic safety is necessary to understand the relationship between their traffic safety and contributing factors on a spatial scale. However, zero crashes exist at the analysis unit for some specific types of crashes, and few studies have considered the spatial heterogeneity between older people’s crash frequency and the influencing variables. To fill these gaps, this study developed an analytic approach to explore the effects of contributing factors for older vulnerable road users’ (VRUs) crashes, with particular attention to the integration of general and local analysis. Socio-economic, road network, public facility, traffic enforcement and older VRU crashes were collected in the grids. The gradient tree-boosted Tweedie compound Poisson models (TDboost) were employed to address zero-inflated crash data from the general aspect. Geographically weighted random forests (GWRF) models were employed to reveal the spatial heterogeneity from the local aspect. The results showed that population and healthcare played an important role in predicting older VRU crashes. Major influencing factors showed nonlinear effects on older VRU crashes. They had a positive correlation with both older pedestrian crashes and non-motorized vehicle (NMV) crashes. This study demonstrated that the TDboost excelled in dealing with zero-inflated crash data and the complex effects of safety contributing factors, compared with conventional statistical models (e.g., negative binomial model and zero-inflated negative binomial model) in both prediction accuracy and parameter interpretation. The local variable importance of major contributing factors for VRU crashes showed a spatial clustering tendency and a block distribution tendency. The findings provided important insights into reducing older VRU crashes. For example, the concentration areas for older people, including healthcare facilities, markets, and bus stops, could be targeted to make safety improvements. The analysis sheds light on the nonlinear effects and spatial heterogeneity of safety contributing factors on older VRU crashes, which are usually disregarded in the older traffic safety. The proposed approach emphasizes that the countermeasures for improvement should be formulated based on the spatial distribution of the variable importance, that is, “adapt to local conditions”.