Grouped random parameters Poisson-Lindley model with spatial effects addressing crashes at intersections: Insights from visual environment features and spatiotemporal instability

This study investigates the unobserved heterogeneity and spatiotemporal variations in the effects of visual environment features on intersection crash frequency. A Grouped Random Parameters Poisson-Lindley model with Spatial Effects is developed to account for spatial variations at both the macro (county) and micro (intersection) levels. The analysis utilizes crash data from 2,044 intersections across 12 Florida counties, collected between 2020 and 2022, along with explanatory variables including traffic flow, geometric design characteristics, and visual environment features (extracted from Google Street View images). Comparing to existing methods (e.g., Fixed, Random Parameters, and Grouped Random Parameters Poisson-Lindley models), the proposed approach, which incorporates both macro- and micro-level spatial effects, demonstrates significantly improved model performance. Additionally, the temporal variations of explanatory variables over the three-year period are clearly identified through out-of-sample predictions and marginal effects analysis. Two visual environment features, Vegetation and Grass, result in the identification of grouped random parameters, highlighting the varying impact of these features on intersection crash frequency across the 12 counties. The findings also reveal a strengthening of micro-level spatial effects, indicating heightened spatial correlations between adjacent intersections following the COVID-19 pandemic. Key factors influencing crash frequency include traffic volume, four-legged intersections, major roads with more than four lanes, wider minor roads, and a higher proportion of vehicles in the drivers’ field of vision. These results provide valuable insights into the influence of drivers’ visual environment on intersection safety and offer policy recommendations for enhancing traffic safety.