Spatiotemporal crash outliers are also noteworthy: evidence from 13 years of pedestrian-vehicle crashes in North Carolina

Introduction Previous research usually focused on high-frequency crash clusters (surrounded by high-frequency crashes), which overlooked outlier locations where high-frequency crashes were surrounded by low-frequency crashes. Neglecting spatiotemporal outliers might overlook critical factors for safety improvements. Methods Using pedestrian-vehicle crash data in North Carolina from 2007 to 2019, this study proposes an enhanced spatiotemporal analysis framework (combined with Approximate Nearest Neighbour and the Global Moran I index) to distinguish spatiotemporal crash outliers from aggregated/dispersed patterns. Random parameters ordered logit models with heterogeneity in means are used to model ordered injury severities, identify significant factors and explore in-depth heterogeneity across observations. Results Likelihood ratio test results indicate significant instability exists in factors across four spatiotemporal patterns: high-low outlier, low-high outlier, high-high cluster and low-low cluster. Also, spatiotemporal variations and shifts in the proportion of crashes that suffer more severe injuries are founded on the marginal effects of several factors. Specific countermeasures and policy guidance are suggested under different patterns. In high-low outliers, drunk drivers increase the probability of more severe crash outcomes in public vehicular areas from 16.3% to 92.8%. Meanwhile, in low-high outliers, 64.8% of the crashes with speed limits between 45 and 75 mph would result in more severe outcomes. By increasing the mean, older pedestrians, drunk pedestrians and urban areas increase this proportion to 91.6%, 87.2% and 78.2%, respectively. Conclusion These findings underscore the importance of strengthening alcohol tests and setting safety crossing facilities for older pedestrians in identified high-speed urban areas. All these propose a calling for spatiotemporal outlier investigation in future crash analysis and prevention.