This study aims to quantitatively examine the variations in effect of road-level factors on crash frequency across different regions. Treating the hierarchical structure existing in the crash data that road entity nested within the geographic region, a hierarchical random parameter model, which allows the coefficients of road-level variables to vary with regions, is proposed. A Poisson lognormal model and a hierarchical random intercept model are also built for the purpose of comparison. A specific roadway facility type, urban two-lane two-way roadway segments in Florida, with crash and road level data including traffic volume, road length, surface condition, and access density for three-year period are used for a case study. The result shows that, in the hierarchical-random parameter model, the local regression coefficients and marginal effects of the road level factors vary over a wide range in the selected counties, which clearly illustrates the non-stationary in the relationships between road level factors and crash frequency across the counties. In regard to the model comparison, the hierarchical random parameter model outperforms the Poisson lognormal model and the hierarchical random intercept model in term of deviance information criterion (DIC). This further confirms the necessity of the use of hierarchical random parameter model in analyzing the crash frequency for road entities in different regions. This study provides a potential in guidance of model construction that considers regional variations (heterogeneities) in safety effects of road-level factors.

这项研究旨在定量检查不同地区道路因素对碰撞频率的影响变化。针对道路实体嵌套在地理区域内的碰撞数据中存在的分层结构，提出了一种分层随机参数模型，该模型允许道路水平变量的系数随区域而变化。为了比较的目的，还建立了泊松对数正态模型和分层随机截距模型。案例研究使用了特定的道路设施类型，即佛罗里达州的城市两车道两路道路网段，并提供了三年期间的碰撞和道路高度数据，包括交通量，道路长度，地面状况和出入密度。结果表明，在层次随机参数模型中，在选定的县中，道路回归系数的局部回归系数和边际效应在很宽的范围内变化，这清楚地说明了在各个县，道路回归系数与交通事故频率之间的关系是不稳定的。关于模型比较，就偏差信息标准（DIC）而言，分层随机参数模型的性能优于Poisson对数正态模型和分层随机拦截模型。这进一步证实了使用分层随机参数模型分析不同地区道路实体的碰撞频率的必要性。这项研究为模型构建的指导提供了潜力，该模型考虑了道路水平因素的安全影响中的区域差异（异质性）。