Reliability analysis has been advocated as a robust methodology to quantify the risk (known as the probability of non-compliance, P_nc) associated with design limitations such as insufficient sight distance on horizontal curves. This risk represents the probability that the current design (e.g., available sight distance) would fail to meet the requirements of the driving population (e.g., required sight distance). Although previous work has quantified the risk and established links between P_nc and safety, P_nc remains a statistical measure that is not informative enough to roadway designers. To overcome this limitation, the impacts of geometric design attributes on the P_nc as well as the direct and indirect (through the impacts on P_nc) impacts of those attributes on safety need to be modelled and understood. To achieve the aforementioned objective, this paper proposes the adoption of Structural Equation Modelling (SEM) to simultaneously model the relationships mentioned above using data collected on horizontal curves in Alberta, Canada. Light Detection And Ranging (LiDAR) data was first used to extract curve information and assess the 3D available sight distance on 244 curved highway segments. Reliability analysis was then used to quantify the P_nc associated with each curved segment. SEM was then used to test the impacts of curve geometric and traffic attributes on P_nc as well as the direct and indirect impacts those variables have on collision frequency. The results showed that P_nc, curve length, and traffic volume all had statistically significant effects on collision frequency. Curve attributes such as the curve’s deflection angle and chord length had statistically significant impacts on P_nc. Both curve deflection angle and the chord length were also found to have an indirect influence on collisions, which was mediated through P_nc. The findings emphasize the importance of considering other curve attributes when designing horizontal curves instead of focusing entirely on the curve radius. The findings of this research provide insights into the indirect impacts of curve attributes of horizontal curves on safety. This could help designers consider curve features that have the highest impacts on non-compliance and safety levels.

可靠性分析一直主张作为完善的方法来量化的风险（被称为非依从性，概率P _NC）与设计相关的局限性，如在水平曲线视距不足。该风险表示当前设计（例如，可用的视距）将无法满足驾驶人员的需求（例如，所需的视距）的概率。尽管先前的工作已经量化了风险并建立了P _nc和安全性之间的联系，但P _nc仍然是一种统计量度，不足以为巷道设计人员提供足够的信息。为了克服此限制，几何设计属性对P _nc的影响以及这些属性对安全的直接和间接影响（通过对P _nc的影响）都需要进行建模和理解。为了达到上述目的，本文提出采用结构方程模型（SEM）来同时利用加拿大艾伯塔省水平曲线上收集的数据对上述关系进行建模。首先使用光检测和测距（LiDAR）数据提取弯道信息并评估244个弯曲的高速公路路段上的3D可用视距。然后，使用可靠性分析来量化与每个弯曲段相关的P _nc。然后使用SEM测试曲线几何和交通属性对P _nc的影响以及这些变量对碰撞频率的直接和间接影响。结果表明，P _nc，曲线长度和交通量均对碰撞频率具有统计学意义。曲线的属性（例如曲线的偏转角和弦长）对P _nc具有统计上显着的影响。还发现曲线的偏转角和弦长都对碰撞有间接影响，这是通过P _nc介导的。这些发现强调了在设计水平曲线时考虑其他曲线属性的重要性，而不是完全专注于曲线半径。这项研究的发现提供了对水平曲线的曲线属性对安全性的间接影响的见解。这可以帮助设计人员考虑对不合规和安全级别影响最大的曲线特征。