Modeling Two-vehicle Crash Severity by Generalized Estimating Equations
|關鍵字:||傷亡嚴重度;號誌化路口;廣義估計方程式;次序普羅比模式;Severity level;Signalized intersection;Generalized estimation equations;Ordered Probit model.|
|摘要:|| 以往有關兩車碰撞事故之研究，大多以單變量模式將較嚴重一方當事人為目標變數加以模化，但忽略雙方當事人間之互動關聯，可能造成模式推估之偏誤。基此，部分研究乃改以雙變量次序普羅比模式，將雙方當事人的關聯性納入模式並分別推估雙方當事人的參數值，但卻導致模式變數過多及共線之缺點。因此，本研究以次序普羅比模式結合廣義估計方程式模化兩車碰撞事故嚴重度，將雙方當事人嚴重程度分為死亡或頭部受傷、中度受傷（四肢及頭部以外部位受傷）、輕度受傷(僅四肢受傷) ，以及財損四等級，並以肇事當事人的駕駛特性、車輛種類、碰撞型態、環境特性，以及路口特性等變數進行模化，並與傳統之單變量及雙變量模式進行比較。
Most of previous studies in modeling two-vehicle crash severity levels use of univariate model which primary focus on the more serious injured party by ignoring the potential correlation between two parties, leading to the estimation bias. To overcome this problem, the other studies adopt bivariate models to accommodate the correlation and incorporate explanatory variables of two parties. However, the bivariate models tend to be too complex and have the problem of collinearity. Based on this, this study attempts to adopt generalized estimating equations (GEE) technique in association with ordered Probit models to identify common key risk factors of two parties of an accident, including the potential risk factors of driver characteristics, type of vehicle, collision types, environmental characteristics and intersection characteristics. The severity levels are classified into four categories: disabled injury and fatality (head injury and death), evident injury (body injury excluded limbs and head), possible injury (only limbs are injured) and property damage only. The estimation results show that the GEE-based ordered Probit model performs best in terms of hit ratio, suggesting the correlation of two parties and the applicability of the proposed model in modeling two-vehicle crashes. Moreover, a total of 10 variables are significantly tested, which imply the parties those who are drunk, speeding, motorcycle riders, involved in a corner collision, in darkness, at intersections with high number of lanes or high number of phases tend to have higher severity level. In contrast, the parties those who are male, at intersections with high numbers of directions, and proportion of motorcycles tend to have lower severity level. In which, motorcyclists and number of phases are identified as the most risky factors. Therefore, the corresponding countermeasures, such as enhancement of illumination, improvement of signal timing plan, increase of motorcycle safety management, increase of police enforcement, and promotion of designated driving service, are then proposed. Keywords: Severity level, Signalized intersection, Generalized estimation equations, Ordered Probit model.
|Appears in Collections:||Thesis|