以離散型計數資料統計模式探討鐵路平交道事故風險因素──以臺鐵為例

INVESTIGATING RISK FACTORS OF TRAFFIC ACCIDENTS AT HIGHWAY-RAILROAD GRADE CROSSINGS USING COUNTED DATA STATISTICAL MODELING: A CASE STUDY OF THE TRA SYSTEM

鐵路平交道為鐵公路交會之處，鐵公路用路人共同使用該路權。儘管臺鐵局所轄的平交道數相對有限，不過近年來國內平交道事故頻傳，造成人民生命財產的損失，以及顯著的交通延滯與社會成本支出。因此，有必要深入探討平交道事故的主要肇因與風險水準，據以研提有效的改善措施。本研究以離散型計數資料統計模式，分別探討鐵路平交道事故頻率、事故衝擊程度，以及事故風險水準。為完整地評估鐵路平交道之事故風險程度，本研究從離散型計數資料模式家族的卜瓦松迴歸、負二項迴歸，以及零膨脹卜瓦松迴歸模式中，選擇較適當之模式，據以構建鐵路平交道事故頻率與事故風險模式；另以排序性羅吉斯迴歸模式，構建平交道事故衝擊程度模式。實證研究結果顯示，每日列車數與公路年平均日交通量等交通曝光量相關因素，為平交道事故的主要肇因；此外，平交道寬度與道路等級等因素，亦為可能的影響因素。最後，針對所構建的模式進行彈性分析與勝算比分析，以提供鐵路平交道主管機關在推動相關安全改善措施前之參考。

A highway-railroad grade crossing (HRGX) is a spatial location where highway and rail users share the same right-of-way. Despite the number of HRGXs managed by the Taiwan Railways Administration (TRA) is relatively few, a significant number of traffic collisions and severe consequences at the HRGXs in Taiwan has signaled an urgent need for appropriate models to identify the key factors that are associated with the casualty risk levels of an HRGX. This study uses counted data statistical models where Poisson, negative binominal and zero-inflated Poisson regression models are evaluated to analyze crash frequency and risk levels in terms of number of fatalities and injuries in a given time period. Besides, crash severity in terms of number of equivalent injuries per traffic collision is modeled by the ordered logistic regression approach. In the empirical study, using the recent crash and HRGX inventory data provided by the TRA, this study aims to analyze the causal relationships between crash frequency and/or severity and a set of influence factors. The empirical study results indicate that traffic exposure related variables such as daily trains and/or highway average annual daily traffic are the key factors. Besides, grade crossing width and highway type also play different roles on the occurrence and/or consequence of a traffic collision(s) at HRGXs. Finally, elasticity and odds ratio analysis based on the calibrated models are also conducted. The ultimate goal of this research is to provide the HGRX administrative office with a reference before implementing related countermeasures for safety improvement.