以存活理論分析臺灣汽機車駕駛人交通違規舉發記錄對未來事故發生時間之影響

Using Survival Theory to Analyze the Influence of Traffic Violation Records to Accidents Occurrence on Car Drivers and Motorcyclists in Taiwan

張洺瑋(Ming-Wei Chang)

指導教授 ： 許添本

國立臺灣大學/工學院/土木工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201901472

摘要

交通違規舉發常被預期可以防止駕駛人未來發生交通事故的手段之一，故本研究旨在找出交通違規舉發與交通事故的發生之關聯性。結合2011-2017年警政署事故資料庫、公路總局第三代公路監理資訊系統駕駛人主檔、違規主檔、車籍異動檔等，以及衛生福利部死因資料庫，進行資料庫清洗和串聯。將交通違規種類分為31種，並根據違規和事故資料之車籍車種欄位，將駕駛人分為汽車駕駛人和機車駕駛人，以進行分析。先以敘述統計分析汽機車駕駛人曾經被舉發違規下，其駕駛人和未來發生事故數的比率關係，再以關聯規則，分析這些被舉發違規的汽機車駕駛人，和未來發生事故的關聯。為了再進一步了解曾經被舉發違規和其不同駕照狀態下的汽機車駕駛人，從第一次違規舉發到第一次發生事故，其發生事故率和時間的關係，先繪製Kaplan-Meier (K-M)存活曲線觀察不同駕照狀態和曾經被舉發違規的汽機車駕駛人之發生事故的存活率，也就是不發生事故的機率。再以Cox等比例風險模式分析影響駕駛人單一事件時間變數的相關變數，包括曾經被舉發的違規種類，以及汽機車駕駛人之駕照狀態。而Cox等比例風險模式卻只能分析駕駛人單一事件的存活時間，但駕駛人在研究期間內，可能不僅僅只有一件交通事故的發生，可能還會伴隨著多件交通事故，使得一個駕駛人會有多個事件的存活時間，無法以Cox等比例風險模式分析。因此，引入存活理論常使用的復發事件邊際模型，包括AG模型和PWP模型，以對擁有多個事件的駕駛人，進行曾經被舉發的違規種類和汽機車駕駛人駕照狀態之變數分析。
Kaplan-Meier存活曲線結果指出，在不同駕照狀態之汽機車駕駛人中，「僅有機車駕照之汽車駕駛人」發生事故存活曲線下降最快，且最終存活率僅有0.845。「有機車駕照之機車駕駛人」和「有汽車駕照之汽車駕駛人」中，機車駕駛人的發生事故存活曲線下降較快，但機車駕駛人之最終存活率為0.849，汽車駕駛人之最終存活率為0.848。在存活理論復發事件PWP模型的結果下，「僅有機車駕照之汽車駕駛人」未來發生事故的時間危險率為1.48。而在曾經被舉發「車輛設備和規格違規」、「抗拒稽查或肇逃」的機車駕駛人，相較於無違反此類違規的人，未來發生事故的時間危險率上升至1.10和1.23。曾經被舉發「酒駕和藥駕」的汽機車駕駛人，其舉發違規對於駕駛人未來發生事故的時間危險率分別下降至0.62和0.50。被舉發違規的駕駛人，未來發生事故的關聯和時間危險率，會因不同的違規種類之舉發，而有所不同。

關鍵字

交通事故 ； 交通違規 ； 存活理論 ； Cox等比例風險模式 ； 復發事件模型

並列摘要

Traffic enforcement is usually expected to add the effectiveness of accident prevention. Therefore, the relationship between traffic violation records and traffic accident needs to be investigated. In this study, the combination of Taiwan’s national traffic accident database, national traffic violation record database, and driver and car registration database is used, and drivers are divided into car drivers and motorcyclists for analysis. This study first analyzes the accident involvement rate of car drivers and motorcyclists with past violation records by the descriptive statistic analysis. Then, the relationship between accident and those drivers is discovered by association rule. Furthermore, it is desired to understand the change of survival time between the first violation records to the first accident of drivers with different driver’s licenses and past violation records, the Kaplan-Meier curve is applied. Then it uses Cox proportion hazard function for further discussion of the relating factors, such as the past violation records and driver’s licenses. However, the Cox proportion hazard function is only used to analyze the single event in each driver. The driver may involve multiple accidents in the research period. The survival time in multiple accidents cannot be analyzed in the Cox proportion hazard function. Therefore, the marginal model of the recurrent events in the survival theory is applied in the study, including the AG model and the PWP model. The survival time of multiple accidents in each driver can be modeled.
Based on the result of the Kaplan-Meier curve for analyzing the different driver’s licenses, the survival curve of car drivers only with motor driver’s license decline quickly than others, and the last survival rate is 0.845. The last survival rate of car drivers with car driver’s license and motorcyclists with motor driver’s license are 0.848 and 0.849 respectively. Based on the further result of the PWP model for analyzing the different driver’s licenses, the hazard ratio of car drivers only with motor driver’s license is increasing to 1.48. For analyzing the past violation records, the hazard ratio of motorcyclists cited by “Improper vehicle equipment” and “Resisting inspection or Hit-and-Run” is increasing to 1.10 and 1.23 respectively. However, the hazard ratio of car drivers and motorcyclists cited by “Drunk and drug driving” is declined to 0.62 and 0.50 respectively. Besides, car drivers and motorcyclists with other violation records can be shown in the different hazard ratio.

並列關鍵字

Traffic Accident ； Traffic Violation ； Survival Theory ； Cox Proportion Hazard Function ； Recurrent Event Model

參考文獻

[1] Kaplan, E. L., & Meier, P. (1958). Nonparametric Estimation from Incomplete Observations, Journal of the American Statistical Association, 53(282).

[2] Cox, D. R. (1972). Regression Models and Life-Tables, Journal of the Royal Statistical Society, 34(2), 187-220.

[3] Prentice, R. L., Williams, B. J., & Peterson, A. V. (1981). On the regression analysis of multivariate failure time data, Biometrika, 68(2), 373-379.

[4] Andersen, P. K., & Gill, R. D. (1982). Cox's Regression Model for Counting Processes: A Large Sample Study, The Annals of Statistics, 10(4), 1100-1120.

[5] Jovanis, P. P., & Chang, H.-l. (1986). Modeling the relationship of accidents to miles traveled, Transportation Research Record, 1068.