應用計數模式分析機車肇事行為

Application of Count Models to Analyze Motorcycle Accident Behavior

本研究利用交通部運輸研究所2001年調查「我國機車持有及使用特性之研究」之問卷資料，探討影響機車旅運者交通肇事之潛在因素，研究中應用之Zero-inflated及Hurdle迴歸計數模式，可解決資料中過多0次肇事件數造成的偏誤，相較於傳統Poisson及NB計數模式可獲得較佳之配適度。研究結果顯示，第一部分「危險駕，駛行為機率模式」共同顯著的變數為：車輛每週使用頻率為六天以上且車齡高於七年以上；第二部分「肇事次數模式」顯著的變數包括：年齡界於23--30歲、個人平均月所得及機車累積行駛里程數。

This paper investigates the factors influencing accident likelihood of motorcyclists in Taiwan by using the data collected by Institute of Transportation, MOTC in 2001. Different models, including Poisson, Negative Binomial, Zero-Inflated Poisson and Hurdle, are applied to model count date, as the dependent variable. Zero-inflated and Hurdle models with two parts have a more general dual regime data generating process in overcoming the problem when the number of zeros in the data exceed what would typically be predicted, compared to Poisson and NB models. The results indicate that the effects of dangerous driving behavior on the accident likelihood appear to be significant in "driving frequency over six days per weed and vehicle age over seven years". The factors that significantly influence the accident likelihood of motorcyclists are "personal monthly income", "age", and "cumulative mileage".