存活分析法應用於機車持有年限之研究

A HAZARD-BASED ANALYSIS OF MOTORCYCLE HOLDING DURATION

本研究利用交通部統計處之機車使用狀況調查及行政院主計處公布之縣市總體社經資料，結合監理車籍登錄系統來觀測特定樣本之機車持有歷程，建構機車持有年限之Cox 等比率危險模式。研究結果顯示，持有中古車、購車之車齡越高、汽缸容量數越小、年維修成本越高、使用者越年輕、每週行駛里程越多、家中僅有唯一1 輛機車，以及車籍登記為大臺北地區等屬性，均會增加結束機車持有的危險率。而縣市總體社經變數中，縣市失業率越低、每千人小客車數越少、平均消費傾向越強，以及恩格爾係數越低，亦將增加結束機車持有之風險。進一步依報廢及過戶事件之風險性質不同分別建構競爭風險存活模式，結果顯示報廢與過戶事件兩類存活模式，在解釋變數顯著性與影響持有年限之危險率均具有明顯差異性存在。研究中並針對存在不確定預期影響方向、競爭相反假設，以及與預期假設相反之解釋變數進行探討，最後對於研究限制包括動態屬性與真實持有期間、資料可能存在異質性，以及競爭風險模式之條件獨立性假設等三方面問題亦分別有所討論。

To explore the association between motorcycle holding duration and its determinant factors, We developed a design which linked the sampled data for motorcycle usage to the registered motorcycle information recorded in the Taiwan’s Vehicle Registration System. The complete ownership duration of the sampled motorcycle owners could be observed by tracing the motorcycle registration records. Two kinds of Cox proportional hazard models, single risk and competing risk models, were separately applied to construct the regression relationships between the hazard rate of ownership termination and its influential factors including motorcycle conditions, user’s attributes, and aggregate economic variables. The results showed that under the single risk model, the following characteristics, such as owning a used motorcycle, buying a older, used motorcycle at the beginning of owning, owning a motorcycle with a lower displacement and with a higher maintenance cost, running a higher kilometers by the users, having only one motorcycle in the household, and the motorcycle being registered in Great Taipei Area, would raise the hazard ratio as compared individually with their counterparts. In addition, aggregating economic variables including a lower unemployed rate, fewer numbers of passenger cars owned, a higher consuming propensity, and a lower Engel’s coefficient in a city increased the difficulty of terminating motorcycle ownership in the city as well. On the other hand, two different natures of risk-specific events —transferring or discarding a motorcycle while ending its ownership— in the competing risk model revealed different patterns for the independent variables and the sensitivities for the hazard rates. Some variables with ambiguously influential direction and research limitations were also discussed in this study.