木柵線捷運系統為國內首次通車之都市捷運系統，本研究即利用其通車前後所調 查之個體運具選擇或轉移行為資料，進行「事前與事後」比較之實證研究。在通車前行為分 析中，本研究利用「顯示性偏好」與「敘述性偏好」資料分別建立二元、三元及巢式選擇糢 式，並結合此兩類資料建立「整合性偏好模式」。在通車後旅運行為分析中，則建構旅運者 實際選擇結構之顯示性偏好模式。首先在通車前後運具選擇行為一致性之檢定結果顯示，僅 「成本項」參數具一致性評價，而整體模式則不具一致性。因此本研究乃利用更新「運具特 定常數」、「參數尺度」及「效用函數尺度因」等方法，進行通車前後行為模式之比較；比 較結果發現，事前捷運特定常數項被高度高估，而事前運具屬性變數之邊際負效用則相對被 低估。另外，在模式更新前後之驗證中發現：（1）木柵地區之個體運具模式以二元/三元選 擇結構較為適用。（2）事前個體旅運需求以「整合性偏好模式」進行預測，當可獲得較理 想之預測結果；而由多元模式之尺度因子可知，「敘述性偏好」資料較「顯示性偏好」資料 有較大誤差。（3）由「更新效用函數尺度因子」法校估結果得知，選擇結構設定顯然會影 響數據誤差程度之判定。（4）由木柵地區實證經驗得知，利用「更新運具特定常數」法修 正事前模式之預測偏差，不失為一易行且效果良好之方法。

　　　　　The purpose of this study is to explore the disaggregate model choice/ switch behavior before and after Mucha line rapid transit operated in Taiwan. In the analysis of pre-choice behavior, the revealed preference (RP) and stated preference (SP) data are used to establish the binary, trinomined and nested choice models, and the combined RP—SP (i.e. joint preference) modle. In the analysis of post-choice behavior, the RP choice model which posses the real choice hierarchy is established. First of all, the consistency between pre-choice and post-choice model is examined. The statistical test shows that only the parameter of cost item is consistent. However, both of these models are inconsistent. Therefore, the pre-choice and post-choice models are compared by employing three methods which are the updated mode-specific costant, the calibrated scale factor of parameter, and the calibrated scale factor of utility function methods. The results of the comparison reveal that the value of rapid-transit-specific constant of pre-choice model is overestimated, but the negative utility value of mode-attribute variable is underestimated. Besides, there are important findings about the calibration of the updating model: (1) The binary and trinominal choice hierarchy of disaggregate model are more suitable to adopt in the Mucha area. (2) In the pre-choice analysis of disaggregate demand, the combined PR-SP model is suggested to be applied to forecast and can gain more accurate forecast result. Besides, the scale parameter of the combined Mutilnominal choice model is less than one, and this result show that the SP data has more estimated that the RP data. (3) As a result of the estimation of the updated scale parameter, the choice structure wil affect the degree of data error. (4) The validation of Mucha Area shows that adopting the updated mode-specific constant methos to adjust the forecast bias of pre-choice model is a easy and good effect method.