通勤運具選擇行為異質性之研析：混合羅吉特模式之應用

Abstract

由於台灣地區之旅運者相當依賴使用私人運具，不僅使得旅運者在有限的道路容量下會產生交通擁塞以及能源消耗的問題，運具的排放亦會造成環境汙染，特別是在都會地區。促進旅運者使用公共運輸無疑是最有效的改善對策之一，因其能夠降低使用私人運具之交通需求，並且使旅運者更有效地使用道路系統。然而，在不了解旅運者對於運具選擇之偏好時，無法提出對公共運輸有效的之市場行銷及改善政策。此外，本研究認為公共運輸服務水準以及使用私人運具的便利程度會因旅運者於不同地區而有所差異，而差異會存在於旅運者對運具選擇的偏好及決定。因此，本研究在分析旅運者之運具選擇行為時會考慮具有異質性之運具選擇偏好。

我們都了解人口密度為解釋提供公共運輸的關鍵因素之一。有鑑於此，本研究旨在透過旅運者於不同人口密度之地區分別建立運具選擇模式，並進一步使用混合羅吉特模式，表現出在同一地區之旅運者對運具之選擇亦存在個體異質性。在模式推估部分，本研究透過郵寄方式寄發全國性之問卷，最終回收有效問卷為5769份。根據人口密度將旅運者居住於台灣348個鄉鎮市區分為五群：高偏遠地區、低偏遠地區、郊區、都會區及市中心。此外，由於高偏遠地區、低偏遠地區及郊區樣本回收數不足較不適合單獨建模，故將此三群進一步合併並且定義為偏遠地區。最後，在建立傳統多項羅吉特模式(MNL)及混合羅吉特模式(MXL)是分為對偏遠地區、都會區及市中心三群進行比較及推估。

推估結果並利用概似比檢定顯示三群所建立的混合羅吉特模式皆明顯優於多項羅吉特模式，表示運具選擇行為皆存在個體異質性。此外，四項共生變數包含步行時間、等待時間、車內時間及旅行成本，及潛在服務變數均為影響旅運者運具選擇的關鍵因素。特別的是，在偏遠地區此四項共生變數均具有顯著異質性，在都市區有三項共生變數(等待時間、車內時間及旅行成本)具有異質性，而在市中心僅有兩項共生變數(步行時間及等待時間)具有顯著異質性。此結果顯示在低人口密度之偏遠地區具有較顯著之個體異質性。因此，不同類型之公共運輸如需求反應運輸(DRTS)應可導入至這些地區以吸引不同旅運者之喜好。最後，本研究進行彈性分析及市場占有率預測，確立影響公共運輸之關鍵成功因素，並提出對應之市場行銷及改善公共運輸之政策。

Due to the high dependency on private vehicles of travellers in Taiwan, the traffic conditions in many surface roadway systems are rather congested and the problems of energy consumption and emissions are serious, especially in urban areas. To promote the usage of public transportation is undoubtedly the one of the most effective countermeasures, which can largely curtail the traffic demand of private vehicles and more effectively use the roadway systems. However, without knowing the preferences of travellers in choosing transport modes, it is impossible to propose effective marketing and improvement strategies for public transportation. Additionally, to acknowledge the remarkably different service levels of public transportation and different convenient levels in using private vehicles in different districts/townships where the traveller live, significant differences must exist in their mode choice preferences and decisions. Thus, the heterogeneity of mode choice preferences should be considered while analyzing their mode choice behaviours.

As we know, population density is one of key factors explain the provision of public transportation. Based on this, this study aims to separately develop the mode choice models for those travellers living in the districts/townships with different levels of population density and further to use of the mixed logit model to acknowledge the heterogeneity of respondents even living in the areas with same level of population density. For model estimation, a nationwide post-mailed questionnaire survey on commuters was conducted with a total of 5769 valid questionnaires returned. According to the population density, respondents living in the 348 districts/townships of Taiwan inland are classified into five groups: high-rural area, low-rural area, suburban, urban, and central business centre. In addition, due to the small number of valid questionnaires, the first three groups are further combined and defined as the rural area. Traditional multinomial logit models (MNL) and mixed logit models (MXL) for rural, urban and CBD areas are separately compared and estimated.

The estimation results show that MXL models for three groups perform significantly better than MNL models in terms of likelihood ratio tests, suggesting the existence of heterogeneity in mode choice behaviours. In addition, four generic variables of walking time, waiting time, in-vehicle time and travel cost and two generic latent variables of comfort and convenience are the key factors affecting mode choices of commuters. Especially, the heterogeneity in these four factors are all significantly tested for rural areas, while only three factors (waiting time, in-vehicle time, and travel cost) for urban areas and only two factors (walking time and waiting time) for CBD areas are significant, suggesting the heterogeneity is more significant in the remote areas with low population density. Thus, different types of public transportation, such as demand responsive transit system (DRTS) should be introduced into these areas so as to attract the commuters with rather diverse preferences. Finally, elasticity analysis and marketing share prediction are conducted to identify the key successful factors for public transportation. Corresponding marketing and improvement strategies for public transportation are then proposed accordingly.