以成對組合羅吉特模式探討城際間運具選擇行為之研究

Abstract

個體選擇模式已被廣泛的應用在交通運輸與其他領域。多項羅吉特（Multinomial Logit）模式是最常被使用的個體選擇模式，因為它的數學型式簡單且易於校估，但缺點是模式具有不相關替選方案獨立特性。最常用於克服不相關替選方案獨立性的是巢式羅吉特（Nested Logit）模式，其可將具有某種程度相關的方案置於同一巢。但在許多情況下，巢式羅吉特也無法反應決策者真實的選擇行為，因為其假設同一巢內方案間具有同等的相似程度。最近被發展出來的成對組合羅吉特（Paired Combinatorial Logit）模式能考慮方案間兩兩的相關性，又比多項與巢式羅吉特模式更具一般化，且易於校估。 本研究探討成對組合羅吉特模式之理論架構、特性與校估方法，並分析臺灣地區城際間運具選擇行為。本研究利用顯示性偏好法與敘述性偏好法蒐集資料，以台北至高雄城際間旅運者為研究對象。考慮的運具包括航空、臺鐵、國道巴士、小汽車與目前尚未完工之高速鐵路。 實證研究結果發現，運具服務水準、個人所得、性別及同行人數是影響城際運具選擇的重要變數。成對組合羅吉特模式同時考慮航空、鐵路及巴士運具間之相似特性，其解釋能力遠優於多項與巢式羅吉特模式。本研究結果可作為規劃者進行運輸需求預測與研擬策略之參考。

Discrete choice model has been widely used in transportation and other fields. The multinomial logit (MNL) model, most used discrete choice models due to its mathematical structure and ease of estimation, has the independence of irrelevant alternative (IIA) property which limits its applicability in many cases. The most widely known relaxation of the IIA property is the nested logit (NL) model which allows interdependence between pairs of alternatives in common groups. However, the restriction on equality of dependence between all pairs of alternatives in a group may be unrealistic in some cases. The paired combinatorial logit (PCL) model which relaxed these restrictions by allowing differential correlation between pairs of alternatives retains the computational advantages of other logit models and has more flexible error correlation structure than the MNL and NL models. The purpose of the research is to describe the structure, properties and estimation approach of the PCL model and to investigate intercity mode choice behavior in Taiwan. This study collected revealed-preference and stated-preference mode choice data in the Taipei-Kaohsiung corridor. The data set contains air, train, bus, car and high-speed rail. The empirical analysis demonstrates that level of service variables, income, gender and group sizes are important factors affecting intercity mode choice. The PCL model with similarity parameters for both air and train and air and bus is statistically and structurally superior to the MNL and NL models. The results provide important implications for travel demand forecasts and policy decisions.