Abstract: The authors proposed a new travel mode choice model to overcome the limitation of traditional logit model. Combining generalized extreme value model and latent class model, the authors present a modeling methodology capable of accounting for individual preference heterogeneity and correlation across choice alterna- tives. Travel cost, travel time, parking fee, and waiting time are defined as utility variables for mode choice, while individual income, travel purpose, and travel distance are selected as variables of segment membership function. This model can depict the correlation among choice alternatives and individual preference heterogeneity simultaneously. Using Beijing traffic survey data of 2005, the model parameters are estimated. Estimation results show that the latent class paired nested logit model outperforms the traditional models. Most travelers are cost-sensitive to travel modes, and thus strategies that reduce the travel cost can be more effective than reducing the travel time.

Key words: travel mode, preference heterogeneity, correlation, latent class, paired nested logit model

摘要： 为克服传统logit模型的IIA缺陷, 构建合适的居民出行方式选择模型。尝试结合广义极值模型与潜在类别模型, 选取出行费用、出行时间、停车费用及等待时间等作为方式选择效用变量, 选取个人收入、出行目的与出行距离作为类属函数变量, 构建一种区分潜在类别的配对巢式logit模型, 该模型能同时刻画备选方式之间的相关性以及出行者的偏好差异。利用2005年北京市第三次居民出行调查数据, 对模型参数进行估 计和检验。参数估计结果表明: 1) 相较于传统MNL模型与不区分潜在类别的配对巢式logit模型, 区分潜在类别的配对巢式logit模型具有更优的统计学特征; 2) 对出行费用敏感的出行者比例大于对出行时间敏感的出行者比例, 提供交通服务时, 降低费用将比缩短时间更为有效。

关键词: 出行方式, 偏好差异, 相关性, 潜在类别, 配对巢式logit