意愿调查设计导致离散选择模型参数估计偏差的实证分析

An Empirical Analysis on the Willingness Survey Design Leads to Bias in Discrete Selection Model Parameter Estimation

为深入研究不同意愿调查问卷实验设计法导致离散选择模型参数估计偏差的原因,以成都市中心城区与卫星城龙泉驿区间新增地铁为研究对象,分别利用正交设计法与D-efficient设计法设计两种意愿调查问卷,获取出行者偏好数据。采用效用均衡法计算两组调查问卷的效用均衡值,定量化描述调查问卷中是否存在包含主导项的选择情景;利用调查数据标定Nested Logit Trick(NLT)模型来验证是否由于主导项的存在引起误差项方差变化,从而导致模型参数估计结果产生偏差。研究结果表明,D-efficient设计问卷中含有主导项的选择情景数量较正交设计问卷多,导致效用函数误差项方差越小,尺度参数值变大,进而使模型参数估计值存在显著的偏差。

In order to analyze the reason that the different willingness survey designs would bias the parameter estimates in discrete choice model.In this paper,the new metro between a centre city and its satellite city was selected as the research subject to study this question.The Orthogonal Design and D-efficient design will be used to design questionnaires respectively and travellers’preference would be obtained by these questionnaires.The utility balance will be introduced to quantitatively describe the dominant alternatives in the different designs.Meanwhile,the Nested Logit Trick model is used to test whether the bias is induced by the changes of error variances resulting from dominant alternatives.Finally,the results show that D-efficient questionnaires have much more dominant alternatives than Orthogonal Design which will lead smaller error variance and larger scale parameter,the bias of the parameter estimates is more significant.