Multinomial Probit with Structured Covariance for Choice Situations with Similar Alternatives

The estimatability of the multinomial probit (MNP) model has been improved greatly by the modeling of the error structure. A modeling idea for travelers’ choice behaviors using the MNP with structured covariances (MNPSC) model is proposed. Choice of route, destination, and time of day are considered. The covariance of the error term of the utility function is structured by introducing the overlapped length between any two routes and the overlapped time length between any two departure times. Simultaneous models such as route and train class, route and mode, and time of day and mode choice model also are introduced by applying the MNPSC modeling idea. These are formulated easily by assuming that the error is generated independently between two choices. An empirical study for a route and train class choice model was conducted by using the passenger survey data in the Tokyo metropolitan region, and its estimatability and applicability are examined.