Modeling complex substitution patterns with variance and covariance heterogeneity in long distance travel choice models

The assumption of Independently and Identically Distributed (IID) error terms in the Multinomial Logit (MNL) model leads to its infamous Independence of Irrelevant Alternatives (IIA) property. Relaxation of the IID assumption has been undertaken along a number of isolated dimensions leading to the development of a rich set of discrete choice models, that are more flexible than the MNL model. In some cases, these more general models lose the mathematically convenient closed-form structure of the MNL.; This research integrates the most flexible closed-form extensions of the MNL and Nested Logit (NL) models in an integrated model structure to yield a behaviorally rich, yet computationally tractable choice model. Specifically, this research combines the Generalized Nested Logit (GNL) model, which allows for non-independent errors across alternatives; the Heteroscedastic MNL, which allows non-constant errors across observations; and the Covariance Heterogenous NL model, which allows for non-constant correlation structure across observations. The resulting model, called the Heterogenous GNL (HGNL) model further extends our ability to represent the complex behavioral processes involved in choice decision-making. It extends the state-of-the-art by accommodating a comprehensive treatment of variance-covariance error structure in a closed-form environment.; The value and need for the additional modeling complexity of the HGNL model is tested in the empirical context of mode and rail service class choice behavior for long distance intercity travel (more than 250 miles), a somewhat neglected area of research. The practical motivation for the study stems from the need to evaluate the potential demand for proposed new classes of rail service for long distance intercity travel. The proposed classes of service may extend or substitute existing classes of service (coach and sleeper class), which include an upgraded or premium coach service and an economy sleeper option. To accomplish this objective, rail travel demand is studied within a market research framework through stated preference surveys of approximately 1,500 long distance intercity travelers including current users and non-users of passenger rail service.; The empirical analysis is conducted using an incremental modeling approach, starting from the simple MNL model structure, and sequentially relaxing some of its restrictive assumptions to estimate progressively more flexible models. The statistical fit and behavioral appeal of the estimated models improve substantially with each additional relaxation, strongly supporting the concept of integrating isolated generalizations of the MNL/NL models. The final preferred HGNL model provides differential and more intuitive behavioral insights relative to the MNL/NL models and will therefore produce substantively different forecasts.