Alternative Model Forms and Statistical Approaches
This chapter examines some alternative estimation procedures and model forms. It deals with the functional form of the representative component of utility given by in the logit model. An alternative logit formulation with potentially less restrictive assumptions on the relationship between alternatives i and j is the dogit model, proposed by M. Gaudry & M.G. Dagenais and tested using both aggregate cross-sectional and time series data by Gaudry & M. J. Wills. Another alternative to the multinomial logit (MNL) model is the multinomial probit (MNP) model which is more general than the MNL model yet involves considerable computational complexity. The MNP model relaxes the assumption of IID random terms in equation so that there may be some covariance between alternatives i and j. A potentially useful aspect of the dogit model is that it allows some pairs of alternatives to exhibit the independence from irrelevant-alternative property, while other pairs may not exhibit the IIA property.