Abstract
This paper derives the exact necessary and sufficient constraints on the dissimilarity coefficients in the Generalized Nested Logit (GNL) model. Previous studies have assumed the Nested Logit (NL) model’s dis similarity coefficients lie within the unit interval, consistent with random utility maximization. However,those previous studies relaxed constraints on dissimilarity coefficients only for the NL model. We derived constraints on the dissimilarity coefficients for the GNL model. We tested the consistency by deriving the constraints of the NL model from those of the GNL model. Then, we proposed the parameter estimation method under derived constraints. We performed the parameter estimation using POS data from supermarkets. As a result, we obtained higher likelihoods and catching rates compared with the previous conditions.
