The GMNL model does not actually separate scale and preference heterogeneity as claimed. It is not a generalized version of the mixed multinomial logit model, but a more flexible version that still has limitations. Random scale and preference can only be identified separately under specific conditions within the logit family of models.