A comparison of generalized multinomial logit (GMNL) and latent class approaches to studying consumer heterogeneity with some extensions of the GMNL model by J. Pancras and D.K. Dey

J. Pancras and D.K. Dey written a paper to demonstrate a different version of the generalized multinomial logit (GMNL) and the widely used latent class multinomial logit (LCMNL) approaches to studying heterogeneity in consumer purchases. They have demonstrated that the GMNL model outperforms the best-fitting LCMNL model for a panel dataset of household ketchup purchases. They have also described extensions to the scale heterogeneity model that include the effects of state dependence and purchase history. The authors have proposed an interesting simulation approach by means of the Bayes' rule, to compute individual level beta coefficients. It will be interesting to compare their results with those It will be interesting to compare their results with those, as the simulation approach is conditional on the maximum likelihood estimates. The proposed approach is also simpler to implement, but the performance of the procedure needs to be investigated in detail.