Comparison of Multinomial Logit and Mixed Logit
This paper concentrates on the comparison of different discrete choice models using data concerning purchases of different brands of crackers in USA. Five logit models are proposed: multinomial logit (MNL), mixed logit (ML), multinomial logit which accounts for heterogeneity between the customers (MNLH), multinomial logit extended with a brand loyalty variable (MNLB), and multinomial logit with a combination of the two extensions to MNL mentioned above (MNLC). Using the Cracker data set from the Ecdat package in the R software, it was found that ML outperforms all the other models, and that the extensions applied to MNL help in improving MNL. Therefore, it was concluded that using a mixed logit model is the best way to model an unordered categorical dependent variable.