Revealing Individual Preferences with Bayesian Hierarchical Multinomial Logit Model

The primary goal of conjoint analysis is to use valuable information to take strategic marketing decisions. In this sense, it helps firms to answer the question how consumers will react to a new product without producing this product. To make this answer beneficial, it is necessary to use the appropriate models. For a long period of time, the analysis of consumers' aggregated preferences was popular. However, the world changed dynamically and the idea of ​considering people on average was realized to be a very poor description of the market. At the present time, individuals are especially important in all their diversity. One way to describe the heterogeneity of consumer preferences is to use the Bayesian hierarchical multinomial logit model. However, the lack of a free flexible platform for conducting such an analysis makes it unavailable for many researchers. Our goal is to change it and create a numerical model accessible to anyone who needs it. Moreover, the current paper analyzes the results, which created model leads to. A weak sensitivity of the posterior distribution to parameter`s priors was observed. Also, this study includes comparing various modifications of models and comes to a rather unexpected conclusion: not complete heterogeneity, but the presence of some number of clusters allows the model to give more accurate predictions.​