CMStatistics 2016: Start Registration
View Submission - CFE
Title: Consumer clustering model based on the time of new product adoption using ID-POS data Authors:  Masataka Ban - Nihon University College of Economics (Japan) [presenting]
Abstract: In marketing literature, the consumer behavior of selecting one brand from various goods is called ``brand choice''. For consumer heterogeneity, in general, brand choice behavior is modeled by hierarchical Bayes discrete choice model like logit or probit which have consumer's individual-level parameter. A brand choice model is proposed for consumer clustering in terms of a new product adoption. In particular, the model is constructed by hierarchical Bayes probit model having a Dirichlet process (DP) prior with time ordering clustering constraint. Features of this model is that (1) the model enables the estimation of the number of clusters, and then it is not necessary to set that before analysis. (2) Time ordering clustering leads to estimation of breakpoints among consumer clusters. The consumers are categorized into an adequate time ordering cluster based on the similarity of their market response. The model estimates provide useful information corresponding to the marketing concepts containing time ordering clusters (e.g. Roger's diffusion of innovation theory, product life cycle management). The model is estimated by Markov Chain Monte Carlo sampling method. A Metropolis-Hastings-based algorithm modified to fulfill the constraint is used.