Correspondence: Jaime R S Fonseca, ISCSP — Higher Institute of Social and Political Sciences, Technical University of Lisbon, CAPP — Centre for Public Administration and Policies, R. Almerindo Lessa, Lisbon 1300-663, Portugal. E-mail: jaimefonseca@iscsp.utl.pt

¹is currently Professor of Statistics/Data Analysis, at Technical University of Lisbon, Institute of Social and Political Sciences–ISCSP. He is the author of two books on Statistics and some technical papers, and has a MSc in Data Analysis and Computation, from Sciences Faculty of Lisbon University. He is a doctoral student in quantitative methods at the ISCTE-Business School, Department of Quantitative Methods, Lisbon, Portugal.

²is an assistant professor at the ISCTE Business School. She holds a degree in Mathematics and a Master's degree as well as a PhD in Systems Engineering, from Lisbon Technical University. Her research interests include multivariate statistics and machine learning techniques in marketing research (segmentation and positioning, in particular).

Received 9 August 2007; Revised 9 August 2007.

Abstract

Stability is a desirable property of a segment structure that has implications in its managerial utility, particularly in what concerns targeting and positioning. The stability of a segment structure is the focus of the present work. We argue that the evaluation of stability should be preceded by an adequate segmentation methodological approach that addresses satisfactorily the issues of selection of segmentation base variables, modelling and determination of an adequate number of segments. We advocate the use of the latent segments model approach (estimation of finite mixtures) and the selection of the best models to be based on the ICL-BIC, CAIC, BIC and L information criteria, which evidence some advantages when dealing with mixed-type variables, commonly used in segmentation. We then address the evaluation of stability. Internal stability is evaluated using split samples procedures and replicating segmentation results. Dynamic stability evaluation relies on analysis of data related to different time periods. An application concerning the segmentation of customers of a supermarket chain illustrates the proposed approach. Two databases related to questionnaires conducted in 2000 and 2003 (with 3,141 and 1,504 observations, respectively) are used. The obtained two-segment structure —Preferential Customers and Occasional Customers— exhibits internal and dynamic nonstability.