Further Considerations on Latent Segmentation Techniques for Customer Heterogeneity Detection

Latent segmentation procedures are usually aimed at detecting the heterogeneity of statisitical units with reference to a system of relationships described by a structural equation model with latent variables (SEM-LV). After a short review of hte main proposals given in the literature, we present a descriptive two-step procedure, wich first considers either an a-priori or a post-hoc segmentation approach: the former is based upon the existence of proper classification variables; the latter defines as classificaiton variables the latent scores estimated by a global SEM-LV model. In the second step a cluster analysis is performed by using the previously defined grouping vairables, in order to improve the classification by means of a cluster-wise Partial Least Squares algorithm; the classification refinement is executed with reference both to the SEM inner and outer models.