JOINT SEGMENTATION ON DISTINCT INTERDEPENDENT BASES WITH CATEGORICAL-DATA

The authors discuss a latent class framework for market segmentation w ith categorical data on two conceptually distinct but possibly interde pendent bases for segmentation (e.g,, benefits sought and usage of pro ducts and services). The joint latent segmentation model explicitly co nsiders potential interdependence between the bases at the segment lev el by specifying the joint distribution of latent classes over the two bases, while simultaneously extracting segments on each distinct basi s. An EM algorithm is used to estimate the model parameters. The autho rs present an empirical application, using pick-any data collected by a regional bank on two popular, conceptually appealing, and interdepen dent bases for segmenting customers of financial services - benefits ( i.e., desired financial goals) and product usage (of an array of banki ng services). A comparative evaluation of the approach on synthetic da ta demonstrates the ability of the modeling framework to detect and es timate the interdependence structure underlying the two segmentation b ases and thereby provide more accurate segmentation than ''traditional '' (single-basis) latent segmentation methods.