Concomitant Variables in Latent Change Models

A basic tenet in modeling preference behavior is that individuals differ in the ways they perceive and evaluate choice options. Latent-class analysis provides a parsimonious and flexible approach to represent these taste differences. This method decomposes a heterogeneous population of decision-makers into several homogeneous classes or subpopulations. Each decision-maker is assigned to one of the latent classes such that preference differences among members of different classes are maximized (Böckenholt, 1993; Croon & Luijkx, 1993; DeSarbo, Ramaswamy, & Lenk, 1993; Winsberg & DeSoete, 1993). The accuracy of this assignment can be greatly improved by taking into account person-specific collateral information (e.g., demographics). The consideration of collateral information is of particular interest in strategic marketing studies. If the segmentation results obtained in a latent class analysis can be characterized by demographic variables, marketing tasks such as positioning and targeting of consumer groups are much simplified (Dillon, Kumar, & Smith de Borrero, 1993; Gupta & Chintagunta, 1994).