Latent Class Analysis for Marketing Scale Development

Abstract

Measurement scales are a crucial instrument in marketing research for measuring unobservable variables such as attitudes, opinions and beliefs. In using, evaluating or developing multi-item scales, a number of guidelines and procedures are recommended, to ensure that the measure applied is psychometrically robust. These procedures have been outlined in the psychometric literature since the late 1970s and are composed of steps that refer to construct and domain definition, scale validity, reliability, dimensionality and generalisability. Various statistical instruments are used in the scale-developing process, almost always referring to metric variables (interval or ratio scales). Instead, items forming scales are rarely measured metrically; items are frequently ordinal and, in some rare cases, nominal. In this paper, it is shown how the implementation of latent class analysis may improve the process of measurement scale development, since it explicitly considers that items generate ordinal or even nominal variables. Specifically, applying appropriate latent class models allows us to assess scale validity and reliability more soundly than traditionally used methods.