Latent class analysis for evaluating a multi-item scale to measure customer satisfaction with reference to a shopping good: a pair of branded jeans

In the field of marketing many objects of interest exist that are not directly observable, nevertheless they can be measured through multi-item measurement scales. As a consequence, this kind of instruments are extremely useful and their importance requires an accurate development and validation procedure. The traditional marketing literature highlights specific protocols along with statistical instruments and techniques to be used for achieving this goal. For example, correlation coefficients, univariate and multivariate analysis of variance and factorial analysis are widely employed with this purpose. However, these kind of statistical tools are usually suited for metric variables but they are adopted even when the nature of the observed variables is different, as it often occurs, since in many cases the variables measured by the items of which the scale is made up are ordinal. On the contrary, latent class analysis takes explicitly into account the ordinal nature of the observed variables and also the fact that the object of interest, that has to be measured, is unobservable. The aim of this paper is showing how latent class analysis can improve the procedures for developing and validating a multi-item measurement scale for measuring customer satisfaction with reference to a shopping good that is a good characterized by a high level of involvement and an emotional learning, linked to the lifestyle of the customer. This latent class approach explicitly considers both the ordinal nature of the observed variables and the fact that the construct to be measured is not directly observable. Especially, applying appropriate latent class models, important features such as scale dimensionality, criterion and construct validity can be better assessed while evaluating the scale.