Multi-Group Latent Variable Models for Varying Numbers of Items and Factors with Cross-National and Longitudinal Applications

Varying sets of items and constructs are a problem frequently encountered in cross-national and longitudinal studies in marketing. We discuss the use of multi-group latent variable models in this situation and describe a method that can be used to handle unequal sets of items and constructs across groups in such models. A simulation study based on cross-national marketing data from Belgium and Great Britain revealed that accurate estimates of differences between latent means can be obtained with this procedure with as few as two common items, although a fairly large sample size is required to obtain small standard errors of the estimates of latent mean differences. A substantive example involving a confirmatory factor model as well as a structural model is also provided, using longitudinal data concerning the quality image of a food product in the Netherlands.