AN EMPIRICAL COMPARISON OF EXPLORATORY VERSUS CONVENTIONAL STRUCTURAL EQUATION MODELLING

In this paper we discuss two SEM approaches: an exploratory structural equation modelling based on a more liberalised and inductive philosophy versus the classical SEM based on the traditional hypothetical-deductive approach. We apply these two modelling techniques to data from a consumer survey and compare them based on several criteria, such as coefficients, parsimony, model fit, plausibility, and consistency with the theory. A comparison of the estimates obtained from the two models clearly indicates that it is not at all a trivial matter whether cross-loadings are allowed in a measurement model or not. The results shed serious doubt on the generally accepted rule of thumb according to which (cross) loadings can safely be ignored if they do not have a "practically significant" loading with an absolute value of at least 0.30 or 0.40.