Formative Versus Reflective Measurement for Multidimensional Constructs

Abstract

Information systems researchers often conceptualize constructs as higher-order multidimensional entities. As is the case with the specification of first-order dimensions, researchers seem to have two basic choices regarding measurement specification at the second-order level: to operationalize second-order factors as either formative (molar) or reflective (molecular). Although several researchers have noted the importance of measurement misspecification when testing measurement models for first-order constructs, limited research has investigated the consequences of measurement misspecification at the second-order level. In this article we empirically investigate whether different measurement specifications at the second-order level can lead to different inferences. Exploiting data from an empirical study investigating the notion of electronic service quality, we test the multidimensionality of the construct, using two different measurement specifications: the formative model which assumes molar relationships and the reflective model which assumes molecular relationships. We find that there is poor convergence between the two models in terms of resulting conclusions, highlighting the need for more careful measurement specification of IS constructs at a higher-order hierarchical measurement level. Furthermore, it seems that the choice of analytical methodology (i.e., PLS versus Covariance-based SEM) is also likely to influence results. We end up with implications for research regarding higher-order constructs measurement.