Formative Measurement Models in Covariance Structure Analysis

Abstract

Many researchers seem to be unsure about how to specify formative measurement models in software programs like LISREL or AMOS and to establish identification of the corresponding structural equation model. In order to make identification easier, a new, mainly graphically orientedapproach is presented for a specific class of recursive models with formativeindicators. Using this procedure it is shown that some models have erroneously beenconsidered underidentified. Furthermore, it is shown that specifying formative indicators asexogenous variables rises serious conceptual and substantial issues in the case that theformative construct is truly endogenous (i. e. influenced by more remote causes). Anempirical study on the effects and causes of brand competence illustrates this point.