Assessing the unidimensionality of measurement: A paradigm and illustration within the context of information systems research

The development and psychometric evaluation of scales which measure unobservable (latent) phenomena continues to be an issue of high interest among researchers within the information systems community. Accurate measurement of structurally complex constructs provides a potentially powerful means for empirically exploring relationships between information technology and individual, organizational, and industrial phenomena. In exploratory contexts, measurement properties of psychometric scales are evaluated using traditional techniques such as item-to-total correlations, reliability analysis, and exploratory factor analysis. In instances of strong theoretical rationale, contemporary techniques, such as confirmatory factor analysis, are utilized as a means of assessing model efficacy. An essential, but often overlooked, property of measurement which is assumed in both exploratory and confirmatory statistical techniques is unidimensionality. Scales which are unidimensional measure a single trait. This property is a basic assumption of measurement theory and is absolutely essential for unconfounded assessment of variable interrelationships in path modeling. In this paper, a paradigm for developing unidimensional scales is presented and illustrated. Built on similar frameworks within the disciplines of psychology, education and marketing research, this paradigm is offered as a means of formally defining unidimensionality, distinguishing the concept from traditional reliability-based metrics, and describing a structured technique for empirically testing its existence. (C) 1997 Elsevier Science Ltd.