Formative measurement, where indicators are frequently seen as causing their corresponding latent variables, is extensively used in information systems research and in such a way as to attract methodological criticism to the entire field. This paper aims to ameliorate this situation.
Anchored on a new measurement residual theory, this paper argues that a latent variable always exists before the corresponding indicators when data is collected via questionnaires, whether reflective or formative measurement is used. Consequently, this paper posits that the direction of causality going from indicators to latent variables normally associated with formative measurement is misguided.
This paper develops a theory-driven set of recommendations for the assessment of formative measurement quality, addressing the following elements: factor reliability, indicator redundancy, significance of indicator weights, indicator effect sizes, Simpson’s paradox instances associated with indicators, model-wide factor redundancy and use of analytic composites.
The new theory and related recommendations are illustrated based on an empirical study of 290 geographically distributed product innovation teams that used various electronic communication media to conduct their work.
The data is analyzed with the software WarpPLS, a widely used structural equation modeling tool that allows for formative measurement assessment and analytic composite utilization in ways that are fully compatible with the theory-driven set of recommendations presented in this paper.
