Empirical Agreement Between Formative and Reflective Measurement Models: A Monte-Carlo Analysis

Abstract

Diamantopoulos and Winklhofer’s (2001) paper on formative index construction in the Journal of Marketing Research changed the perspective of many marketing researchers on construct measurement. Since 2001, a significant stream of literature has developed which explores the extent to which existing reflective measures should be specified as formative. This work raises the possibility of significant negative consequences arising from the so-called ‘incorrect’ specification of measures as reflective when they should be modeled as formative. For example, Jarvis, MacKenzie, and Podsakoff (2003, p. 212) suggest that if a formative measure is specified as reflective, “the structural parameter estimates within that model [may] exhibit very substantial biases that would result in erroneous inferences.” This paper examines the evidence for such claims, first by considering the nature of validity, then by conducting a simulation analysis of the effects of reflective treatment of formative indicators. The results demonstrate that despite their conceptual differences, and the results reported in prior empirical literature, formative and reflective specifications may reveal very similar substantive relationships between constructs when the comparisons are made appropriately.