Higher-order models with reflective indicators

Abstract

Purpose – Higher-order factor models have recently been dismissed as a ‘misleading’, ‘meaningless’, and ‘needless’ approach for modeling multidimensional constructs (Lee and Cadogan, 2013; L & C, 2013 hereafter). The purpose of this paper is to show that – in contrast to L & C’s (2013) verdict – higher-order factor models are still a legitimate operationalization option for multidimensional constructs. Design/methodology/approach – Basic conceptual and statistical premises of L & C’s (2013) arguments against higher-order factor models are scrutinized both conceptually and statistically as to their logic and validity. Findings – A thorough analysis of L & C’s (2013) arguments shows that they are fundamentally flawed both conceptually and statistically, rendering their conclusions invalid. Research limitations/implications – Researchers should not remove the well-established higher-order factor models from their methodological toolkit. Furthermore, empirical findings should not automatically be considered suspect simply because higher-factor models have been used to model multidimensional constructs. Originality/value – So far, L & C’s (2013) arguments against higher-order factor models have gone unchallenged in the literature. This rejoinder is a first, much needed attempt to protect applied researchers from getting the false impression that by using higher-factor models, they rely on a “misleading” or “meaningless” modeling approach.