Theory-driven structural equation modeling (SEM) is an increasingly popular technique for analyzing quantitative data in Information Systems research. Since 1994, 20% of all papers in top-tier journals use structural equation modeling [Urbach and Ahlemann, 2010]. Higher-order factor structures have been widely discussed from a number of theoretical perspectives [e.g., Bagozzi and Edwards, 1998; Hayduk, Ratner, Johnson and Bottorff, 1995; Law, Wong and Mobley, 1998]. Our intention is not to contradict these theoretical discussions but to postulate that empirical analysis can assist in the discovery of emergent higher-order structures where perhaps not initially proposed. As constructs have become both more numerous and specific, the existence of higher levels of multicollinearity, even when discriminant validity is evident, are becoming problematic for assessing the role of individual constructs. Thus, there becomes a need for a higher-order structure to represent these relationships. In this paper, we present a six-step methodology for researchers who have structural models suffering from multicollinearity, positing that multicollinearity can obscure an underlying higher-order structure. We present findings from an empirical study where multicollinearity had masked the presence of a higher-order structure. We conclude with a proposed methodology for discovering higher-order factor models.
Schwarz, Colleen; Schwarz, Andrew; and Black, William C.
"Examining the Impact of Multicollinearity in Discovering Higher-Order Factor Models,"
Communications of the Association for Information Systems:
Vol. 34, Article 62.
Available at: http://aisel.aisnet.org/cais/vol34/iss1/62