Contemporary Reporting Practices Regarding Covariance-Based SEM with a Lens on EQS

Introduction

“By the end of the 1990s, [covariance-based] SEM had ascended to the ranks of the most commonly used multivariate techniques within the social sciences,” (Hancock and Mueller 2006, p. 2). This interest in structural equation modeling (SEM) extended into the Information Systems discipline, but not without reporting deficiencies. As early as 1998, at the request of the editor-in-chief of MIS Quarterly, Wynne Chin was invited to submit a paper (Chin 1998) relating to the appropriate use of structural equation modeling. While the paper made a few references to Partial Least Squares (PLS), the paper itself was focused on what the IS field often refers to as covariance-based SEM. In 2000, Gefen, Straub, and Boudreau (2000) compared PLS, LISREL, and regression techniques, but positioned EQS and AMOS as a different type of analysis method from LISREL (ibid, p. 7). AMOS, EQS, and LISREL all belong to the same family of covariance-based structural equation modeling programs. LISREL (Joreskog and Sorbom 1984) began with the modeling of eight unique matrices. AMOS (Arbuckle 1989) extended this type of computational approach, but added visualization to improve modeling ease of use. While EQS and LISREL have added visualization modeling improvements, EQS (Bentler 1985) architected its SEM computational abilities along the equations-based modeling orientation already utilized by behavioral researchers.

Gefen et al. (2000) continued the no longer valid assumption (Treiblmaier et al. 2010) that covariance-based SEM cannot support formative constructs (ibid, p. 10 & p. 31). We now see formative measures used in a number of contemporary SEM studies (Diamantopoulos and Windlhofer 2001; Edwards and Bagozzi 2000; Kline 2006; Mackenzie et al. 2005; Qureshi and Compeau 2009). Unlike PLS (an analytical alternative to covariance-based SEM), EQS, AMOS, and LISREL share the factor analytic measurement model computation approach versus PLS’ principal components computation (Rigdon 1996).