Variance-Based Structural Equation Modeling: Guidelines for Using Partial Least Squares in Information Systems Research

Structural Equation Modeling In Information Systems Research

From a philosophical position, research that applies Structural equation modeling (SEM) usually follows a positivist epistemological belief (Urbach & Ahlemann, 2010). In this vein, SEM emerges as a result of the conjunction of two traditions (Chin 1998a). On the one hand, an econometric perspective (linear regression models), on the other hand, a psychometric approach (factor analysis). SEM thus combines the usage of latent (unobserved) variables that represent the concepts of theory, and data from measures (indicators or manifest variables) that are used as input for statistical analysis that provides evidence about the relationships among latent variables (Williams, Vandeberg, & Edwards, 2009). SEM is particularly useful in IS research, where many if not most of the key concepts are not directly observable. Indeed, a large portion of IS research during recent years has mainly applied SEM as an analytical methodology for theory testing (Gefen, et al., 2000; Gerow et al., 2010). In turn, an expansion to the breadth of application of SEM methods can be also noted, including exploratory, confirmatory and predictive analysis, as well as the generating of an increasing diversity of ad hoc topics and models (Westland, 2010).