Operationalizing Multidimensional Constructs in Structural Equation Modeling: Recommendations for IS Research

Abstract

Although scholars have provided advice regarding how to conceptualize multidimensional constructs, less attention has been directed on how to evaluate structural equation models that include multidimensional constructs. Further, the extant information systems literature has provided little, and sometimes contradictory, direction on how to operationalize multidimensional constructs. This gap in how we approach multidimensional constructs merits attention because: (1) establishing construct validity is critical to testing theory and (2) recent advances in software enable testing models with multidimensional constructs more readily. Therefore, this tutorial (1) describes different forms of multidimensional constructs and (2) illustrates how to integrate superordinate and aggregate multidimensional constructs in structural equation models. In doing so, we offer guidelines and examples for how to conduct and evaluate research using multidimensional constructs.

The effects of technostress and switching stress on discontinued use of social networking services: a study of Facebook use

“…The multidimensional stress creator constructs were conceptualized as superordinate constructs (reflective first‐order and reflective second‐order construct; Polites et al ., ) and evaluated as suggested by Wright et al . (). This means that we observed content validity, indicator validity, construct reliability and discriminant validity to validate the measurement model (Bagozzi, ).…”

Competitiveness and environmental performance in Spanish small and medium enterprises: is there a direct link?

“…Because the variable used to evaluate competitive performance was considered a subordinate construct 1 , the two-step approach described by Wright et al (2012) was followed. Thus, following these authors, two assessments of the measurement models (first and second order 2 ) were performed before testing the direct and mediating effects.…”

An explanatory and predictive model for organizational agility

“…The choice of PLS owes to the following reasons (Roldán & Sánchez-Franco, 2012): (1) The focus of the study is as explanatory as predictive of the main dependent variable; (2) the sample (n = 172) is not very large; (3) the research model is complex according to the type of relationships (direct, mediation, and moderation) within the hypotheses and the levels of dimensionality; (4) this study uses latent variables scores in the subsequent analysis of predictive relevance, particularly in the implementation of the two-stage approach for modeling multidimensional constructs (cf. Wright, Campbell, Thatcher, & Roberts, 2012); and (5) this research defines the nature of most theoretical constructs as defined: This approach relies on a composite measurement model with a reflective design approximation (Mode A), which means that indicators and dimensions represent different facets, although correlations exist among them (Henseler, 2014). Finally, Mode A is the advisable option for out-ofsample prediction when sample size is not large and correlation exists between the indicators and the dimensions (Becker, Rai, & Rigdon, 2013).…”