Content area
Full Text
Introduction
Structural equation modeling (SEM) has become the norm for analyzing cause–effect relationships between latent variables. When conducting business research, authors often desire to test complete theories and concepts. This is one of the major reasons why they have embraced SEM (Babin et al., 2008). While SEM is a general term encompassing a variety of statistical models, covariance-based SEM (CB-SEM; Jöreskog, 1978) is the more widely used approach, with many researchers simply referring to CB-SEM as SEM. However, research has suggested alternative SEM techniques, most notably partial least squares structural equation modeling (PLS-SEM; Lohmöller, 1989; Wold, 1982), which has recently come to the attention of a variety of disciplines, such as hospitality management, human resource management, international business, marketing and tourism (Ali et al., 2018a,b; do Valle and Assaker, 2016; Hair et al., 2012b; Richter et al., 2016; Ringle et al., 2018). Reviews of PLS-SEM use in these fields show univocally that researchers often justify their choice of method on the grounds of their measurement model specification, because PLS-SEM allows for estimating formatively specified constructs without identification concerns (Sarstedt et al., 2016b)[1].
Unlike their reflective counterparts, indicators in formative measurement models form the construct by means of linear combinations (Diamantopoulos et al., 2008). For instance, a change in a respondent’s assessment of the trait that an indicator is capturing alters the indicator’s value, which, in turn, results in a change of the construct’s value. Formative measurement has become important in business research (Bollen and Diamantopoulos, 2017; Diamantopoulos et al., 2008), as it allows for isolating indicators’ relative impact when forming a construct (i.e. in the context of a certain nomological net represented by the PLS path model), which makes more nuanced managerial recommendations possible (Albers, 2010).
The convergent validity assessment is a requirement for the empirical evaluation of formative measurement models in PLS-SEM. Convergent validity is the extent to which a measure relates to other measures of the same phenomenon (Hair et al., 2017a, Chapter 5). A redundancy analysis, in which each formatively specified construct is correlated with an alternative measure of it, provides this assessment (Chin, 1998). The result allows for determining whether, jointly, the formative indicators represent the construct of interest adequately.
When running...