I think it Is decided based on theory. So, what happens if the theory said that dimension reflectively measures higher-order construct BUT there are low correlations among the dimensions? If so, should we ignore the theory and go to a formative higher-order construct? And is there any cut point of correlations among the dimensions for a higher-order construct to be reflective?
@Hasan The basic difference between a reflective and formative scale is that reflective items are highly correlated with each other (within the scale) and consistently go in one direction thereby establishing the scale reliability and convergence, where as for a formative scale, items are not sufficiently correlated with each other and do not tend to show the consistency of the scale and neither tend to show convergence. In response to your query about the correlations cut-point for higher-order reflective constructs, usually correlations above .6 or .7 are considered enough to go for higher order latent constructs and the rules regarding the loadings of first-order (convergence) to second order latent construct are same as of loadings of observed to first order latent constructs (.708).
Reflective measures are caused by the latent construct, whereas, formative measures cause the latent construct. ... As researchers, we often rely on the testing and validation of published works using those same measurement models to theoretically test our structural (path) models.
Please make your choice a-priori based on theoretical considerations: Article Estimation issues with PLS and CBSEM: Where the bias lies!
Ex post you can use the confirmatory tetrad analysis in PLS-SEM (CTA-PLS) to test whether your measurement model is reflective or formative (e.g., via the CTA-PLS implementation in SmartPLS):
Article Confirmatory Tetrad Analysis in PLS Path Modeling
Book Advanced Issues in Partial Least Squares Structural Equation Modeling
Christian M. Ringle Dear Sir, thanks for adding your valuable answer and resources.
Dear Shahedul Hasan according to my understanding treatment of a construct as reflective or formative depends primarily on the Nature of the Construct with reference to a specific area of research and "common sense" developed through rigorous literature review of the topic. Statistical justifications I think come later. For example, if a researcher considering Subjective as well as Objective dimensions of Career Success, it would be unrealistic to put totally different scales as reflective measurement of the construct of Career Success. Tq
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence