forum.smartpls.com

Dear SmartPLS community,

Diamantopoulos, Riefler, and Roth have shown that a PLS path model with only formative latent variables (LVs) is non-identifiable. It is also written here on the discussion forum.

I love SmartPLS, but, recently, I discovered the 'plspm' R package too. It offers two modes: A and B, which correspond to reflective and formative measurement, correspondingly. How is that possible that the 'plspm' R package estimates models with only formative LVs? How could you explain that? Are the results there correct and reliable?

For example, a model with an exogenous second-order formative LV and an endogenous formative LV can be estimated in 'plspm' R package. Then, I also added moderation there.

I think that Diamantopoulos, Riefler, and Roth discuss the identification of model with formative measure in the context of CB-SEM not PLS-SEM.
In PLS-SEM you can have a model with only formative measures, because it is a composite-based estimator. Identification rules in PLS are different than in CB-SEM and you can also specify and estimate a model with only formative measures in SmartPLS.