Reflective-Formative Construct - Unique Proportion of Variance

[fkreitne]

Hello together,

according to MacKenzie et al. (2011) the reliability of formative first-order constructs can be assessed by excluding one of the first-order constructs one after another to calculate the unique proportion of variance they explain in the second-order constrct. However, this recommendation is for CB-SEM only. Does a similar technique exist for PL-SEM?
The only comparable method that comes to my mind would be a redundancy analysis.
Since always 100 percent of the variance is explained in the second-order construct, when using the repeated indicator approach it is probably not possible to just exclude the first-order constructs, since the secondary loadings would also have to be deleted in the second-order construct, which then result again in an R-Square of 1.

Best regards,
Fabian

[jmbecker]

I think it is confusing to talk about reliability in terms of formative measures, because reliability is a concept that only works well for reflective items.

You are right that the idea of MacKenzie et al. (2011) relates best to the redundancy analysis. You are also right that in PLS it does not work to delete items and assess R² as the composite is always a complete function of the indicators and hence its R² from measurement model is always 1. But in a redundancy analysis you could delete dimensions/items and see how R² in the target construct changes. That would give you an intuition of the items relevance. However, that should also be closely connected to the strength and significance of the item weights in the original model.