AVE for the non-associated items

[elisabettar]

I am seeking your help concerning the validation of a measure I am developing. I have 4 factors and 12 items and I use AVE for evaluating construct and discriminant validity. I tested CR, CA, AVE. For discriminant validity I computed the correlation matrix where diagonal elements are square roots of the average variance extracted.

A reviewer told me to perform an additional analysis by computing the AVE for the non-associated items. This is his comment:

According to Fornell and Larcker (1981), the AVE is only useful as a metric if the loadings on the non-associated items are constrained to zero (the metric was developed for CFA, not EFA), which is not the case when the loadings are obtained from an EFA or from SEM approaches that are not covariance-based (such as PLS that was employed in the present study and does not constrain cross-loadings to zero). Hence, the AVE values for non-associated items should be computed and reported in addition, and they should be very small, preferably close to 0. The larger the difference between the AVE for the associated items and the AVE for the non-associated items, the more useful the AVE is as a metric to evaluate construct validity (both convergent and discriminant) in EFA or PLS. If the AVEs for the non-associated items are indeed close to zero and the AVEs of the associated items are above 0.5, then the AVE constitutes a good metric for construct validity evaluations. However, in the present paper it remains unknown whether the AVE constituted a good metric.

The AVE of the non-associated items ranges between 0.217 and 0.323. Are they too high?

Fornell, C., Larcker, D.F., 1981. Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18, 39–50.

Thank you very much in advance.

Elisabetta

[jmbecker]

It is an interesting suggestion by the reviewer, but as I have never seen such a test before, there are also no guidelines on what values are good or bad.
Moreover, the values of the AVE for the non-associated items may not be very useful without the information on the normal AVE. As the reviewer notes: "The larger the difference between the AVE for the associated items and the AVE for the non-associated items, the more useful the AVE is as a metric to evaluate construct validity.." [emphasis added].

However, to satisfy the reviewer, you might also consider using the new HTMT criterion to test discriminant validity. https://www.smartpls.com/documentation/ ... assessment