Interpretation of Fornell-Larcker Table

[kchristy]

I understand that the criteria for demonstrating discriminant validity with the Fornell-Larcker criterion is that if the AVE is greater than the squared latent variable correlations, then an assumption of discriminant validity has been supported.

What I'm having difficulty with, however, is interpreting the Fornell-Larcker table in the SmartPLS discriminant validity output. I know the AVEs are along the diagonal, but are they the "pure" AVEs or the square root of the AVEs, given that the correlations below appear to be the raw correlations, not the squared correlations?

[jmbecker]

SmartPLS 3 reports the square root of the AVEs along the diagonal and the "raw" correlation on the off-diagonal elements. If you compare the diagonal elements of the Fornel Larker table with the AVE values from the "Construct Reliability and Validity" table you can see the difference.

[kchristy]

Thank you so much! I was pretty sure that was the case, but wanted to be absolutely certain before I reported it.

[Ingrid Snijders]

What does it mean when the correlation (not the square root of AVE) has a negative value?