Hello,
I'm faced with a problem I'm currently not able to solve: Within a reflective measurement model (3 constructs) all evaluation criteria are perfect (high AVEs, high and significant loadings, high Cronbach's Alpha), high composite reliability - everything far better than 0.9 - BUT testing for discriminant validity, my AVEs are NOT higher than the squared latent variable correlations (which are themselves higher than 0.9).
What does this mean, and what can I do about it?
Could one reason be that all constructs (derived from a principal component analysis) are components of the same "higher" latent variable? Just to explain it quickly: I want to measure the creativity-support of a work environment, but theoretically, it is not possible to include all relevant parameters into a formative model (there are much too many supposed in literature) - that's why I reduced the parameters I found (limited to 15) to underlying components (the above-mentioned 3 constructs) which then can be measured reflectively - so all the constructs are part of a "higher" construct. Could this be the reason for the lacking discriminant validity, and can I use this as an explanation/reasoning to ignore it?
Thanks in advance for your help!
Best regards,
Evelyn.