Andy Luse posted on Wednesday, November 16, 2016 - 3:18 pm
I have a model that has four latent variables that have both individual and between level variance. Typically when I look at psychometrics I run a measurement model and use the standardized loadings to get the average variance extracted and composite reliability. I obtained CR and AVE values at the between level, but when looking at discriminant validity at the between level, the correlations between the variables is higher than some of the square root of the AVE values. Is this because the within level accounted for much of the residual variance? Maybe I shouldn't even be doing it this way? Should I be getting separate CR and AVE values at both levels for each variable? I can't seem to find any research describing what to do.