Hello Till,

I'm not completely certain what your aim is here, as you have obviously modified the two measurement models based on the results of previous analysis(es). The inclusion of one covariance across two error terms (for AF01_01 and AF01_02) also suggests that you've used results from prior models to justify inclusion of this term.

So, any fit index you might identify is likely to be biased upwards since you've used the results from your data set to modify the model. In other words, unless you gather new data or use previously held-out data, you're likely to have overfitted and a fit index may be inaccurate for the target population.

The fact that the two constructs have entirely different variables associated with them and the constructs apparently don't correlate strongly (in your attached measurement model results) suggests that these are distinct entities. So, the matter of comparison of fit may not be an altogether useful enterprise.

In any event, your original approach to quantifying the differences in loadings across factors won't be a fruitful path, unless you went back what I presume was your original model wherein all variables are hypothesized to load on both factors.

Apologies if I've misunderstood your query.

Good luck with your work.

Hello David,

thanks for your answer.

Exactly, I modified the model, to achieve a satisfactory model fit. The original model consisted of seventeen indicators for both constructs, however, I "kicked" out indicators with a factor loading lower than 0.5 to achieve model fit. Same with the added covariance arrows.

I understand that the constructs are distinct entities, however, it would actually benefit (or at least be quite interesting for) my research if they have only a low fit.

I will follow your advice by using all variables, as seen in the second Screenshot, to measure the image fit index.

Again, thanks for your help!