Thanks - I wondered if it were possible to get them from ESEM, when I have additional parameters (such as theoretically based correlated errors) which I can't model with EFA. If it's possible, I wanted to plot a scree plot with and without additional some additional regression parameters.
The idea of this is to show that a multi-factor solution is an artifact of a couple of very similar items - specifically that item order influences answers.
I tried things like calculating the eigenvalues using the residuals or the sums of squared (standardized) loadings, but I don't think the 'true' sum of eigenvalues is the number of items, as the additional regression parameters. The eigenvalue of the first factor extracted is lower when there are regression parameters in there, but I want to scale that up, to make it a proportion of the sum of the eigenvalues.
Does that make sense? If it does, is it possible, or do I need to think of a different way to make my point.
Sounds like you want eigenvalues for a model-estimated covariance (or corr) matrix. You can print the matrix and then submit it to a program getting eigenvalues. But I'm not sure that answers your question - seems like your point is more easily made by working with say BIC as a result of using fewer factors.