Predicting factors scores after CFA

[Adrian Pfammatter]

Hello

I am currently working on confirmatory factor analyses with my survey data using the command -sem-.
After the actual analyses, I tried to extract factor scores for my latent variables using -predict, latent-.

There are no issues with the commands but I was very confused about the variances of the obtained factor scores and their correlations.
The reason is that they differ from the output of the -sem- command for my original CFA.

Could anyone please explain to my why the -predict, latent- command produces factors with different variances and correlations than in the models?
Thank you .

[Alecia Cassidy]

The sem command estimates the covariance matrix for the true latent factors, whereas I think what you are doing is getting estimates of the true latent factors and finding the covariances of those estimates.

[Adrian Pfammatter]

Thank you very much for your comment: I just don't see why the estimates (covariances, variances) from the sem command in the model should differ from those that I get when extracting the factors and calculating the estimates again ('by hand'). In believe the retrieved estimates should still be based somehow on the original covariance matrix?