comparing Sem and Gsem with their AIC

[iman gharraie]

Hello,
I wanted to know if it is possible to compare a sem model with a gsem model by their AIC ? I tried a simple path estimation with both sem and gsem, they produce identical results however different AIC values, so does it mean that its wrong to compare two models in sem and gsem with AIC ?
Thanks.

[iman gharraie]

yes, they have totally different log-likelihoods, what is crazy here is, when I am doing a normal path analysis without any latent variables, I get different log-likelihoods between sem and gsem with the same estimates. however, if I try a sem model with latent variables, then both sem and gsem give me the same log-likelihoods, same AIC and same estimates!
if this normal?

[Phil Bromiley]

If you get identical results but different AIC, that reflects that GSEM makes different assumptions about the error structure than SEM. If you look carefully, you might find they have different numbers of parameters or different log likelihoods.