2
$\begingroup$

How would I begin to determine if my model is "identified" in SEM process? I've already completed step 1 for model specification where I have created the path diagrams showing connections between observed and latent variables. I'm not sure how to do the second step, model identification.

$\endgroup$
1
$\begingroup$

Model identification is a pretty complex area.

The first thing to check is to count the degrees of freedom. Count parameters, count covariances (and mean)s. There should be fewer parameters than covariances (and means).

Then there are rules that you can learn - Bollen's book 'Testing structural equation models' covers them.

One problem is that there is model under-identification, and empirical under-identification. A model might be identified with one dataset, but not identified with a different dataset.

Here's a very simple example:

The simplest thing to do is to run the model. 99% of the time [wild guess there] the model will fail, with an error that says that the model is not identified.

The way to be sure that the model is identified is to rerun it with slightly different starting values. An identified model will always converge at the same place. A non-identified model won't.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.