How should I know whether my model is under identified, just identified or over identified in SEM? As per standard method I am interested to find the just identified model. How could I do that ?
Hi Dear Sir,
You mean if I run the estimate and it give me these values (as you mentioned above), it would mean that the model is identified, irrespective of the fact whether chi square value is insignificant/significant or the model fit is good or poor. Is that what you mean ?
Thank you for your response, but still I am confused. What I got is that if the model is just identified , it will run, otherwise it will not run both in case of over-identified and under-identified... Am I right ?
What statistical program are you using?
Rimantas' explanation is correct. You want an over-identified model, and it would be best to compare several models that are theoretically defensible to see which has the best fit.
Are you having problems with your model converging? It is also possible that a model won't converge based upon a number of factors, such as lack of discriminant validity among latent variables, Heywood cases, etc.. If you are having problems with model convergence, you can check for these issues.
I am intended to use AMOS software for this purpose. Currently I am trying to learn this technique, I will start my analysis of the data later on. You mean i should estimate many models and chose the one with best fit indices? I am following the book of Randall E. Schumacker and Richard G. Lomax (A beginner's guide to Structural Equation moedling, 3rd Edition)
Many potential models can fit your data well, so a good approach is to compare two or more models to determine which has the best fit. I haven't read the book you cite - I'm sure it's fine. I would also check out Kline's book on SEM, and also David Kenney's website, which has many great explanations of issues, fit tests, and general methods. . http://davidakenny.net/cm/causalm.htm
Yes, it is Kline's recommendation to select alternative models to compare. It isn't a random choice - they should be theoretically informed.
Hi all and Remantas,
As you said that "difference between over-, just and under-identification will become clear to You very quickly". For the IRT models that I have fitted so far, it is only clear to me whether under-identified versus just- or over-identified, but it is not clear to see just- and over-identified.
Could you make your statement more clearly to me? Thank you so much!
P/S: I fit IRT models in Bayesian way.
Tran.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling