Estimation of saturated model for LRT PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
 Anonymous posted on Wednesday, June 29, 2005 - 5:17 pm
Hello -- I am fitting a standard growth curve model to 10 repeated measures organized on chronological age, not wave of assessment. Given this design, I have no observed covariances between a number of pairs of ages (e.g., age 5 and age 15, age 6 and age 16, etc).

In Mplus 3 I get a saturated model & an omnibus chi-square. In Amos 4, Amos 5, and LISREL 8.72, the saturated model can not be estimated, and I get a larger model df in Amos than in Mplus. The difference in df is precisely equal to the number of covariance elements with no paired observations.

Could I please ask if you could briefly describe how the saturated model is being estimated in Mplus 3? Am I being charged a df for each cell with no observed coverage?

Thank you very much for your time.
 bmuthen posted on Saturday, July 02, 2005 - 6:11 pm
If I recall this correctly, the saturated (H1) model estimates for covariance matrix elements that have exactly zero coverage are not meaningful, but they also do not hurt the model estimation and the correct chi-square test of H0 with the correct number of df is produced because these elements are not contributing to the H0 model fitting. The df count should be equal to the number of H1 parameters minus the number of H0 parameters - if you don't include the cells with zero coverage in this count, which in my opinion one shouldn't, you should get what Mplus prints.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message