Dear Muthen and Muthen, I implemented a non-linear latent growth curve model that has one time-invariant and 5 time-varying predictors. The model has 11 time points and involves a sample 109 countries. The resulting output is generally plausible, especially in terms of the individual predictors. The major challenge I am encountering, though, is that only one of the goodness of fit statistics meets the conventional cutoff criteria, that is, the SRMR. I have carried out several theory-driven modifications of the model and still came up short in terms of the other model fit statistics. I just can't seem to find a model that fits the data better than the one I currently have. Now here are my questions:
1) I have read elsewhere that what I discussed above tends to be the case with complex growth curve models, but I am wondering if this is true with models estimated with Mplus. 2)Do you have any suggestions about improving the model fitness? 3) Are there studies using Mplus for growth curve modeling that you will recommend I review?
1) There is nothing Mplus-specific to the growth modeling.
2) You can try to add residual covariances among adjacent time points. If N=109, however, you don't have much power to reject the model so the model misfit would seem to be large. You can try using Modindices (see UG).
3) See our Short Course videos and handouts for Topics 3 and 4 on our website.