PPP as computed in Mplus for SEM is based on fitting the sample mean and variance, assuming independent observations.
In DSEM the observations are not independent across time and the models are fitting means variances and auto correlations. New posterior predictive checking must be developed that is appropriate for that model. It has not been done yet.
That is correct. It is not possible to use that same methodology because the fit for Means and Variances is not sufficient given the time-series nature of the data where covariance across time is in play. I think that the current trend is to estimate as unrestricted model as the data allows and then walk it back to obtain a more parsimonious model by removing insignificant effects. You can use credibility intervals, model constraints, model test and DIC for testing of nested models.
AH posted on Wednesday, January 22, 2020 - 12:53 am