Yeah, I understand why people use this, Mplus, but it's still strange people find this okay. One reason I enjoy Bayes is I don't have to do grid search for tuning parameters. But I didn't know I could do a 3-point search and be done with it.

SEM researchers create misspecification as: clean model and modeler misses important parameter, not so realistic IMO. But the problem is now a variable selection one and priors that do that work fine

Also, "small-variance priors" should learn variance not assume it's known, but Mplus can't do this

Looooool I think I might be one of those people who will still be there till its very last day.

ha! maybe, all right.

looool I'm still questioning your innocence in all this

my reaction was one ohhh after the other lol. I'm barely on here, so it's the first time I'm coming across this feature. Seems very useful, thank you!

It's been a very useful experience trying to implement the models in the package. The core of the models is pretty basic LISREL but there are lots of practical choices to make that make doing something like this worthwhile.

The biggest problem using the package is it relies on CmdStan which can be difficult for some to install or use (e.g. there may be a spaces-in-username problem lol). Also, there may be mistakes in the package implementation, but that's hard to fully resolve developing alone.

2. Bayesian analysis of correlation structures, including polychoric correlation matrix input
3. always assume total latent variable variance is 1, important for specifying loading, coefficient priors across different types of models

Other features but these are unique AFAIK