Closed jpritikin closed 1 week ago
Thanks for the reports. This required some further modifications to lavaan, but I believe all your examples should now be working. Please try it out using my fork of lavaan, https://github.com/ecmerkle/lavaan, along with updated blavaan. Then I will send the lavaan PR once I test it out a bit more.
Okay! I'll try it. BTW, where can I learn more about ppp
? Is that what you are estimating in the post-sampling phase? I suspect it is a posterior predictive p-value, but I couldn't find it documented in the blavaan man pages. Did I miss something?
Yes, there is a posterior predictive p-value that is computed, and not much documentation about it. Some general detail about ppp appears in the appendix of the JSS paper. For models with ordinal variables, things get more complicated because there are multiple likelihoods that could contribute to a ppp. Some details about the likelihoods are here, and a student and I are assessing the best way to get a ppp out of all these.
Ah, okay. Good it know that ppp is not ready yet. I guess the only criteria that I need to report is zero divergent transitions, R-hat < 1.015, and tail/bulk ess > 100*numChains. I think my data is too small to expect much from the loo package. Any other model fitness measures I should be aware of?
Slowly working on this one. It is easy enough to turn off the check for this in lavaan, but then I encounter other errors elsewhere...
Thanks for letting me know that you're working on it. I can imagine that there is quite a lot of code to untangle. I really appreciate your efforts. I really need this model to work. I don't know how to analyze the anticipated data without this model.
Ok, try again with my fork of lavaan and github version of blavaan. I needed to turn off the variance check and then also turn off some starting value computations in lavaan. It is quite possible that I missed something else, especially for models beyond cfa.
I've run about 100 datasets. So far so good. Is there a way to turn off Computing post-estimation metrics (including lvs if requested
? This step seems to take a long time.
With missing data, I'm getting:
Here is my script again:
New simulated data with some missing entries are here: pm.csv psi.csv