Ordinal Data EFA - PML estimator takes too long

[Zhao]

Hi all,

Greetings. I am a newbie using lavaan here. I am using cfa() function to perform a CFA on ordinal data using the efa("efa")*f1 + efa("efa")*f2 + ... syntax based on the article here https://solomonkurz.netlify.app/blog/2021-05-11-yes-you-can-fit-an-exploratory-factor-analysis-with-lavaan/. I have around 527 observations with 146 variables.

By default, lavaan uses DWLS as the estimator running

lavaan::cfa(model = model_expression, data = df, rotation = "oblimin", ordered = TRUE)

It works fine but I was unable to compare models using AIC() and BIC() because DWLS is not ML-based.

So, I tried the pairwise ML as the estimator:

lavaan::cfa(model = model_expression, data = df, estimator = "PML", rotation = "oblimin", ordered = TRUE)

So far, the PML job has run for around 16 hours and is still running (compared to around 1 hour completion for the DWLS).

I am wondering if there is any tip to speed up the processing? Will multi-CPU help? Or using GPUs or other parallel processing?

Otherwise, is there another estimator in lavaan that allows me to use AIC() or BIC() with ordinal data?

Another question: According to https://lavaan.ugent.be/tutorial/tutorial.pdf, page 29, at the very bottom of the page, it says " Full information maximum likelihood is currently not supported." for ordinal data, is that correct? 

Thanks!

Regards,

Zhao