CFA model with categorical items using MLR estimator

369 views
Skip to first unread message

Shubhabrata Mukherjee

unread,
Jun 5, 2019, 1:10:10 PM6/5/19
to lavaan
Hi All,

 I have a cognitive data set with around 2400 individuals and 10 items. Seven of the items have good coverage (non-missing on more than 2300 individuals). Three of the items are non-missing on round 700 individuals. All of the items are categorical in nature.

 When I ran a CFA in Mplus, I had to use MLR. A run with WLSMV as the estimator would not converge due to pairwise missingness.

  I want to run this model using lavaan. Can I run a model with all categorical items using MLR estimator in lavaan? If this feature going to be added soon if it is not there yet?

 Thanks,

Shubhabrata

Mauricio Garnier-Villarreal

unread,
Jun 5, 2019, 1:57:09 PM6/5/19
to lavaan
Shubhabrata

First a clarification on what is Mplus doing, with estimator = MLR and categorical data Mplus is NOT using MLR as the estimator, actually is using Marginal Maximum Likelihood (MML), which is the ML estimator for IRT models. This can handle missing data in categorical data.

Now, in SEM family of analysis, WLS cannt do FIML to handle missing data because is not an ML method, you would have to use Multiple Imputations and then do the analysis with the runMI family of functions in semTools andWLSMV

If you want to use the MML method, now in R the best option would be the package mirt. I believe lavaan has an estimator="MML", but this is really slow (at least the last time I use it)

Balal Ezanloo

unread,
Jun 6, 2019, 7:55:37 AM6/6/19
to lavaan
Hi Mauricio 

I read your post. are you sure MLR estimation method in MPLUS is the same as MML and FIML  in IRT? do you have any reference for this claim?

Terrence Jorgensen

unread,
Jun 6, 2019, 8:24:25 AM6/6/19
to lavaan
are you sure MLR estimation method in MPLUS is the same as MML and FIML  in IRT? do you have any reference for this claim?

Yes, but they are deliberately vague about how they actually program (their software is not open source, and they don't want anyone else providing competition by implementing the same algorithms they design).  I've never seen them explicitly call it MML, but their description of it is exactly how MML works.


In Ch. 8 of their user guide, scroll down to the ESTIMATOR section to find a big table.  


Notice the asterisk by ML(R/F) under the column when at least 1 variable is categorical, indicating that numerical integration is required (i.e., how MML operates in practice).

Terrence D. Jorgensen
Assistant Professor, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam

balal izanloo

unread,
Jun 6, 2019, 8:49:09 AM6/6/19
to lav...@googlegroups.com
OK, this is Great. Thanks for your clarification. if MLR=MML so what is the difference between them with FIML method? please let me know.

Best Regard

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To post to this group, send email to lav...@googlegroups.com.
Visit this group at https://groups.google.com/group/lavaan.
To view this discussion on the web visit https://groups.google.com/d/msgid/lavaan/9d85581a-686f-4d56-a2f8-9d53048d420f%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Mauricio Garnier-Villarreal

unread,
Jun 6, 2019, 3:35:41 PM6/6/19
to lavaan
with MML, the handling of missing data would be functionally FIML



On Thursday, June 6, 2019 at 7:49:09 AM UTC-5, balal izanloo wrote:
OK, this is Great. Thanks for your clarification. if MLR=MML so what is the difference between them with FIML method? please let me know.

Best Regard

On Thu, Jun 6, 2019 at 4:54 PM Terrence Jorgensen <tjorge...@gmail.com> wrote:
are you sure MLR estimation method in MPLUS is the same as MML and FIML  in IRT? do you have any reference for this claim?

Yes, but they are deliberately vague about how they actually program (their software is not open source, and they don't want anyone else providing competition by implementing the same algorithms they design).  I've never seen them explicitly call it MML, but their description of it is exactly how MML works.


In Ch. 8 of their user guide, scroll down to the ESTIMATOR section to find a big table.  


Notice the asterisk by ML(R/F) under the column when at least 1 variable is categorical, indicating that numerical integration is required (i.e., how MML operates in practice).

Terrence D. Jorgensen
Assistant Professor, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lav...@googlegroups.com.

Shubhabrata Mukherjee

unread,
Jun 6, 2019, 4:43:56 PM6/6/19
to lavaan
Thank you for all your insights.

-Shubhabrata
Reply all
Reply to author
Forward
0 new messages