Three-level CFA? PreviousNext
Mplus Discussion > Confirmatory Factor Analysis >
Message/Author
 Sylvia Morelli posted on Monday, September 09, 2013 - 1:49 pm
Is it possible to do a CFA with a three-level model? If yes, is there an example dataset and MPlus code anywhere?
 Bengt O. Muthen posted on Monday, September 09, 2013 - 3:24 pm
See UG ex 9.22 - just drop the model parts for the x, w, z covariates.
 Sylvia Morelli posted on Monday, September 16, 2013 - 3:49 pm
Is the three-level CFA only available in MPlus 7? I am using MPlus6 and am getting this error:

"*** ERROR in ANALYSIS command
Unrecognized setting for TYPE option:
THREELEVEL "
 Linda K. Muthen posted on Monday, September 16, 2013 - 4:08 pm
It was introduced in Version 7.
 Md Zabir Hasan posted on Sunday, April 29, 2018 - 9:02 am
Is it possible to perform a Three level CFA with categorical and ordinal indicator variables in MPLUS v.8.

I have six binary and six orderdinal variables with three categories. My data is nested within three levels (Individual -> Household -> Community).

Thanks in advance.
 Bengt O. Muthen posted on Sunday, April 29, 2018 - 2:02 pm
Yes, see our Short Course Topic 10 videos and handouts on our website.
 Md Zabir Hasan posted on Monday, April 30, 2018 - 11:29 am
Thank you so much.

After reviewing the short course, I am using the three level CFA with BAYES estimator. The model is taking long time to converge (3 hours and still running. Is there any solution?

Analysis:
TYPE = THREELEVEL;
ESTIMATOR = BAYES;
PROCESSORS = 2;
BITERATIONS = (1000);

MODEL:
%WITHIN%
fw1 BY GMB CA CAD;
fw2 BY SSE SSF SSI;
fw3 BY TRUSTL TRUSTS TRUSTN;
fw4 BY SCR SCSB SCF;

%BETWEEN hhid%
fb1 BY GMB CA CAD SSE SSF SSI;
fb2 BY TRUSTL TRUSTS TRUSTN SCR SCSB SCF;

%BETWEEN psu%
fc1 BY GMB CA CAD SSE SSF SSI;
fc2 BY TRUSTL TRUSTS TRUSTN SCR SCSB SCF;
 Bengt O. Muthen posted on Monday, April 30, 2018 - 11:47 am
Perhaps your sample size is very large or the third level has very small variances. Also, use Tech8 to study how the PSR values progress, seeing how long each iteration takes, and seeing if a particular parameter is often the one with the worst PSR (use Tech1 as well for this purpose).
 Md Zabir Hasan posted on Monday, April 30, 2018 - 12:23 pm
Thanks,

I have 7186 Individuals in 5186 Households which are nested in 346 villages. On an average I have 1.385 observation from each household and observation from household ranges from 1 to 6.

26% households have one observation, 53% have two and 16% have three (which is cumulatively 95% of the sample). I have calculated the household level ICC which are quite high across the 12 items. But my dilemma is there are mostly 1-3 observation from each household.

What will be the implication of considering the two level model with individual and the village and not accounting for the household?

Thanks in advance.
 Bengt O. Muthen posted on Monday, April 30, 2018 - 6:03 pm
I think you should try to do the 3-level modeling assuming you have perhaps only a random intercept parameter specific to the household level.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: