Multiple group analysis in complex su... PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Shang-Min Liu posted on Tuesday, April 08, 2008 - 2:40 pm
Dear Dr. Muthen,

I did MIMIC with 7 binary indicators and 2 binary covariates in complex survey with n=1,422. I got significancy result between these two covariates and 2 out of 3 factors, so I want to do Multiple groups (4 groups) analysis to know where the differency from. Since there is no direct effect in my MIMIC model, does it make sence to do Multiple analysis? (because the example in the handout is with direct effect) The other question is how to use "variable" commend for 4 groups? Thank you!
 Linda K. Muthen posted on Tuesday, April 08, 2008 - 3:02 pm
The MIMIC model allows you to test measurement invariance of the intercept parameters. A multiple group analysis allows you to test measurement invariance of other measurement parameters, for example, factor loadings and residual variances.

The VARIABLE command works the same in multiple group analysis as in single group analysis. It names the variables in the data set.
 Shang-Min Liu posted on Wednesday, April 09, 2008 - 8:39 am
Thanks for your reply! About "model" command, could we use overall model for all groups or we have to specify differences between the overall analysis model in each group(in step 2: fit the model in all groups)? How could we know the specific model would be in each group? Also in the step 2, how could we write "grouping" statement under "variable" commend for four groups? The other question is about the numbers of people in these four groups. They are from size 230-589. Is it large enough? Really appreciate for your help!
 Linda K. Muthen posted on Wednesday, April 09, 2008 - 10:59 am
Please see Chapter 13 where the use of the overall and group-specific MODEL commands is described as are other issues with multiple group analysis. The GROUPING option works the same no matter how many groups there are. Please see the GROUPING option in the user's guide. The only way to know if a sample is large enough is to do a Monte Carlo study.
 Shang-Min Liu posted on Wednesday, April 09, 2008 - 2:24 pm
Dear Dr. Muthen,

Thanks. After reading Chapter 13 in user guild, I still not clear about how to choose which indicator not to be constrained in each specific model because I don't have any direct effect between indicators and covariates in my MIMIC model. Would you please let me know how to deal with this? The other question is about the sample size. Since Monte Carlo can't work on design variables (psu, stratum, etc.) in Mplus, is there any other way to figure out whether the sample size in each group is large enough in Mplus? Thank you!
 Linda K. Muthen posted on Thursday, April 10, 2008 - 9:22 am
I would do a Monte Carlo study taking clustering into account. That would be the best you can do.

Please send your input, data, output, and license number to and be specific about what you want to do.
 James McMahon posted on Saturday, September 12, 2009 - 11:57 am
I am new to MPlus and have a basic question on subgroup analysis. I simply want to run analyses on subsets of data in a single dataset based on some variable attribute. For example, suppose I have a dataset that includes both males and females (in a GENDER variable) and I wanted to run the same analysis on males and females separately. In SAS I would simply include


or use the BY command.

How is this done in MPlus? Thank you.
 Bengt O. Muthen posted on Saturday, September 12, 2009 - 12:21 pm
You would use the USEOBSERVATION option in the VARIABLE command to select the subgroup to be studied, and do a run for each gender.
 Daniel Schwarzkopf posted on Thursday, February 04, 2010 - 1:31 am
Dear Dr. Muthen,
I have Multilevel-Data and want to do multigroup analysis. I want to do it with a grouping variable on the within level. In my multilevel model I have random slopes which are predicted by between level variables. Using within level variable for grouping brings me an error message.
Is it only possible to do the grouping on the between level? Or is there a way to do it with a variable from the within level also?
 Linda K. Muthen posted on Thursday, February 04, 2010 - 7:41 am
In multiple group analysis, the groups should contain independent observations. Using a within variable as a grouping variable violates this independence.
 Rachel posted on Monday, March 09, 2015 - 4:22 pm
Hello Dr. Muthen,

I am doing multi-group SEM with complex data but I have unequal sample sizes. I was going to use bootstrapping to deal with this however I understand with TYPE=COMPLEX Mplus does not allow this. are there any other commands I can use to deal with unequal sample sizes while doing this type of analysis?

Thank you for your time.
 Bengt O. Muthen posted on Monday, March 09, 2015 - 6:23 pm
Multiple-group analysis is not hurt by unequal sample sizes.
 Rachel posted on Tuesday, March 10, 2015 - 4:48 pm
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message