Multiple groups PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 Petras Blakely posted on Thursday, March 20, 2008 - 11:47 pm
Hi,

I have a disaggregated SEM with covariates, some measured at the between level and some measured at the within level. I would like to see whether the structural paths are invariant across multiple groups. I am using FIML and I have categorical data.

When I use the type=mixture, knownclass option to specify that I have multiple groups I cannot test group invariance of paths originating from covariates at the between level.

Is there a statistical reason why these paths cannot vary over groups or is it something computational?
 Linda K. Muthen posted on Friday, March 21, 2008 - 12:28 pm
I would need more information to answer this. Please send your input, data, output, and license number to support@statmodel.com.
 Mihyun Park posted on Monday, May 09, 2011 - 1:19 pm
Dear. Linda

I would like to see whether the trajectories of stres are different across male and female. I tried to use multiple group analysis for the analysis. Following is the syntax. Is that right? And then, i have to do chi-square test. Is that right? But as you see, I found second following warning message. What's the problem?

Variable: names are sum1 sum2 sum3 sum4 axde4 par1 gra1 peer1 ....gra3 peer3 ape3 mat3 par4 gra4 peer4 ape4 mat4 sexw1 axde1 axde2 axde3;
Grouping is sexw1(1=male 2=female);
Missing are all(-99999);
AUXILIARY = (m);
Analysis: ESTIMATOR = ML;
Model:
f1 by par1 par2 par3 par4;
f2 by gra1 gra2 gra3 gra4;
f3 by peer1 peer2 peer3 peer4;
f4 by ape1 ape2 ape3 ape4;
f5 by mat1 mat2 mat3 mat4;
axde4 on f1 f2 f3 f4 f5;
[f1 f2 f3 f4 f5@0];
f1 f2 f3 f4 f5(1-5);
Model female: f1 by par2 par3 par4;
f2 by gra2 gra3 gra4;
f3 by peer2 peer3 peer4;
f4 by ape2 ape3 ape4;
f5 by mat2 mat3 mat4;
[f1 f2 f3 f4 f5];
[f3];
OUTPUT:sampstat stand mod(3.84);
TECH4;

Second,
*** WARNING in SAVEDATA command
DIFFTEST is available only for the estimators MLMV and WLSMV.
Request for DIFFTEST will be ignored.
 Linda K. Muthen posted on Monday, May 09, 2011 - 2:05 pm
You are using the ML estimator. You don't need the DIFFTEST option for ML. You should run the two models and look at the difference between the chi-square values and the degrees of freedom. See pages 434-435 of the Version 6 user's guide.
 Mihyun Park posted on Monday, May 09, 2011 - 2:34 pm
Thank you for your response.

You mean..what...
I should LGMs for each male and female? Or I should do using grouping option, multiple group analysis as original.
 Linda K. Muthen posted on Monday, May 09, 2011 - 4:22 pm
You should always analyze each group separately. If the same growth model does not fit in both groups, it does not make sense to compare across groups. Comparisons across groups is done with multiple group analysis.
 Luci M. posted on Tuesday, May 10, 2011 - 4:25 am
Does one look at each path for males and females in the unconstrained and constrained model? Or do the paths only get interpreted if the chi-square diff test is non-significant? This is a more general question as it applies to multiple group analysis? Thanks.
 Linda K. Muthen posted on Tuesday, May 10, 2011 - 12:02 pm
See the Topic 1 course handout and video under the Topic multiple group analysis. There are examples of doing difference testing across groups that should help you.
 Mihyun Park posted on Tuesday, May 10, 2011 - 3:26 pm
Thank you for your advice.
Gender can't be independent variable?
That is, following can't be impossible?

Model: i s on gender;
y on i s;

Only using male data, I should analyze LGM and then, I shoudl do LGM using female data seperately?
Model: y on i s;

Second question, if independent variable don't have in some study model, that is, the model only has dependent variables and continuous latent variables(I S), the model can apply LGM?
 Linda K. Muthen posted on Tuesday, May 10, 2011 - 8:33 pm
You should do a growth model for each group without covariates or anything else in the model. You must see if the growth model fits in each group.

A growth model does not need to have covariates. Please see Chapter 6 of the user's guide and the Topic 3 and Topic 4 course videos and handouts on the website.
 Mihyun Park posted on Wednesday, May 11, 2011 - 7:13 am
Thank you for your comments. Your advice is helpful for me.
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: