Pooled Correlation Matrix for SEM Met... PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Paul Silvia posted on Tuesday, February 17, 2009 - 12:18 am
Hi:

I've been reading Cheung's work on meta-analytic SEM, which pools correlation matrices via multiple-group analysis.

Will Mplus output such a pooled matrix, given summary correlation data?

(To tinker, I used summary data in correlation matrix form for 3 groups (same variables, no missing data) using NGROUPS and NOBSERVATIONS. The SAMPSTAT and TECH1-4 output reports separately for each group, but not for the combined sample.)

Thanks, Paul
 Linda K. Muthen posted on Tuesday, February 17, 2009 - 10:48 am
I can't think of any way to do this in Mplus. I'm assuming the pooled correlation matrix is the average of the three correlation matrices.
 Paul Silvia posted on Tuesday, February 17, 2009 - 3:43 pm
Yup, it's just the average of the matrices. It sounds like it's time for the old-school approach: pen, paper, and calculator. :-) Paul
 Mike Cheung posted on Saturday, March 28, 2009 - 8:48 am
Hi Paul,

Mplus can be used to pool correlation matrices. The following is the sample Mplus code:
TITLE: Pooling correlation matrices

DATA: FILE = test.cov;
NGROUPS=2;
TYPE IS COVARIANCE; ! Pretend correlation matrices as covariance matrices
! See Cheung and Chan (2005)
NOBSERVATIONS = 100 100;
VARIABLE: NAMES ARE x1-x3;
USEVAR ARE ALL;
MODEL:
latent1 BY x1*; ! Estimated standard deviations
latent2 BY x2*;
latent3 BY x3*;
latent1@1;
latent2@1;
latent3@1;
x1@0; ! No measurement errors
x2@0;
x3@0;
latent1 WITH latent2* (1);
latent1 WITH latent3* (2);
latent2 WITH latent3* (3);
MODEL g2:
latent1 BY x1*; ! Estimated standard deviations
latent2 BY x2*;
latent3 BY x3*;
latent1 WITH latent2* (1); ! Constrain correlation matrices
latent1 WITH latent3* (2);
latent2 WITH latent3* (3);
OUTPUT: SAMPSTAT;

Regards,
Mike
 fok hung kit posted on Wednesday, August 24, 2011 - 3:05 am
Hi....
I have also been reading Cheung's work on meta-analytic SEM.
Might I ask if I can use the raw data to "create" a pooled correlation matrices or covariance matrices?

And if I have obtained a pooled correlation matrices or covariance matrices in Mplus, how I can use it to do a CFA?

Thank you very much for your kind attention.

Kit
 Linda K. Muthen posted on Wednesday, August 24, 2011 - 12:53 pm
See the SAMPLE option in SAVEDATA. You can use this to create a pooled-within covariance matrix.

See Example 13.1 for how to use a covariance matrix as data. For a pooled-within covariance matrix the sample size is the number of observations minus the number of clusters.
 fok hung kit posted on Thursday, August 25, 2011 - 6:15 am
Thank you very much for your reply.
 Ke Anne Zhang posted on Sunday, February 19, 2012 - 10:03 pm
Dear Drs. Muthen & Muthen,

I am performing a meta-analytic SEM on a pooled correlation matrix with many missing cells. Is Mplus capable of taking in summary data (such as FULLCORR) with missing cells at all?

Here's my input code:
DATA: FILE = MASEM2.csv;
TYPE IS FULLCORR;
NOBSERVATIONS = 45925;
VARIABLE: NAMES ARE AGE1-AGE26;
USEVARIABLES ARE AGE1-AGE26;
ANALYSIS: TYPE = GENERAL;
ESTIMATOR=ML;

If Mplus is not designed to handle summary data with missing cells, what would you recommend I do to handle the missing cells? Impute them with another software? If so, is there a software that you would recommend?

Thank you very much!

Best regards,

Ke Anne Zhang
 Linda K. Muthen posted on Monday, February 20, 2012 - 1:56 pm
No, Mplus cannot handle summary data with missing cells. I don't know of a program that can.
 ehrbc1 posted on Wednesday, March 23, 2016 - 11:57 pm
Hello,

I have tested the same path analytic model across a number of studies, and now the next obvious step is to run a MASEM.

I’m interested in the significance of individual direct pathways, indirect effects and total effects. Across my studies, I have also tested the relative magnitude of these various paths (using the model constraint option in MPLUS).

So I would be interested in examining these relative magnitudes based on the pooled correlation matrix, but I understand that MPLUS can’t handle Cheung’s TSSEM and the WLS estimation method? Any suggestions as to how I can conduct this analysis in MPLUS?

Thank you.
 Bengt O. Muthen posted on Thursday, March 24, 2016 - 3:50 pm
You can do meta analysis as shown in our Topic 9 handout (last version) on slide 150 and on.
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: