RESIDUAL COVARIANCE MATRIX NOT POSIT... PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Hao Duong posted on Thursday, May 22, 2008 - 7:09 pm
Dr. Muthen,
I compare three models: linear, quadratic, and piecewise model. When I run a quadratic model, the result appears as follows:
THE MODEL ESTIMATION TERMINATED NORMALLY
WARNING: THE RESIDUAL COVARIANCE MATRIX (THETA) IS NOT POSITIVE DEFINITE. THIS COULD INDICATE A NEGATIVE VARIANCE/RESIDUAL VARIANCE FOR AN OBSERVED VARIABLE, A CORRELATION GREATER OR EQUAL TO ONE BETWEEN TWO OBSERVED VARIABLES, OR A LINEAR DEPENDENCY AMONG MORE THAN TWO OBSERVED VARIABLES. CHECK THE RESULTS SECTION FOR MORE INFORMATION. PROBLEM INVOLVING VARIABLE BMI12.

Does this warning influence the result accuracy? Should I say this model is not appropriate or I can use its BIC to compare with BIC in other models?
Thank you
Hao
 Bengt O. Muthen posted on Thursday, May 22, 2008 - 7:27 pm
Consider the text of the printed WARNING, for instance look at the Theta estimate for BMI12 - is it negative? If so, you may want to modify the model.
 cchien posted on Monday, April 26, 2010 - 8:04 am
I got the same warning messages when running an EFA with complex survey design. I am just wondering what else I can do if the variable showing in this warning message should be included in my analysis.
Thank you.
 Linda K. Muthen posted on Monday, April 26, 2010 - 10:29 am
Please send the full output and your license number to support@statmodel.com so I can see the exact situation.
 Joshua Wiley posted on Monday, August 06, 2012 - 11:20 am
When the theta matris is not positive definite due to a residual correlation greater r where |r| >= 1, are there any general suggestions for what to look at? I am hoping for pointers and likely culprits to change or check to address the residual correlation issue.

I can certainly send output and license info for this particular model, but I have run into the issue before and searched the discussion boards looking for general suggestions without much luck. I thought a general question might be useful to me and other users.
 Linda K. Muthen posted on Monday, August 06, 2012 - 1:43 pm
This message is usually due to some model misspecification. With a residual correlation greater than one, it may point to the need for more factors.
 Joshua Wiley posted on Monday, August 06, 2012 - 8:05 pm
Thanks Linda!
 Danyel A.Vargas posted on Saturday, September 14, 2013 - 3:29 pm
Dr. Muthen,

I am receiving this error:
WARNING: THE RESIDUAL COVARIANCE MATRIX (THETA) IS NOT POSITIVE DEFINITE.
THIS COULD INDICATE A NEGATIVE VARIANCE/RESIDUAL VARIANCE FOR AN OBSERVED
VARIABLE, A CORRELATION GREATER OR EQUAL TO ONE BETWEEN TWO OBSERVED
VARIABLES, OR A LINEAR DEPENDENCY AMONG MORE THAN TWO OBSERVED VARIABLES.
CHECK THE RESULTS SECTION FOR MORE INFORMATION.
PROBLEM INVOLVING VARIABLE D1HILED.

This variable is being used as an auxiliary variable. I cannot see where the problem is in my output.

Can you please help me?

Thanks,

Danyel
 Linda K. Muthen posted on Saturday, September 14, 2013 - 5:26 pm
Please send the output and your license number to support@statmodel.com.
 Eiko Fried posted on Tuesday, June 24, 2014 - 2:21 am
Dear Dr Muthens,

We are running a comparison of two groups regarding their correlation structure, using the WLSMV estimator.

> grouping is cond (1=control 2=widow)
> model:
> s1 with s2-s11;
> s2 with s3-s11;
etc.

We receive the "RESIDUAL COVARIANCE MATRIX (THETA) IN GROUP x IS NOT POSITIVE DEFINITE" warning for two variables. In tech1, there seems nothing wrong with the theta values of those variables.

What could be the problem? Can the error be ignored?
 Linda K. Muthen posted on Tuesday, June 24, 2014 - 5:45 am
Please send the output and your license number to support@statmodel.com.
 Jahun Kim posted on Monday, December 14, 2015 - 2:06 pm
Dear Dr. Muthens,
I got the same warning messages, when running an LCA with (6- year) longitudinal repeated design. The error was involved with a variable (gsic) should be included in my analysis. I ran gsic@6 then replaced with gsic*. Please see written syntax below.

PLOT:
SERIES= asic(0), bsic(0.5), csic(1), dsic(1.5), esic(2),fsic(3), gsic(6);
Type=plot3;
MODEL:
i s q|asic@0, bsic@0.5, csic@1, dsic@1.5, esic@2, fsic@3, gsic@6;

I'm wondering what else I can do to fix the error and get a better model fit.

Thank you!
 Linda K. Muthen posted on Tuesday, December 15, 2015 - 5:55 am
Please send the output and your license number to support@statmodel.com.
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: