Constraining residual variance
Message/Author
 Liz Woodruff posted on Tuesday, April 05, 2011 - 11:38 am
Below is the syntax for a CFA I'm running. I'm trying to constrain residual variance for variables y2 and y10 to zero as they are negative but non-significant. Can you help me understand what syntax to use. I'm having a hard time understanding based on example 5.20 in the manual.

TITLE: Measurement Model CFA

DATA: FILE IS C:\DissData.dat;
VARIABLE: NAMES ARE y1-y20;
MODEL: F1 BY y1, y2, y20;
.
.
.
F7 BY LogY10, LogY11;

OUTPUT: STANDARDIZED;
 Linda K. Muthen posted on Tuesday, April 05, 2011 - 11:54 am
y2 WITH y10@0;

You should read Chapter 17 of the user's guide which describes all options that can be used in the MODEL command.
 Liz Woodruff posted on Friday, April 08, 2011 - 11:29 am
From reading the discussion board, it seems that if there are residual variances that are negative, then it is possible to constrain these variances to 0 (or is it to 1?) and re-run the CFA.

However, it also seems that this is not considered "best" practice and may be more appropriate to run an EFA. Can you help me understand how running the EFA will be helpful? What information will it provide?
 Liz Woodruff posted on Friday, April 08, 2011 - 11:39 am
Sorry to keep bothering, one more question (for now anyway)

Is the following syntax correct for constraining variables y2 and y10 to 0?

TITLE: Measurement Model CFA

DATA: FILE IS C:\DissData.dat;
VARIABLE: NAMES ARE y1-y20;
MODEL: F1 BY y1, y2, y20;
F2 BY LogY10, LogY11;
y2@0;
y10@0;
 Linda K. Muthen posted on Friday, April 08, 2011 - 4:04 pm
There is a thorough discussion if EFA in the Topic 1 course video that should answer your questions.

You have fixed the residual variances of y2 and y10 at zero.
 Liz Woodruff posted on Monday, April 18, 2011 - 6:31 pm
My original CFA was not converging due to negative residual variance. I conducted an EFA (in SPSS) and modified four of the indicators to fit my data more appropriately. I re-ran the CFA and 3 out of 4 indicators were no longer negative. I decided to remove the factor associated with that indicator that still has negative residual variance. Then ran the CFA and although there is NO negative residual variance, the model is still not converging because of too many iterations. Can you help me determine why? And what to do????

Thank you so much
 Linda K. Muthen posted on Tuesday, April 19, 2011 - 6:25 am
Try freeing all factor loadings and fixing the factor variances to one:

f BY y1-y4*;
f@1;

It may be that the first factor loading which is fixed at one as the default is estimated at a value very different from one.
 Liz Woodruff posted on Thursday, April 21, 2011 - 10:41 am
I tried that. I have 7 factors and fixed each of them to 1 and I freed all of the factor loadings. Now one of the indicators has a negative residual variance. Suggestions?
 Linda K. Muthen posted on Thursday, April 21, 2011 - 11:43 am
This points to the need for a change in your model. I would do an EFA in Mplus. There is a big difference between an EFA and a CFA. If the EFA is not clean, fixing all of the factor loadings to zero in a CFA may not be the best way to go. You can send the EFA output, CFA output, and your license number to support@statmodel.com if you want further guidance.
 Liz Woodruff posted on Friday, April 22, 2011 - 10:32 am
Great, thank you so much Dr. Muthen.