Correlated Errors in CFA using ML PreviousNext
Mplus Discussion > Confirmatory Factor Analysis >
Message/Author
 Dallas posted on Wednesday, June 29, 2011 - 7:05 am
I have a model with 10 categorical (binary) items. I'd like to run the model allowing local dependence between two of the items. I'm using ML estimation. So, I get the following message "Variances for categorical outcomes can only be specified using PARAMETERIZATION=THETA with estimators WLS, WLSM, or WLSMV."

How can I go about specifying correlated errors in Mplus using ML estimation?

Thanks!
 Linda K. Muthen posted on Wednesday, June 29, 2011 - 7:30 am
The Theta parameterization is for weighted least squares analysis. With maximum likelihood each residual covariance requires one dimension of integration. To specify a residual covariance in this case, say

f BY u1@1 u2;
f@1; [f@0];

The residual covariance is found in the factor loading for u2.
 Dallas posted on Wednesday, June 29, 2011 - 8:12 am
Thanks! I see. In this case, then, wouldn't one want to constrain the covariance between the general factor and the factor representing the local dependence to zero?

Say:
f1 by u1-u10;
f2 by u1@1 u2;
f2@1; [f2@0]; f1 with f2@0;

To make the model equivalent to a WLSMV specified as follows:
f1 by u1-u10;
u1 with u2;
 Linda K. Muthen posted on Wednesday, June 29, 2011 - 8:20 am
If the factor covariance is not zero as the default, then it should be fixed at zero.
 Dallas posted on Wednesday, June 29, 2011 - 8:21 am
Thanks. When I estimate the model, it does not appear fixed to zero as default. So I asked to make sure my understanding that it should be set to zero made sense. Thanks again.
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: