IRT with Theta/Delta parameterization PreviousNext
Mplus Discussion > Categorical Data Modeling >
Message/Author
 Laura Stapleton posted on Wednesday, October 15, 2008 - 11:23 am
Folks-

When modeling with variables identified as categorical, I understand that the choice of scaling of the latent item response variances (to 1.0) or the item residual variances (to 1.0) is somewhat arbitrary.

With WLSMV estimation, it appears that the delta parameterization is used as a default (latent item response variances are set to 1.0).

With ML estimation (with numerical integration), when requesting LINK=PROBIT, it appears that the theta parameterization is used (item residual variances are set to 1.0).

Is this correct?
Is there any particular reason why this is the case?
Thank you!
 Linda K. Muthen posted on Wednesday, October 15, 2008 - 3:06 pm
In maximum likelihood, scale factors are not used and residual variances are fixed to one for probit and pi squared divided by three for logit. The former is not the Theta parameterization because in multiple group or growth models, the residual variances cannot be estimated. They are fixed in all groups and at all timepoints.
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: