Should variables be correlated when d... PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 SCOOPER posted on Thursday, October 09, 2008 - 10:55 pm
I'm embarking on my first attempt at doing an LCA. Specifically, I'm doing a LCA predicting juvenile justice system contact among girls. There are about 14 variables going into the profiles. I've done some preliminary analyses and I get the following for a couple of my class solutions:

"WARNING in Model command
Variable is uncorrelated with all other variables within class: MNNEIGH
*** WARNING in Model command
Variable is uncorrelated with all other variables within class: ANXIETYM
*** WARNING in Model command
Variable is uncorrelated with all other variables within class: DEPRESSM
*** WARNING in Model command
Variable is uncorrelated with all other variables within class: RELSERV
*** WARNING in Model command
Variable is uncorrelated with all other variables within class: SEXEVER
*** WARNING in Model command
Variable is uncorrelated with all other variables within class: SUBCGSMK
*** WARNING in Model command
Variable is uncorrelated with all other variables within class: SUBHRDLQ
*** WARNING in Model command
Variable is uncorrelated with all other variables within class: SUBMJ
"

Should my variables be correlated when conducting a LCA? This is happening for many of the variables. Thanks in advance for feedback regarding this question.
 Linda K. Muthen posted on Friday, October 10, 2008 - 6:05 am
This is a warning in case you have variables in the analysis that should not be. If it is as intended, you can ignore it.
 Seana Golder posted on Wednesday, November 05, 2008 - 8:27 am
I am running a LCA with a mixture of categorical and continuous latent class indicators. Conceptually, there is reason to have the continuous latent class indicators correlate within class. When I run the 2 class model, the model estimation terminates normally (I have not used starting values). However, when I move to run the 3 class model (without starting values) , I am getting an error beginning with: THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO AN ILL-CONDITIONED FISHER INFORMATION MATRIX. CHANGE YOUR MODEL AND/OR STARTING VALUES.
In response to this error message I used starting values as shown in example 7.22 of the Mplus Userís Guide (the 3 class model then terminated normally).

I have several related questions: Am I correct in my assumption that I needed to specify starting values for the model to run? Why are starting values necessary? Finally, can you provide any citation or other guidance as to how to select appropriate starting values for continuous latent class indicators?

Thanks in advance.
 Linda K. Muthen posted on Thursday, November 06, 2008 - 8:45 am
You should not need starting values. Please send the two- and three-class outputs without starting values 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: