Multilevel Growth Mixture Modeling Sy... PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Janice Kooken posted on Saturday, October 24, 2015 - 7:40 am
I am analyzing multilevel longitudinal data with observations nested within students and students nested within teachers. I have structured the model so that class assignment is occurring at the student level. I am freeing most variances to vary by class, but I have not been successful in getting the teacher level intercept variance to vary by class. When I set up the syntax, I get the following error message: "Variances of between level variables are not allowed to vary across classes." Is there a work around for this? I am not using any covariates in the model. Thank you.
 Bengt O. Muthen posted on Saturday, October 24, 2015 - 8:07 am
You can capture the variance by a loading for a latent variable that you introduce. So with y referring to the latent intercept on level-2 say:

f by y* (lam);
f@1; y@0;

That lam parameter can then vary across classes, where the variance is lam*lam.
 Janice Kooken posted on Saturday, October 24, 2015 - 10:15 am
Thank you for your quick reply. I am missing something because when I tried to run the syntax you provided, I got the error message, "unknown variable: lam".
Here is my between cluster syntax for a 2 class model.
%between%
%overall%
Ib1 Sb1| y1@0 y2@1 y3@2 y4@3 y5@4;
sb1@0;
f by ib1*(lam);
f@1;
ib1@0;
%c#1%
lam;
%c#2%
lam;
 Bengt O. Muthen posted on Saturday, October 24, 2015 - 4:52 pm
Sorry, lam is a parameter label that I chose to communicate more easily to you. You should replace that in your class-specific statements with

f by ib1*;
 Janice Kooken posted on Sunday, October 25, 2015 - 6:07 am
Thank you again for your quick reply. I am getting closer. To make sure I understand, I will provide the syntax. It does not contain class specific parameters.
%between%
%overall%
Ib1 Sb1| y1@0 y2@1 y3@2 y4@3 y5@4;
sb1@0;

f by ib1*;
f@1;
ib1@0;

1. Is this correct?
2. Is the square of the factor loading the variance of ib1?
3. How would this change for a parallel process model with ib1, ib2, and ib3? Here are two possibilities:
f by ib1* ib2* ib3*;

or
f1 by ib1*;
f2 by ib2*;
f3 by ib3*;
4. Will this change the latent class structure?

Thank you.
 Bengt O. Muthen posted on Sunday, October 25, 2015 - 4:12 pm
Yes on 1 and 2.

3. The latter is correct.

4. Yes.
 Janice Kooken posted on Wednesday, November 11, 2015 - 7:08 pm
Thank you for your response. Could you please provide an interpretation of the between cluster variance when the above mentioned technique is not used? Is it the between cluster between class or is it the between cluster within class? I ran a set of models, and the between cluster variance is not statistically significant even though the ICC is close to 30% in the multilevel model without latent class or growth mixture. I found this surprising. Could you recommend any literature on the interpretation of the variance for multilevel growth mixture models? Thank you again.
 Bengt O. Muthen posted on Thursday, November 12, 2015 - 6:04 pm
Unless you can point to an example in the UG or in Henry-Muthen, we need to see the output to know exactly which model you refer to - please send to Support along with your license number.
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: