Higher-Order Factors

2 replies [Last post]
blitz's picture
Offline
Joined: 02/13/2013

We have four time-points, and at each time-point there are two indicators. Is it possible to use the two manifest indicators per time-point to create one latent variable per time-point, and run a growth mixture model on those four latent variables?

If that is possible, suppose that we are actually missing one manifest indicator at one time-point. Is it possible to create a latent variable in place of that manifest indicator? A drawing of our model, as well as our attempted script, is attached. We are getting the following error:

Error: The job for model 'Growth Mixture Model' exited abnormally with the error message: Objective function returned an infinite value at iteration 0.1.

AttachmentSize
Forum Model.pdf66.81 KB
R Syntax.R14.22 KB
Ryne's picture
Offline
Joined: 07/31/2009
Sorry for the delay in

Sorry for the delay in answering you.

I see no reason OpenMx can't estimate the model you've drawn. I'd add that the sets of the PTSD variables and the growth factors both need variance terms, and you haven't put variances or covariances on the factors in the script you've attached. Also remember that the classes should have different starting values for at least one parameter, if possible. One of the bigger difficulties in second order factor models is identification, which you've taken care of by fixing the PTSDx to CAPSx loadings to 1.0.

I can't replicate the error without data of some kind; I actually receive a different one, as you've renamed the default growth mixture dataset with one of different size.
Running Growth Mixture Model
Error: The MxMatrix object 'Class1.F' has specified dimnames with dimensions 0 x 13 but the matrix is of dimensions 7 x 15

Could you change your model to add the latent variances/covariances? If you still get this error, attach some data that yields the same error message and I'll give it a look.

Ryne

blitz's picture
Offline
Joined: 02/13/2013
Thanks for your help!

Don't apologize--thanks for responding at all!

The growth factors did not have variance terms because we wanted to try a latent class growth analysis, not growth mixture model. I shouldn't have written "growth mixture model" in my original question. Sorry!

We managed to get the model to work in Mplus, and used the Mplus results as starting values in R. The syntax is attached. It worked!

AttachmentSize
Forum Response.R 8.96 KB