Wim Beyers posted on Tuesday, February 01, 2005 - 12:43 am
Is it possible to estimate a Growth Mixture Model based on multivariate data? I mean, a classical cluster analysis often is based on multiple (relatively) independent variables (e.g., in parenting research: warmth and control). So, is this also possible in a LGMM (e.g., finding latent classes of trajectories in warmth ànd control)? I do not mean parallel LGMM, but rather 'combined' LGMM.
Wim Beyers posted on Thursday, February 03, 2005 - 2:55 am
OK, but do examples exist of such an analysis? To make myself more clear: I want to estimate latent classes that combine measures of at least two relatively independent variables (X1 and X2), measured repeatedly. So that, for instance, Class 1 consists of persons with 'stable high levels of X1' (high intercepts and low slopes) COMBINED with 'increasing X2 (low intercepts and high slopes).
bmuthen posted on Thursday, February 03, 2005 - 10:15 am
I don't know that we have exactly such examples, but the idea and the modeling is clear. You would use 2 sets of growth factors (I assume you don't think they are the same for the 2 outcomes) but one latent class variable, where the latent classes are determined by the joint growth trajectories. You can use ex 6.13, deleting the last 2 lines of the model input, adding type = mixture and letting the defaults give you different growth factor means.
As a follow up to this original post, I was wondering if it is possible to conduct an LGMM analysis with multiple outcomes (anxiety, depression, self-esteem). In other words, is there a way to model three outcomes at the same time to test whether there are different classes of individuals who might show differing patterns across these variables?
You can start from UG ex 7.14 which shows 2 sets of outcomes, each with its own latent class variable. Just turn that into a growth situation. Then generalize to 3 sets. I can't think of a paper that describes this in detail but there is nothing difficult about it.