Sam Craft posted on Wednesday, September 18, 2019 - 9:58 am
I'm attempting to conduct either an LCGA or GMM of a categorical outcome measured over 3 time points. I would like to include 2 time varying covariates, am I right in thinking that at this point it becomes a GMM (due to LCGA not allowing within class variation of covariances)?
Also, Muthen 2004 & 2006 suggest to firstly use an unconditional LCGA to estimate the number of classes, then add any class-specific variances (if necessary) & covariates into a GMM model. Is this because of the computational demands of running several iterative GMM models? Or are there are other reasons why it would be inappropriate to only use GMM models with within class variances and TVCs from the outset?
Q1: You can have covariates without having within-class variances - the equation the covariates is part of simply won't have a residual variance.
Q2-Q3: That but also the fact that there are so many possible direct effects from covariates to latent class indicators that the correct model may be difficult to find. See also
Nylund-Gibson, K. & Masyn, K. (2016). Covariates and mixture modeling: Results of a simulation study exploring the impact of misspecified effects on class enumeration. Structural Equation Modeling: A Multidisciplinary Journal, DOI: 10.1080/10705511.2016.1221313
Sam Craft posted on Thursday, September 19, 2019 - 10:11 am
But do the TVCs not require ALGORITHM=INTEGRATION due to the random slope on the outcome?