Anonymous posted on Sunday, March 06, 2005 - 10:04 pm
I recently finished the Li, Duncan, Duncan, and Acock (2001)article in Structural Equation Modeling. After reading the article I looked at their website, www.ori.org/methodology/sempaper.home.htm. On the website, they present some model Mplus programming. I understand the programming is for version 1.0, I am left to wonder is it still necessary (i.e., using Mplus 3.11) when building the models to include the means of the intercepts and slopes? For instance, after running a model with 1 class, they used the mean for the intercept and slope in the next run for class 1 in a 2 class model. Is this necessary?
Also, I noticed in the Li et al. article and the Nagin (1999) article that part of the process is to make comparisons of the latent class variable on covariates. Both seem to use logit. I am familiar with the logit function in Mplus, but not in this context. How would I do this so that I maintained my groups from the Latent class analysis? Would it be necessary to add to the LCGA program the logistic regression syntax were the latent class variable is the dependent variable and then use the covariates as predictors? Will I still be able to keep the groups separate?
Any help with these issues will be outstanding. Thanks in advance.
It is no longer necessary to give starting values for mixture models. As the default, Mplus provides automatic starting values and random starts.
The regression of the categorical latent variable on a set of covariates is a multinomial logisitc regression. It is specified as follows for two class model where the last class is the reference class:
c#1 ON x1 x2;
See the following paper for a discussion of covariates in mixture models:
Muthén, B. (2004). Latent variable analysis: Growth mixture modeling and related techniques for longitudinal data. In D. Kaplan (ed.), Handbook of quantitative methodology for the social sciences (pp. 345-368). Newbury Park, CA: Sage Publications.
Anonymous posted on Monday, March 07, 2005 - 10:38 am
One more point of clarification on the multinomial logistic regression. Is this performed in conjunction with the LCGA analysis? Or is this a separate run on data that have been save using the savedata command?