milan lee posted on Wednesday, January 15, 2014 - 11:23 pm
Hi Dr.Muthens, I tried THREE types of 4-class GMM: a unconditional one, one with covariates on linear and quadratic factors, one with covariates on the latent class variable and both growth factors. When I looked at the outputs and estimated-mean plots, there is one class in which both linear and quadratic factors were non-sig in unconditional gmm but significant in gmm with covariates and latent class variable. Also, the plots look different in these three types of models. 1) Is this change in significance of growth factors across models normal? If so, why? 2) How much of the changes in estimated means across models is acceptable? Thanks a lot!
1) By significance of growth factors perhaps you refer to their variances. If so, this phenomenon often happens also in non-mixture growth models, perhaps due to having more power with covariates included.
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. download paper contact author show abstract
which is on our website.
milan lee posted on Thursday, January 16, 2014 - 12:08 am
Hi Dr.Muthen, By significance of growth factors, I meant the regression estimates for linear slope mean and quadratic mean. So they turned out to be significant in the GMM with covariates but non-sig in the unconditional one. May I have your advice on the first question again? Thanks a lot!
When you have covariates, the parameter you are looking at is the intercept for the growth factor (so e.g. an intercept for the intercept growth factor), not its mean. With covariates you find the mean in TECH4.