Message/Author |
|
|
Hi, I am conducting GMM with 2 parallel growth processes, and I have 4 time-invariant covariates; 2 of them are temperament variables and the rest of 2 are parenting variables. I first started with unconditional models without any covariates and reached at stable solutions. Then, I sequentially added covariates, first 2 temperaments and then added 2 more parenting variables. 1) However, the slopes of some classes have changed when I added 2 parenting covariates along with 2 temperament covariates. For example, with just 2 temperament covariates, one of the class had a significant negative slope but with 2 more parenting covariates, the slope of the same class turned out to be significantly positive. How could this happen and how should I explain this kind of results? 2) Is it possible to examine hierachical effects of covarites in GMM? In polynomial logistic regression part of GMM, I want to examine additive effects of parenting beyond temperaments. How can I incoporate this in my input syntax? Thank you for help in advance! |
|
|
1. Are you regressing the classes on the covariates or the growth factors on the covariates. 2. Mplus does not do stepwise regression. |
|
|
I am both regressing the classes on the covariates and growth factors on the covariates. My input looks someting like this; Model: %overall% i_dep s_dep | dep12@0 dep12h@1 dep13@2 dep13h@3 dep14@4; i_delin s_delin | delin12@0 delin12h@1 delin13@2 delin13h@3 delin14@4; i_dep WITH s_delin; i_delin WITH s_dep; i_dep WITH s_dep; i_delin WITH s_delin; i_dep WITH i_delin; s_dep WITH s_delin; s_delin@0; s_dep@0; i_dep s_dep ON NS HA; i_delin s_delin ON NS HA; c ON NS HA; Can you provide any tips on this? Thank you again. |
|
|
You should not include WITH statements for variables that have the residual variance fixed at zero. |
|
|
Dear Linda, Thank you for the useful tip. As you have told, I excluded WITH statements for variables that have the residual variance fixed at zero from my input. However, the problem I mentioned at the very top, changing slopes within the class from positive to negative and vice versa depeding on the covariates included in the model, still remains. Do you have any suggestions for this? Thanks again! |
|
|
How the slopes change depends on the relationships among the covariates and the relationship of the covariates to the dependent variable. The coefficients when there is more than one covariate are partial regression coefficients. |
|
Back to top |