Message/Author |
|
|
Hi, I am estimating a latent growth model (three continuous variables), and have multiple time-invariant covariates predicting latent intercept and slope. How can I tell if adding the covariates significantly improved model fit, or if adding certain covariates to the slope, but not intercept is a better model. Thank you. |
|
|
You can see how BIC changes. But I would also simply go with the covariates that have significant effects. |
|
|
Thank you very much for your reply. Just one follow up - is in inappropriate to use chi-square difference testing in this case (i.e., does this not qualify as nested model comparison)? Thanks again. |
|
|
You can use the chi-square difference test. |
|
lotti posted on Wednesday, November 12, 2014 - 6:40 am
|
|
|
Dear Professor Muthen, I have a dual-process latent growth modeling. In univariate modeling, I found: (1) a significant linear increase in the first process over time (i.e., positive slope), and a linear decrease in the second process over time (i.e., negative slope). In the multivariate model, I specified a direct path from the slope and the intercept of the first (increasing) process to the slope of the second process. To my surprise, under this model the slope of the second process became positive. What does this mean? I'm very grateful for your help in understanding this. |
|
|
Are you sure you are looking at the mean of the slope or only its intercept? When the slope is a DV it is the intercept in that regression that is printed and you find the mean in TECH4. |
|
Back to top |