|
|
 |
| Growth modeling with many time points |
 
|
| Message/Author |
|
|
| Yisheng Li posted on Friday, September 26, 2008 - 10:23 am
|
|
|
Dear Linda and Bengt,
In a study, I'm interested in the association between certain latent covariates and a binary outcome variable. The latent covariates, in my case, are the random intercepts and slopes from a multi-level model. I believe Mplus can deal with such a model. In my understanding, Mplus needs to specifiy each measurement in the longitudinal process in the graph. Would it be a problem if I have many time points, say, more than 100 for each study subject? Can Mplus still deal with such a model?
A further question: When Mplus deals with such a joint model (e.g., if ML option is used for predicting the binary outcome using the random intercept/slope factors), does it estimate both factors first and use those estimated factors to predict the binary outcome, or does it truly do a joint modeling? If it's the latter, could you please provide a reference for the method?
Many thanks! |
|
|
| Yisheng Li posted on Friday, September 26, 2008 - 12:22 pm
|
|
|
Sorry, guess it's a quite dumb question. The model specification should be just the same.
Still I'm interested in knowing what method Mplus is based to do the joint modeling, if it is doing so, and the references.
Thanks!
Yisheng |
|
|
|
|
With many time points you can switch from the wide, multivariate approach to the long, univariate approach of growth modeling. The latter is twolevel modeling with time as level 1 and person as level 2. So you predict the binary outcome on level 2 as as function of the random effects.
The estimation is via ML and done jointly, not in two steps. References are Muthen et al (2002) in Biostatistics, Muthen (2004) in the Kaplan handbook, and Muthen & Asparouhov (2008) in the Longitudinal Analysis handbook. You find all of them on our web site. |
|
|
| Back to top |
|
|
