Message/Author |
|
|
Hi there, I am currently plotting trajectories in kidney function (continuous variable) over time. There are four timepoints available, each of them are five years apart, but due to the wide age range of the study population (26-65 years at baseline) I structured the dataset according to age. In short, this means that instead of 4 kidney function measurements, I now have a kidney function variable for every year of age (i.e. kidney26 – kidney80). A lot of missings are however being created. This is because every participant has a maximum of only 4 out of 55 measurements available. I am in doubt whether MPlus can handle this; the analysis (1000 starting values) takes more than 7 hours and there are more than 600 patterns of missings. Also, the first 30 loglikelihood values are not equal, even though they should. Is there any way to come around this problem? Let me know if you're interested in seeing the editor. Thanks heaps for your reply. Gerrie-Cor |
|
|
A better approach in this "multiple-cohort" case is to use multiple-group modeling. See UG ex6.18. |
|
|
Thanks for your reply. I have read the section about multiple-group modeling, but it seems that it is able to estimate the general (overall) trend in outcome. I am actually interested in longitudinal latent classes of kidney function. Do you know if this method is capable of doing that? Thanks, Gerrie-Cor |
|
|
This can be extended to mixture modeling. |
|
|
OK, that's good news. Where can I find the syntax for this extension? |
|
|
I don't think we have specific examples for this case, but as usual one can combine pieces from different UG examples. In the case of mixtures, you would represent the groups by latent classes designated as Knownclass so start from UG ex8.8. This example shows how to combine the known and unknown latent class variables. |
|
Back to top |