Message/Author |
|
|
I have used a growth mixture model on my dataset and identified a model with 4 classes which fits the data well and also runs in line with my hypothesis. However, if I now want to use those classes on a new set of patients (for whom I only have a few datapoints), how would this be done? I would like to for example be able to identify at cut-off point for the variable of interest at which I could allocate patients to the groups. On a related note, how can I best identify which patients were in which groups in the analysis? |
|
|
For your first question you would fix all parameters of the model to the estimated parameter values from the full run when running it with the new set of patients, using missing data indicators for the data points you don't have. This gives the posterior probabilities of class membership for these new patients which can be used for classification. To identify patients' group membership you use Save=cprobabilities (see UG). That file will give you most likely class membership. |
|
|
Seeking some clarification of the answer above as I am a newer user of Mplus and LCA. I have run an LCA and identified a well fitting model with 3 classes from one set of data and would also like to use these classes in a different set of patients. Are the parameter values you are referring to, the threshold estimates from the model results? |
|
|
It refers to all free model parameters. You can get input with parameter estimates as starting values using the SVALUES option of the OUTPUT command. Change the * to @ to fix the parameters to these values. Use this MODEL command and STARTS=0; with the new data set to generate posterior probabilities for the new data set. |
|
|
As discussed above, when applying the free model parameters identified in one dataset to another, are sample size requirements similar or do they not apply if they were initially established with a sufficiently large group e.g. n=500? Is anything known about SS requirements in this scenario? many thanks |
|
|
There are no requirements for the sample size in the second data set. In fact, you can have n=1. |
|
Back to top |