Message/Author |
|
James Algina posted on Wednesday, November 12, 2014 - 8:50 am
|
|
|
Hello: In Appendix K In Appendices for Auxiliary Variables in Mixture Modeling: 3-Step Approaches Using Mplus you do not include c2 on c1 in %OVERALL%. Is this an essential feature of the procedure or can c2 on c1 be included in %OVERALL%? The inclusion/exclusion of c2 on c1 affects the estimates that will be used in Step 2. Thanks, Jamie |
|
|
Yes it is essential. Do not include it. The idea here is that the latent class variable should be determined only by its indicator variables at the corresponding time point. If you include c2 on c1 that will no longer be the case. If you don't care about the invariance of the measurement models at point 1 and 2 you can run the two Step 1 models for c1 and c2 separately - they are put together in Appendix K only to enforce measurement invariance. See page 14 and 15 https://www.statmodel.com/download/webnotes/webnote15.pdf |
|
James Algina posted on Wednesday, November 19, 2014 - 9:57 am
|
|
|
Hello: Thank you for your response. In Step 1 should replication of the log likelihood be monitored by using a starts = # # command? And if it should, should the optimum seed be used in subsequent steps? Thanks, Jamie |
|
|
In Step 1 you should use sufficient number of starts because in that step you are searching for the best latent classification. You should not use the optimum seed for later stages (the parameters are essentially fixed for the later stages). So for all later steps I would recommend using starts=0 and not using any optseed. |
|
Back to top |