|
|
Inconsistent BLRT across comparable m... |
|
Message/Author |
|
Jon Heron posted on Thursday, August 14, 2008 - 11:47 pm
|
|
|
Hi, I have 3 continuous measures and I am throwing a bunch of different mixtures models at them. I am fitting two types of LCGA, one with a quadratic describing changes through time and the other a piecewise linear change. For mixture models with the same number of classes these two models achieve exactly the same fit figures - likelihood, AIC, BIC, aBIC, Entropy (not that surprising) However, when it comes to BLRT I am finding differences - the BLRT supports the 3 class model for the quadratic LCGA however the conclusions are different for the piecewise model (in fact the BLRT changes are non-monotonic for this model, favouring a 2 or a 6 class). To make sure that these are not just fluke results I have increased the starts considerably: k-1starts = 1000 100; lrtstarts 0 0 1000 100; lrtbootstrap 1000; however the results seem robust. Perhaps this is nothing to be surprised about and there is in fact nothing wrong here at all? many thanks Jon |
|
|
Please send your files and license number to support@statmodel.com. |
|
Jon Heron posted on Saturday, August 16, 2008 - 9:17 am
|
|
|
Thanks Linda, I'll have to track down my license invoice first - I fear it may still be sitting in a secretary's in-tray |
|
|
I created a 2 class latent class model, and now want to use the classes as independent variables in regression analyses. when i try to save the classes to a file, it will only save variables in the usevariables command, and the classes/cprobs but not the rest of the variables in the original input file. i don't want to add the other variables in usevariables command because they are not relevant to creating the classes. is there a way to do this, or merge the two files? any help is appreciated. |
|
|
See the AUXILIARY option of the VARIABLE command. |
|
Back to top |
|
|