|
|
Order of latent classes, GMM |
|
Message/Author |
|
Aaron Fisher posted on Thursday, December 03, 2009 - 6:56 pm
|
|
|
Hello. I have a GMM with 4 latent classes. When I estimate the model using this syntax: classes = c(4); analysis: type = mixture; starts = 25 2; model: %overall% i s | d2@0 d3@1.75 d4@3.75 d5@7.50; c#1 on d1; c#2 on d1; c#3 on d1; %c#2% i s; %c#3% i s; %c#4% i s; I discover that there are negative error variances for the slopes for classes 2 and 3. I would like to fix these to zero however, every time I try to do so the classes change order and I am fixing the variances for the wrong class. I have seen advice on this message board re: using starting values to get a particular order for the latent classes. Nevertheless, I have been unable to effectively manipulate the order of the classes. Any help you can give will be greatly appreciated. |
|
|
I would try using growth factor means as starting values. If this does not help, please send the input, data, output, and your license number to support@statmodel.com. |
|
|
Thank you for your response. I believe I have tried that. I specify those within the class-specific commands? for example: %c#3% [i*3.4]; [s*-.27]; i s; And this has not worked. If this is what you were suggesting, please let me know and I will send along the data, output, and lic #. Thanks again. |
|
|
Yes, this is what I was suggesting. |
|
Back to top |
|
|