Tim Forsyth posted on Wednesday, June 04, 2014 - 10:36 pm
This is a very basic question. How does the model used for Latent Class analysis give weight to the subgroups? Obviously they aren't equal. I haven't been able to find a clear cogent explanantion. I have a Masters degree in biostatistics so I understand the basic principles.
Think of a latent class analysis for one continuous, skewed variable. The class sizes are chosen to best fit that observed distribution - so smaller class sizes for classes further out in the tail.
The class sizes are estimated simultaneously with the other parameters using say ML. A good mixture book is McLachlan and Peel (2000).
Tim Forsyth posted on Thursday, June 05, 2014 - 10:40 pm
I am still not clear why some subclasses get more weight than others. Is it that the classes that are more sparsely populated get less weight? How some classes fit the distribution better especially if they are binomial (not continuous). Sorry to be so dense about this. Is there a good resource that I could look at that would help explain this to me.