I recently purchased the full combination version. I've had a lot of fun so far!
I'd like to explore potential latent classes for controversial comorbid disorders with the addition of covariates (e.g., age, gender, race, income, etc.) and distal outcomes (e.g., use of services) by taking my total sample, randomly divide it, explore latent classes with one half, and then test those classes with the other half.
However, I'm not sure if it is legitimate to 1) explore latent classes of DSM-IV-R comorbid diagnoses, for an example only, let's say, attention deficit disorder and conduct disorder -- my data is retrospective, multi-year, cross sectional and ONLY indicates whether a child is/isn't (1/0) diagnosed with the disorder, and then 2) test classes explored with one half of the sample with the other.
2) Exploration-validation can be useful if you have a large sample to split. I don't know what you have in mind when you say "test classes", but there are many ways the validation part can be done and it is not clear to me which is best.
anonymous posted on Saturday, October 10, 2009 - 8:06 pm
Hello, I am attempting an exploratory LCA of different types of DSM-IV comorbidity in a subpopulation (only individuals with one type of disorder) taken from a large, nationally representative sample. When I attempt to obtain 3 or 4 classes, I receive the warning that the log likelihood estimate was not replicated and more random starts are needed. After changing the number of random starts from 500 to 10000, however, I continue to obtain the same warning. I've also tried to increase the number of iterations with no solution. Both males and females are included in the subpopulation sample and I know that the rates of diagnoses differ considerably across gender. Could this be the problem? What do you suggest? thanks.
If you had no trouble extracting two classes, this may indicate that you are trying to get too much information from the data. You may be trying to extract too many classes. For further comments, please send the full output and your license number to email@example.com.