Hi, If the LRT results prefer 2 to 1 classes and 3 to 2 classes, can I conclude that 3-class is the optimal answer or I have to say 2 and 3 classes are all optimal answers under certain conditions? Thanks
With LRT, you look for the number of classes where the p-value is greater than .05. This points to the best solution as that with one less class. So if the p-value is first large with 4 classes, this points to 3 classes. Once a large value is obtained, further interpretation should not be done.
Ruixue Wang posted on Monday, April 04, 2011 - 12:16 pm
thanks.I have another question I met this WARNING: THE LATENT VARIABLE COVARIANCE MATRIX (PSI) IN CLASS 1 IS NOT POSITIVE DEFINITE. THIS COULD INDICATE A NEGATIVE VARIANCE/ RESIDUAL VARIANCE FOR A LATENT VARIABLE, A CORRELATION GREATER OR EQUAL TO ONE BETWEEN TWO LATENT VARIABLES, OR A LINEAR DEPENDENCY AMONG MORE THAN TWO LATENT VARIABLES. CHECK THE TECH4 OUTPUT FOR MORE INFORMATION. PROBLEM INVOLVING VARIABLE I. is it appropriate to set i-s@0 to avoid this?
1.what's the range for small? [-1,0]? 2 If the variance is too large negative value,such as -10,I can not fix it to zero,what should I do to avoid negative variance? 3If there is negative variance, are the values of estimated parameters still trustworthy?
Then it does not help you make a decision. I think you may benefit from listening to the Topic 5 and Topic 6 course videos where these issues are discussed. Also, please see the following paper which is on the website:
Nylund, K.L., Asparouhov, T., & Muthén, B. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling. A Monte Carlo simulation study. Structural Equation Modeling, 14, 535-569.
Regarding your negative varance problem: I bet you estimated class invariant growth factor variances across latent classes, which might be sometimes too restrictive leading to inadmissable variance estimations. Try to relax the equal variance assumption for the latent intercept across classes first.
LRT in your last post simply points to 4 classes. You could further test against 5 classes. But I would suggest to solve your variance problem first, before interpreting LRT.
Hi, 1.I have 1500 subjects, I want to analyse 500 every time. How can I input 1500 then let Mplus do this? Is there loop commands in Mplus? 2.If I want to do a 1-4 classes, I have to change the class number every time, is there a way to output them together without manually changing the number? 3.How can I extract specific outputs such as BIC,P-value of LMR,BLRT,Entropy and then do some caclulation about this. Here I mean if question 1 can be solved, then extracting all of values I need for the 3 data set at a time. Thank you.