|
Message/Author |
|
|
I am working on fitting a multivariate mixture model in Mplus (fully unconstrained MVN mixture model) and having a problem that I am struggling with. There are 5 variables in my dataset. They are all in the same continuous scale. The lowest score possible is 55 in all variables and one variable has quite a few cases with the lowest score (~10%). While the 2-class solution converges without problems, the 3-class solution doesn't converge due to the problematic cases that I just described. All of those cases (with the same lowest score in that variable) were assigned into one of the 3 classes which results in no variability in the estimates for that variable. Since there is no variability for one variable, the model is having convergence problems we believe. Is there any way that I can solve this problem? Thanks! Seniz |
|
|
Can't say for sure without looking at your runs and data. But you can try the new skew-t distribution option in Mplus Version 7.2 which allows a class with a heavily censored distribution. |
|
ksk posted on Monday, October 16, 2017 - 10:55 am
|
|
|
Dear Dr. Muthen, I'm working on LPA with 5 indicators. I have a question about the analysis. Below is the command I used to run 6-class model: ========================== .... usevariables = ZMEdu ZINCTNR ZBDI ZSTAIT ZPDHchpa; missing are all (-99); classes = c(6); ANALYSIS: Type = mixture; STARTS = 200 50; LRTSTARTS = 0 0 200 40; MODEL: %OVERALL% PLOT: type = plot3; series = ZMEdu(1) ZINCTNR(2) ZBDI(3) ZSTAIT(4) ZPDHchpa(5); OUTPUT: SAMPSTAT TECH1 TECH11 TECH14; ======================= 1- to 5-class models ran without any convergence issue. But only for the 6-class model, I got a warning below: THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS 0.232D-14. PROBLEM INVOLVING THE FOLLOWING PARAMETER: Parameter 5, %C#1%: [ ZPDHCHPA ] I double-checked the ZPDHCHPA variable. The distribution was perfectly normal, but it was fine, not too skewed. When the warning appears, what would you recommend to resolve the issue? Thank you, Sue |
|
ksk posted on Monday, October 16, 2017 - 10:56 am
|
|
|
Dr. Muthen, Another questions related to the posting above is about an warning shown below. The warning appears consistently throughout 1- to 6-class models: *** WARNING in MODEL command All variables are uncorrelated with all other variables within class. Check that this is what is intended. 1 WARNING(S) FOUND IN THE INPUT INSTRUCTIONS I meant that some variables should be correlated to some extent within class. Would the warning be ignorable? Or do I need to do something to specify some correlations between the variables? Thank you again, Sue |
|
|
Regarding your first posting, please send your full output to Support along with your license number. Re your second q., mixture models do not necessarily allow for correlations within class. Factor mixture models do - see our website for many papers and short course videos/handouts. |
|
Back to top |
|
|