Message/Author 

Mark posted on Thursday, November 07, 2013  12:53 pm



Dear Dr. Muthen: I am seeking to run a job in MPlus to conduct a finite mixture modeling analysis that will replicate precisely what is done in McLachlan's EMMIX program as an "unrestricted" model. Such an unrestricted model would allow both the means and variances of the latent class indicators to vary. Covariances among the latent class indicators would be allowed to vary. I am working with a data set of 150 observations (cases) with 2 quantitative indicators of the latent classes. I have made a stab at the code in MPlus (which I am new at doing) and this is what I have come up with, but I think I am missing the statements to relax all the constraints I seek to relax. Would you be so kind as to take a peek at this code to let me know what I am missing? Thank you! TITLE: TESTDATA mixture problem DATA: FILE IS testdata.csv; VARIABLE: NAMES ARE y1y2; USEVARIABLES ARE y1y2; CLASSES = c (2); ANALYSIS: TYPE = MIXTURE; ESTIMATOR = MLR; STARTS = 1000 50; LRTSTARTS = 0 0 100 20; Model: %OVERALL% y1 WITH y2; %c#1% [y1y2*1]; y1y2; %c#2% [y1y2*1]; y1y2; OUTPUT: TECH1 TECH11 TECH14; SAVEDATA: FILE IS TESTDATA12.savedat; SAVE=CPROBABILITIES; 


You are close. You want to add y1 WITH y2; within each of the two classes. Your setup holds the covariance equal across classes. 

Mark posted on Thursday, November 07, 2013  1:36 pm



Thank you! How would I allow the covariance to be estimated across classes and not held equal. Thanks again. 


Add y1 WITH y2; within each of your 2 class statements: %c#1% [y1y2*1]; y1y2; y1 WITH y2; %c#2% [y1y2*1]; y1y2; y1 WITH y2; 

Mark posted on Thursday, November 07, 2013  1:59 pm



Thanks for added clarification  I appreciate your help. 

Mark posted on Thursday, November 07, 2013  3:14 pm



A final query Dr. Muthen, as I move up through more latent classes with this program, say 3 or 4 with these data, does the code develop in a predictable fashion? I realize I only worked out the code for 2 classes, but a 3 class model will clearly make sense in this dataset. (Again, being new to MPlus, I want to proceed carefully). Thanks! 


Yes, the code develops in a predictable fashion. 

Back to top 