

Problems with mixture Rasch model 

Message/Author 

Tongyun Li posted on Friday, September 27, 2013  10:01 am



Hi Drs.Muthen, I'm trying to use Mplus to estimate Mixture Rasch model. When I generate person parameters from N(0,1) distributions for both latent classes and I specified the model as follows, Mplus could recover the model parameters very well. MODEL: %OVERALL% f BY u1u15* (1); f@1; [f@0]; %c#1% [u1$1u15$1]; %c#2% [u1$1u15$1]; However, when I generated the person parameters from different distributions (e.g. LC1:N(0,1);LC2:N(1,1)) and I specified the model as MODEL: %OVERALL% f BY u1u15* (1); f@1; %c#1% [f@0]; [u1$1u15$1]; %c#2% [f]; [u1$1u15$1]; The parameter estimates I obtained seems not correct. Could you point out what was wrong with my code? And if both the factor means and variances are different for the two latent classes(e.g. LC1:N(0,1);LC2:N(1,.25)), how can I specify the model syntax? Thank you very much! 

Tongyun Li posted on Friday, September 27, 2013  10:01 am



Another question is about the person parameter estimates. I saved FSCORES in a separate file, but in that file there are two columns of person parameter estimates, the values are very close for each person. I'm wondering why there are two columns of person parameter estimates. Also,does Mplus provide standard error for person parameter estimates for mixture Rasch model? Thank you! 


Your setup MODEL: %OVERALL% f BY u1u15* (1); f@1; %c#1% [f@0]; [u1$1u15$1]; %c#2% [f]; [u1$1u15$1]; is correct. If you have different factor variance in the two classes you say f; within each class. Perhaps you are not using enough STARTS? It is not clear if you are doing a Monte Carlo simulation in which case you don't want to use STARTS, but want to use good starting values to not end up in local optima. If that doesn't help, send output to Support. Two columns of factor scores are given: For the most likely class (denoted by underscore C) and for the mixture over classes. 

Back to top 

