Message/Author 

Ali posted on Tuesday, February 18, 2014  3:20 am



Hello I am using Mixture IRT model to fit 1 calss, 2 class, and 3 class with 20 data set. In 2 class, there is no any problem to fit, also, THE BEST LOGLIKELIHOOD VALUE HAS BEEN REPLICATED. But , in 3 class , sometimes it showed THE BEST LOGLIKELIHOOD VALUE HAS BEEN REPLICATED , but meanwhile shows ：THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NONPOSITIVE DEFINITE FIRSTORDER DERIVATIVE PRODUCT MATRIX.； and ：ONE OR MORE PARAMETERS WERE FIXED TO AVOID SINGULARITY OF THE INFORMATION MATRIX.； Are results trustful such item parameters and the values of BIC? Or does it mean that model can・t fit the data? In addition, sometimes the results of 3 classes showed ：THE BEST LOGLIKELIHOOD VALUE WAS NOT REPLICATED. ： and I increased the starting value = 100 20 to 1000 200 , but it still showed that ：THE BEST LOGLIKELIHOOD VALUE WAS NOT REPLICATED. ： Does that mean the model can・t fit the data too? Or should I increase starting values more? 


The best loglikelihood must be replicated. Even when it is replicated, there can be other problems. I would need to see the output with the error message and your license number at support@statmodel.com to determine what the problem is. 

Ali posted on Monday, February 24, 2014  1:45 pm



Thanks! Can I obtain the estimate of individual ability while I am using mixture IRT model? Or, I just can only get the membership and mean of group ability ? 


You get the individual ability estimate for each latent class and also the ability estimate mixed over classes. For a factor names F, they are called F_C and F, respectively. 

Ali posted on Monday, February 24, 2014  10:49 pm



So, do you mean the output which I can get the the estimated of ability for each class and the whole sample's estimated ability ? But, could I get the estimated of ability of each person in a class? For example, if there are 200 people in class 1 , could I also get 200 people's estimated ability in class 1? 


I am talking about each person's factor score  you get 2 versions of each person's factor score (classspecific and mixed over classes). You get that using the Savedata command and Save  fscores. 

Ali posted on Tuesday, March 04, 2014  1:04 am



Thank you. I am trying to find the fitted model in a mixture Rasch model. I fitted the data in 2 classes , 3 classes and 4 classes. BIC for 2,3,4 classes are 45190.796,45053.329 45083.17.From the BIC value, 3 classes should fit well, but the output gave me the message"ONE OR MORE PARAMETERS WERE FIXED TO AVOID SINGULARITY ...". Should I choose the 3 classes model? 


It depends which parameter was fixed and why. Please send the output and your license number to support@statmodel.com. 

Back to top 