Multilevel mixture IRT simulation PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 şeyma posted on Tuesday, March 11, 2014 - 8:57 am
Dear Mplus users,

1. I am trying to running example 10.5 monte carlo study. But I want this data include a covariate variable. (for example a gender variable: 600 females in latent class 1, 400 males in latent class 2; 400 females in latent class 2 and 600 males in latent class 1=2000 observations totaly gender splşt is 50% and . How can I add such a variable in this study?

2. Another question is: Which prior paramter used for this simulation study? Can we generate response patterns in this example based on item response model? Because I want to item diffuculties differ between latent classes.

Thank you.

Şeyma
 Bengt O. Muthen posted on Tuesday, March 11, 2014 - 12:10 pm
First, note that this is an advanced analysis that calls for experience with Mplus Monte Carlo simulations.

1. You add a binary x variable (see UG chapter 11 for how to do that) which influences the latent class variable ("c ON x").

2. You can use parameter values from a real-data analysis.
 şeyma posted on Tuesday, April 08, 2014 - 4:34 am
Dear Bengt Muthen,

I try to fit my data to the multilevel mixture item response model. For this reason I used example 10.5. I guess the syntax is for 1-PL model or Rasch model. I have now class spesicific item diffuculties.

I want to use these parameter for differential item functioning detection. But I wonder that these item diffuculties directly comparable? Should I bring the parameters the same scale?

Thanks,

Şeyma
 Bengt O. Muthen posted on Tuesday, April 08, 2014 - 8:29 am
Ex 10.5 uses a 2PL model.

To have a comparable latent variable scale you need to start with equality of loadings and thresholds across the classes. Then you can free up some of those equalities to capture DIF.
 şeyma posted on Thursday, September 18, 2014 - 2:39 am
Dear Muthen,

For multilevel mixture irt model I generated the data in C#.
I generated two student latent classes and two school latent classes. The diffuculty parameters differ across student latent classes, so I generate 20 diffuculty parameters for student class one and 20 parameters for student class two.
For dif detection,
Student class 1 school class 1 ,I use the first 20 diffuculty parameters.
Student class 1 school class 2, I use the first 20 diffuculty parameters but only some of them differ.
Student class 2 school class 1 I use second 20 diffuculty parameters.
Student class 2 school class 2 I use second 20 diffuculty parameters but only some of them differ.

For analyzing the simulated data I use example 10.5. I wonder that in output the threshols are so.
1 1;1 2; 2 1 and 2 2. Which is school class and which is student class. How is the order?

In my output 1 1 and 2 2 were almost same estimated.. And 1 2 and 2 1 were almost same estimated.
 Linda K. Muthen posted on Thursday, September 18, 2014 - 11:05 am
The order of the thresholds corresponds to the order of the categorical latent variables in the CLASSES statement.
 Xiaorui  posted on Thursday, December 03, 2015 - 5:30 pm
Dear Dr Muthen,

Is there anyway to simulate a multilevel IRT model and run the simulation data using single level IRT simultaneously?

Thanks!
Xiaorui
 Linda K. Muthen posted on Thursday, December 03, 2015 - 5:56 pm
You would need to do this in two steps. First generate and save the data. Second, analyze the date using TYPE=MONTECARLO.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: