Extract information for MAP Estimation
Message/Author
 Luc Watrin posted on Tuesday, August 23, 2016 - 6:30 am
I analysed a multidimensional forced-choice questionnaire with a Thurstonian IRT Model. Everything worked well and now I'm trying to get a better understanding of how exactly the MAP factor scores are computed. More specifically, I'd like to know if I can extract the relevant information from the estimated model to be able to compute them manually from the raw scores. Is that possible at all? As I understand it, the computation is an iterative process. In that case, I'd probably use Mplus or R to compute them separately. Thanks in advance!
 Bengt O. Muthen posted on Tuesday, August 23, 2016 - 11:47 am
With ML, Mplus uses EAP. And, yes, this is an iterative process which would have to be programmed.
 Luc Watrin posted on Wednesday, November 16, 2016 - 9:12 am
The estimated covariance matrix for the latent variables I get in my model is almost singular. While I understand that I'll have to figure what exactly causes this, I am wondering why Mplus isn't running into problems computing the factor scores. Does Mplus alter the matrix in any way before proceeding with further computations?
 Bengt O. Muthen posted on Wednesday, November 16, 2016 - 5:57 pm
No alteration is done. If it isn't deemed singular by Mplus it goes through.
 Dirk Pelt posted on Wednesday, August 02, 2017 - 6:16 am
I am dealing with a similar challenge as Luc. To help with my algorithm, may I ask which optimization procedure is used by MPLUS in MAP (BFGS or L-BFGS-B or the Newton-Raphson method for example)?

Best, Dirk
 Bengt O. Muthen posted on Wednesday, August 02, 2017 - 4:32 pm
It's a Fletcher-Powell type optimizer.
 Dirk Pelt posted on Thursday, August 03, 2017 - 2:05 am
Thank you! Is the prior weighed by any means? I am getting kind of similar results but I get much more "shrinkage" (i.e. smaller variances of the estimated traits). Best, Dirk
 Bengt O. Muthen posted on Thursday, August 03, 2017 - 5:33 pm
No weighting.
 Dirk Pelt posted on Friday, September 01, 2017 - 5:19 am
Thank you for your answers. Is there any way to let Mplus output the standard errors (SEMs) of the factor scores? Best, Dirk
 Bengt O. Muthen posted on Friday, September 01, 2017 - 12:35 pm
We give SEs in most cases. And with Bayes we give the whole distribution for a person's factor score. Be sure to use Version 8.