Longitudinal IRT - non-positive definite matrix

56 views
Skip to first unread message

Claire Chen

unread,
Aug 19, 2020, 3:56:03 PM8/19/20
to mirt-package
Dear Professors,

I am a novice mirt user and am trying to obtain trait scores for a repeated measure (5 items, 5-point likert scale) across 3 time points. The ultimate goal is to use the trait scores as indicators of further latent growth modeling.

Using the R code adapted from  https://philchalmers.github.io/mirt/html/Longitudinal-IRT.html ),  I got the warning message as below:
Iteration: 1, Log-Lik: -23702.913, Max-Change: 1.01782
Iteration: 2, Log-Lik: -23833.759, Max-Change: 0.29002
Iteration: 3, Log-Lik: -23683.620, Max-Change: 0.19118
Iteration: 4, Log-Lik: -23611.674, Max-Change: 0.13221
Iteration: 5, Log-Lik: -23574.670, Max-Change: 0.09855
Iteration: 6, Log-Lik: -23564.176, Max-Change: 0.08492
Iteration: 7, Log-Lik: -23584.722, Max-Change: 0.79980Warning messages:
1: Latent trait variance-covariance matrix became non-positive definite. 
2: In log(eigen(sigma, symmetric = TRUE, only.values = TRUE)$values) :
  NaNs produced
3: In log(eigen(sigma, symmetric = TRUE, only.values = TRUE)$values) :
  NaNs produced
4: In log(eigen(sigma, symmetric = TRUE, only.values = TRUE)$values) :
  NaNs produced
5: In log(eigen(sigma, symmetric = TRUE, only.values = TRUE)$values) :
  NaNs produced
6: In log(eigen(sigma, symmetric = TRUE, only.values = TRUE)$values) :
  NaNs produced
7: In log(eigen(sigma, symmetric = TRUE, only.values = TRUE)$values) :
  NaNs produced

My concern was that there is an item (s2) not having the same number of response categories as the other time points. I still got the same warning after I removed this item or collapsed the response category. Could you please take a look at the code and raw data (not collapsed) uploaded in the link below and kindly let me know how to move forward?

Another question is about another set of repeated measure (4 items, 3 time points) - I was able to the run the same code without getting any warning; however, some of the covariances were >1. Could I trust the derived scores in this case?

Any suggestions would be much appreciated. Thank you!
Claire

Phil Chalmers

unread,
Sep 3, 2020, 3:46:55 PM9/3/20
to Claire Chen, mirt-package
Hi Claire,

It looks like your model was slightly too unstable to use such a low quadrature point number for numerical integration, which was causing the vcov matrix between the latent traits to take unrealistic values during the successive EM iterations. However, I bumped it up to quadpts = 9, which took notably longer, but the model did end up converging after 81 EM cycles. HTH. 

Phil


--
You received this message because you are subscribed to the Google Groups "mirt-package" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mirt-package...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mirt-package/b40cb7e2-71b7-4d0d-95fb-cf39cb1e8820n%40googlegroups.com.
Reply all
Reply to author
Forward
0 new messages