Attachment | Size |
---|---|
infotest.R [6] | 6.05 KB |
I'm experimenting with OpenMx's capabilities to calculate the information matrix (and standard errors) in the context of IFA models with priors on some item parameters. e.g., see for example, the 3PL model in the documentation: https://openmx.ssri.psu.edu/docs/OpenMx/latest/ItemFactorAnalysis.html#a-3pl-with-bayesian-priors
Say one does the following changes to the example inside mxComputeSequence (e.g., see also end of attached code):
Remove mxComputeNumericDeriv()
Add mxComputeOnce('fitfunction', 'information', "sandwich")
If I do this, I do get standard errors, but no information matrix in the output. Just wondering how the standard errors are calculated in this case and how to obtain the information matrix. I do see there being a note in the example possibly suggesting that the above approach ought not to work or is not recommended in the context of priors, but at this point I'm just experimenting to figure out how things work.
Many thanks!