standard errors cannot be computed as GroupPars.MEAN_1 is released

44 views
Skip to first unread message

KK Sasa

unread,
Feb 22, 2016, 2:08:49 AM2/22/16
to mirt-package
Hi Phil,

I got a problem that may be a bug. Please find the attached file.

The GroupPars.MEAN_1 and GroupPars.COV_11 are released and put constraint on sum(item categories)=0. However, the s.e. cannot be computed.

After testing, I think the problem results from releasing GroupPars.MEAN_1.

Best,
Keith
data.R
data

Phil Chalmers

unread,
Feb 22, 2016, 8:25:37 AM2/22/16
to KK Sasa, mirt-package
Confirmed, it's definitely a bug. I'll see if I can fix it before the next release.

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.
For more options, visit https://groups.google.com/d/optout.

Phil Chalmers

unread,
Feb 24, 2016, 10:19:09 AM2/24/16
to KK Sasa, mirt-package
On second look, this actually seems fine to me and is really just a consequence of including non-linear constraints. The hessian is looking at the full parameter space and computed that way, which is not identified, while the model itself imposed a nonlinear restriction to actually estimate it. So the information matrix from solve(extract.mirt(mod, 'vcov')) needs to be fixed to include information about this constraint so that it's full rank. Seems pretty complicated if you ask me, but maybe there's an easier approach that I'm not quite seeing.

Phil

Keith Lau

unread,
Feb 27, 2016, 1:25:27 AM2/27/16
to mirt-package, genw...@gmail.com
Hi Phil,

Perhaps you could try to fix GroupPars.MEAN_1 to zero by modifying two lines:

~~
pa = para[1:(length(para)-2)]          ----->  pa = para[1:(length(para)-1)]
sv2[c(ind,ind2),"est"] = TRUE        ------>  sv2[c(ind),"est"] = TRUE
~~

The SE could be computed in such case. So, that's why I suspected the inner code does not treat GroupPars.MEAN_1 well.

Phil Chalmers於 2016年2月24日星期三 UTC+8下午11時19分09秒寫道:

Phil Chalmers

unread,
Feb 27, 2016, 10:36:37 AM2/27/16
to Keith Lau, mirt-package
Yes, the baseline model here is identified when the mean is constrained, but the standard errors are still not correct (they don't take into account the equality information in heq, so some may be to large/small). 

There are really only two solutions in my mind to fix this: 1) some transformation matrix is built up based on the heq() function so that pre-post multiplication of the unconstrained hessian is possible, or 2) don't use the EM, and instead adapt the BL method to compute this matrix numerically. The latter seems more feasible at the moment because conceptually the former is harder to obtain (there could be a trick that I'm not seeing though). Either way, I think for now I'll have to disable the SE computations when the alabama or Rsolnp algorithms are used.

Phil
Reply all
Reply to author
Forward
0 new messages