DIF: the #of categories within each group is not equal to the total # of categories

[yingyi....@gmail.com]

Hi Phil and all,

I was trying to do DIF in Uni-dimensional IRT, but so far encountered an issue that the # of response categories are not the same for each group.

For example, for item 1, the response range should be a 5-point scale (1, 2, 3, 4, 5); group 1 responses were 1-5, while group 2 no one choose 5 (so group 2 response were 1-4). 

And when I was trying to do DIF in R, error message occurred:  Error: Multiple Group model will not be identified without proper constraints (groups contain missing data patterns where item responses have been completely ommited or, alternatively, the number of categories within each group is not equal to the total number of categories) (which is basically the same error I got when I was trying to do the same analyses in other IRT software, e.g. IRT PRO).

So Phil, I am wondering if R could handle this issue? Or is there any alternative way to do this analysis in mirt package?

Thanks in advance; any comments are appreciated!

Yingyi

[Felix Fischer]

A straightforward option would be to collapse category 5 into category 4 for the whole dataset (and i can't think of anything else).

Best, Felix

[Phil Chalmers]

Felix's suggestion is a good one. Trying to estimate parameters freely within a given group that has no data obviously leads to non-identification, so this must be avoided. This is all well and good when all the parameters are constrained across groups for these items as information is essentially 'borrowed' from the groups with data, but otherwise it spells trouble. 

As an aside, I'm glad to hear IRTPRO has this bevaviour too, as I believe it is the correct default to choose rather than forcing something weird in the likelihood equations. 

Phil

On Thu, Apr 26, 2018 at 4:55 AM Felix Fischer <drhans...@gmail.com> wrote:

A straightforward option would be to collapse category 5 into category 4 for the whole dataset (and i can't think of anything else).

Best, Felix

--
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.

[yingyi....@gmail.com]

Hey Phil and Felix,

Thanks so much for your prompt reply. I thought of that option, too. 

But I ended up not doing so because I feel like forcing 5 categories into 4 not quite makes sense, though. 

I feel like groups choosing 5 score values behave differently than groups who did not choose 5...

Yeah but I agree that that is the only option I have now, and I will give it a try! 

Thanks so much again :)

Yingyi