How to applying rotation criterion on IRT parameters via mod2values in context of an Exploratory model?

[Seongho Bae]

Dear all,

Anybody knows how to applying rotation criterion on IRT parameters via mod2values in context of an Exploratory model?

--

Seongho Bae

[Phil Chalmers]

I don't think you can, but you can always combine the output from coef(mod, rotate = 'rotation') and simply replace the slope values from the data.frame returned from mod2values().

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.

[Seongho Bae]

Humm, In exploratory MIRT context, May I have to rotate IRT parameters for Full-information Item Factor Analysis purpose?

If I have to rotate IRT parameters for make comparable with Full-information item factor analysis, I'll use coef() method.

Seongho Bae

2015년 3월 13일 금요일 오전 9시 6분 13초 UTC+9, Phil Chalmers 님의 말:

[Phil Chalmers]

On Thu, Mar 12, 2015 at 9:27 PM, Seongho Bae <seongh...@gmail.com> wrote:

Humm, In exploratory MIRT context, May I have to rotate IRT parameters for Full-information Item Factor Analysis purpose?

Usually for exploratory models you don't care about the raw coefficients anyway because they are rotationally indeterminate, so they aren't really of much use. The same issue happens in linear factor analysis, though the raw coefs are often hidden from the user by default and only present the standardized loadings (e.g., see the psych package). Better to use a standardized metric for exploratory models, which is what summary() is for.

Phil 

[Seongho Bae]

Okay, I got it!

My paper reviewer says, "You may check parameter a (for discriminant people) over |.4| even you were used exploratory multidimensional item factor analysis model. Because, basically that is multidimensional item response model. You can also use standardize factor loadings, but check IRT parameters first." So I just need to check.

Seongho Bae

2015년 3월 13일 금요일 오전 10시 37분 41초 UTC+9, Phil Chalmers 님의 말:

[Phil Chalmers]

They are completely incorrect about the |.4| rule for exploratory models, because of the rotational indeterminacy issue. Unless they are referring to the raw slopes after rotation, but even then I have not idea where that rule of thumb came from (there is no reference of that anywhere to my knowledge, and I would completely advise against it's use anyway since it's wholly arbitrary). 

You should check the standardised solution first, not the raw coefficients, and more specifically check the standardized commonality value (h2, which is essentially an effect size ranging from 0 to 1) and the rotated factor loadings. The h2 value is actually the standardized MVDISC coefficient, which they may have heard of before. So, perhaps you should let them know, politely, that they are incorrect in their recommendation on where to start. MIRT is often just nonlinear factor analysis, and in that setting one should be looking at standardized estimates for EFA, and IRT is no different in that respect. Cheers.

Phil

[Seongho Bae]

Okay, That is import advice to me!

Unfortunately, My reviewer may not recognize differences among Principle Component Analysis and Common Factor Analysis. So I describe what is exploratory/confirmatory full-information item factor analysis technique in reply. Maybe, my reply affected that comment what check raw IRT coefficients based on the context of UIRT (He/She referenced articles of F. B. Baker's book and any other UIRT researches published in Korean.).

When I consider standardized MVDISC coefficient as 'effect size', Can I use linguistic expression who suggest the Cohen (1992)?

Seongho

2015년 3월 13일 금요일 오전 11시 5분 2초 UTC+9, Phil Chalmers 님의 말: