BILOG Vs. MIRT

[Flavio Fiuza]

Hello Phill! Hope you're doing fine!

I'm trying to migrate my estimation procces from BILOG-MG to MIRT, but i'm having some issues when comparing results from both softwares!

I'm in this trouble from a couple of weeks with no luck, so maybe you could help me. 

For this example I generated a dataset with two groups of examinees (5000 individuals each) and two forms. The first form has 30 unique itens, the second has 23 unique itens and 7 itens extracted from the first form. All the 53 itens have their real parameters and all the 10k examinees have their real scores. From these infos I generate their probabilities of responding correctly each item and then, using bernoulli distribution, I generate a dichotomous response vector for each individual.


Using this data I run a multiplegroup with the first 30 itens fixed (I need it to be fixed) to generate estimates for the itens parameters and individuals scores.

Then I repeat this procedure in BILOG-MG.

I used the same priors, theta_lim, quadpts, and so, in both softwares. But still I have some discrepant/off-the-wall results. Could you help me?

I've attached my code (mg test), the two response vector datasets (group1 and group2), the real parameters of the two forms (form1 and form2) and the real scores from both groups (scores_g1 and scores_g2).

The group1/group2 datasets are the response vector of the individuals, the first line consist of the itens names. So G1-1 is the first item answered by the first group, while the item G-6 is the 6th item answered by the second group (and so on). In the group2 dataset you will be able to see that the last 7 itens are from the form1. Apart from the first line, each row represents an individual

In the scores_g1/scores_g2 datasets you will find, at the second column, the real scores of group1 and group2, respectively.

In the form1/form2 datasets you will see: the item ID in the first two columns (it's duplicated, sorry), the real a1 parameters in the third column, the real b parameters in the fourth column, and the guessing parameter in the fifth column.

How would you perfom this calibration in MIRT? Am I doing right while using the mirt and multipleGroup commands?

Thanks in advance!

Best,
Flavio

[Phil Chalmers]

On Wed, Nov 8, 2017 at 7:17 PM, Flavio Fiuza <flavio...@gmail.com> wrote:

Hello Phil! Hope you're doing fine!

I'm trying to migrate my estimation procces from BILOG-MG to MIRT, but i'm having some issues when comparing results from both softwares!

I'm in this trouble from a couple of weeks with no luck, so maybe you could help me. 

For this example I generated a dataset with two groups of examinees (5000 individuals each) and two forms. The first form has 30 unique itens, the second has 23 unique itens and 7 itens extracted from the first form. All the 53 itens have their real parameters and all the 10k examinees have their real scores. From these infos I generate their probabilities of responding correctly each item and them, using bernoulli distribution, I generate a dichotomous response vector for each individual.

Using this data I run a multiplegroup with the first 30 itens fixed (I need it to be fixed) to generate estimates for the itens parameters and individuals scores.

Then I repeat this procedure in BILOG-MG.

I used the same priors, theta_lim, quadpts, and so in both softwares. But still I have some discrepant/off-the-wall results. Could you help me?

I rather doubt that the two packages will give the same answers in multiple group settings. BILOG-MG, as far as a I know, uses the classical (and largely suboptimal) parameterization of the IRT models, while mirt uses the slope-intercept form for everything. Moreover, applying priors under different parameterizations won't work either, because these are not transformation invariant. So even here I imagine the software packages will differ.

 

I've attached my code (mg test), the two response vector datasets (group1 and group2), the real parameters of the two forms (form1 and form2) and the real scores from both groups (scores_g1 and scores_g2).

How would you perfom this calibration in MIRT? Am I doing right while using the mirt and multipleGroup commands?

It looks fine with mirt. Though in all honesty, I've never actually used BILOG (sources tell me that mirt provides more accurate empirical results, but I can only take that as second-hand reassurance). 

Phil

 

Thanks in advance!

Best,
Flavio

--
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+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Flavio Fiuza]

Thanks for the quick answer! 

So, just to reassure, the way I performed the fixed parameter estimation on MIRT (the way I encoded, and all) is the right way. Yes?

Thanks a lot!!! 

To unsubscribe from this group and stop receiving emails from it, send an email to mirt-package...@googlegroups.com.

[Seongho Bae]

I had experienced the fixed item parameter linking using the mirt and BILOG-MG both.

Bilog-mg uses prior default values calibrate free parameters, but mirt is not until you do not specify. However, if you define them during fixed parameter linking, you may get identically equal values in mirt::mirt.

Seongho

[Sohee Kim]

Hello,

I am a graduate student in Oklahoma State University.

Now I'm studying about the FIPC with mirt package.

I had experienced the FIPC using the PARSCALE only, thus I am struggling for doing the FIPC linking with mirt package.

I just wonder do you have any reference (R codes, articles, dissertation, and whatever do you have) for FIPC linking in mirt package? I really need your help.

2017년 11월 9일 목요일 오전 11시 34분 22초 UTC-6, Seongho Bae 님의 말:

[Phil Chalmers]

FIPC = Fixed item parameter calibration?

Perhaps describing the process would help us understand what exactly you are trying to do. There are many ways to link tests, so making sure we have the correct definitions and algorithms to work with is important. 

Phil

--

[Seongho Bae]

Hello, Phil.

I answered a few months ago to Sohee Kim (She asked privaitely to me).

She just wanted to calibrate a model like this way.

Best,

Seongho

2019년 2월 13일 수요일 오전 1시 51분 21초 UTC+9, Phil Chalmers 님의 말:

To unsubscribe from this group and stop receiving emails from it, send an email to mirt-p...@googlegroups.com.