which.par

[jksha...@gmail.com]

I am currently attempting to run a DIF analysis using the MIRT package, however I am struggling to understand one particular section. I am not sure I completely understand the “which.par” variable.

In the example provided in the PDF file on MIRT, the parameters that are put in there are “a1” and “d1” which are referring to the slope and the intercept.

However, what does that exactly mean? Is “a1” referring to the discrimination of the items? Then what exactly does “d1” mean? That is not the difficulty of the item is it?

Concerning my own example, I am also getting an error when I just run “d1” as the only parameter in the “which.par”. The error is as follows:

“Error in checkForRemoteErrors(val) :
4 nodes produced errors; first error: Item 1 does not contain any of the parameters defined in which.par.
Consider removing it from the item2test input or adding relevant parameters
to which.par”

How is that possible?

I am also struggling to understand the section where you are controlling for the latent variable. Do I need to use both “free_means” and “free_var” to control for laten variable? What does it really mean when I am controling for latent variable? Is that not the whole point of the DIF analysis, to check whether or not people with the same ability have the same probability of getting the item correct?

[Phil Chalmers]

On Mon, May 14, 2018 at 8:37 AM <jksha...@gmail.com> wrote:

I am currently attempting to run a DIF analysis using the MIRT package, however I am struggling to understand one particular section. I am not sure I completely understand the “which.par” variable.

In the example provided in the PDF file on MIRT, the parameters that are put in there are “a1” and “d1” which are referring to the slope and the intercept.

However, what does that exactly mean? Is “a1” referring to the discrimination of the items? Then what exactly does “d1” mean? That is not the difficulty of the item is it?

It depends entirely on the item types that you fitted. If you fit, for example, a 3PL model then your coefs would be a1, d, g, and u (fixed at 1); for a graded response model, you'd have a1, d1, d2, ..., dk-1. See coef(mod) to understand what you've fitted.

 

Concerning my own example, I am also getting an error when I just run “d1” as the only parameter in the “which.par”. The error is as follows:

“Error in checkForRemoteErrors(val) :
  4 nodes produced errors; first error: Item 1 does not contain any of the parameters defined in which.par.
                 Consider removing it from the item2test input or adding relevant parameters
                 to which.par”

How is that possible?

d1 is not a valid parameter for your IRT model selected. You probably want something else.

 

I am also struggling to understand the section where you are controlling for the latent variable. Do I need to use both “free_means” and “free_var” to control for laten variable? What does it really mean when I am controling for latent variable? Is that not the whole point of the DIF analysis, to check whether or not people with the same ability have the same probability of getting the item correct?

Freeing the latent mean/variance equates the groups so that the item parameters are on the same metric. Otherwise, observed differences in item response behaviour could be attributed to general group information differences rather than item bias (one would expect, for example, a 10th grad math class to have higher probabilities of answer questions correctly than a 9th grade class....but this doesn't mean the items are biased). 

 

--
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.