[Rasch] DIF studies for polytomous items

Kenji Yamazaki yk0271 at yahoo.co.jp
Tue Apr 20 02:34:18 EST 2010

Dr. Linacre,
  Thank you very much for your answer to my previous question.  I was wondering if I could ask you another question regarding the non-uniform DIF detection procedure in FACETS.  
  The data set I have contains 139 residents whose competencies were evaluated by their supervisors.  In the step 1 analysis for non-uniform DIF detection, I found that 23 of the 139 residents were assessed with the maximum performance rating by all the raters and for all the items.  Facets
 provides these residents with estimated measure values with an indicator, "Maximum."
  My question is whether or not I could include these maximum measure values as anchors in the 2.  Instead, should I let the program estimate parameters freely for these high achievers?
  Kenji Yamazaki
  "Mike Linacre (RMT)" <rmt at rasch.org> wrote:
  Thank you for your question, Kenji.

DIF analyses are complicated, and often difficult to interpret.

The items are "polytomous" (= many cuts), not "polychromous" (= many colors).

Anchoring is always done with Rasch-Andrich thresholds, because they produce a simpler expression of the Rasch model. Conversion between Rasch-Andrich and Rasch-Thurstone thresholds requires some algebra, http://www.rasch.org/rmt/rmt164e.htm

With the Facets software,

For uniform DIF, in your step 1: Models = ?B,..,?B produces the DIF estimates. Step 2 is not needed. One "?" is for the group-identifying facet.

For non-uniform DIF, perform your step 1. Output an Anchorfile=.
In your step 2, do not anchor the thresholds or the item difficulties, but do anchor the person measures in the two group analyses at their step 1 values. The person measures are in the step 1 Anchorfile=.


Mike L.

At 4/16/2010, you wrote:
Dear colleagues,
Thank you all very much for your supports for my previous question regarding DIF studies for polychromous items.  Today, I would like to ask another question that is partly related to the previous question that I asked.
The question is about which threshold values--Rasch-Andrich threshold or Rasch-Thurston threshold--I should use to investigate DIF of polytomous items.  
My understanding of a DIF analysis process is like this.  Step 1,  given that there are two groups&#65533;-reference and focal groups--for DIF investigation, I need to combine the data of these groups as one to estimate a set of category threshold values.  The step 2 will be that I use this set of
 threshold values as anchors to recalibrate item difficulties for the two groups separately (and so on$B!D(B).   
My question is regarding the step 2.  Which threshold values--Rasch-Andrich threshold or Rasch-Thurston threshold悠 should use?  I use FACETS program to conduct this investigation based on the rating scale model.  In my situation, I think I need to use the Rasch-Thurston threshold values, but it
 seems that I am not able to use them as anchor values.  
So, my question is about the following two issues:
(1)    Is it correct to use the Rasch-Thurston threshold values as anchors to make item difficulties comparable across two groups in a DIF investigation?
(2)    If the first question is true, does FACETS have a capability to make Rasch-Thurston threshold values as anchors?
Thank you very much.
Kenji Yamazaki
  Mike Linacre
Editor, Rasch Measurement Transactions
rmt at rasch.org www.rasch.org/rmt/ Latest RMT:  23:4 Spring 2010 


Get the new Internet Explorer 8 optimized for Yahoo! JAPAN
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://mailinglist.acer.edu.au/pipermail/rasch/attachments/20100420/02b11b1d/attachment.html 

More information about the Rasch mailing list