[Rasch] OPLM - imputable discrimination

Stephen.Humphry at det.wa.edu.au Stephen.Humphry at det.wa.edu.au
Thu Jul 6 18:31:15 EST 2006


Indices such as Outfit and Infit are sensitive to differential levels of
discrimination, although they don't give invariant information about the
level of discrimination. Inspection of ICCs is important for assessing
differential levels of discrimination for various reasons, including
sensitivity to sample size. If available item discrimination estimates were
genuinely measurements of relative levels of discrimination then they would
necessarily indicate misfit from the Rasch model.

 

Steve

 

-----Original Message-----
From: rasch-bounces at acer.edu.au [mailto:rasch-bounces at acer.edu.au] On Behalf
Of Anthony James
Sent: Thursday, 6 July 2006 3:32 PM
To: rasch at acer.edu.au
Subject: RE: [Rasch] OPLM - imputable discrimination

 

Hi All,

Are Rasch fit statistics sensitive to items which have descrimination
indices that deviate too much from one?

I mean do such items appear as misfitting items in Rasch analysis?

If this is not the case, then apart from checking fit statistics, item
discriminations should also be checked and items with deviating
discriminations should be deleted, to make sure that the Rasch model holds. 

Anthony

"Mike Linacre (RMT)" <rmt at rasch.org> wrote:

Colleagues:

Has anyone found that imputing item discrimination (so giving the
'flexibility' of the 2PL model) is useful?  Winsteps already reports an item
discrimination index. It would be straightforward to make item
discrimination imputable. Of course, if you estimate => impute => estimate
=> impute => .... then the process diverges.

MIke Linacre

At 7/5/2006, you wrote:



Rama, the OPLM is formally the same as the 2PL model. However, item
discrimination in the OPLM is treated as an index rather than a parameter.
The index is imputed rather than being estimated. So the 'one' in OPLM
implies there is still only one item parameter. The weighted score in the
2PL model is sufficient for the person parameter but the problem is that the
values of discrimination parameters are unknown, meaning the weighted raw
score "is not a mere statistic, and hence it is impossible to use CML as an
estimation method" (Verhelst & Glas, 1995, p. 217). The approach in the OPLM
is to impute the values rather than estimating them, though estimates are
still used as starting point. The index is restricted to integer values
between 1 and about 15. The OPLM is intended to retain the properties of the
Rasch model while giving the 'flexibility' of the 2PL model (stated at the
beginning of the manual).

Mike Linacre
Editor, Rasch Measurement Transactions
rmt at rasch.org www.rasch.org/rmt/ <http://www.rasch.org/rmt/>  Latest RMT:
19:4 Spring


_______________________________________________
Rasch mailing list
Rasch at acer.edu.au
http://mailinglist.acer.edu.au/mailman/listinfo/rasch

 

  

  _____  

Do you Yahoo!?
Get on board. You're
<http://us.rd.yahoo.com/evt=40791/*http:/advision.webevents.yahoo.com/handra
isers>  invited to try the new Yahoo! Mail Beta.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://mailinglist.acer.edu.au/pipermail/rasch/attachments/20060706/c5c92fc6/attachment.html 


More information about the Rasch mailing list