[Rasch] logistic regression vs UCON

Mike Linacre (RMT) rmt at rasch.org
Wed Feb 3 12:45:42 EST 2010


It sounds like PROX is good enough for your CAT application:

For N persons administered the pilot item, each with an ability estimated 
from items with known calibrations:

Pilot item difficulty = mean ability(N) - sqrt (1 +( [S.D.(N)]**2 / 2.9)) * 
ln ( count of successes  / count of failures)

See www.rasch.org/rmt/rmt83g.htm for the derivation

Update mean(N), SD(N) and counts after each administration of the pilot item.

Mike L.

At 2/2/2010, you wrote:
>Thanks for proposing using SAS and/or Winsteps, this means that I have to 
>clarify my question.
>I am trying to do "on the fly" item calibration in a CAT context where new 
>items with known locations are mixed in with pilot items, i.e., without 
>known locs. I am looking for a way to also target the selection of pilot 
>items dynamically during the regular CAT so as to speed up their 
>calibration and I wanted everything to be self contained. The estimation 
>of the pilot items' locs wouldn't have to be too precise I think to be 
>useful. But, being self-contained may not be worth the hassle, perhaps I 
>could run Winsteps/SAS dynamically as well ...

Mike Linacre
Editor, Rasch Measurement Transactions
rmt at rasch.org www.rasch.org/rmt/ Latest RMT:  23:3 Winter 2009

Please consider the environment before you print
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://mailinglist.acer.edu.au/pipermail/rasch/attachments/20100202/b859822b/attachment.html 

More information about the Rasch mailing list