[Rasch] Item Re-calibration
ranganath.s at excelindia.com
Tue Jun 22 22:24:57 EST 2010
I have a question in item re-calibration. I was reading a white
paper (http://www.ncme.org/pubs/items/17.pdf) IRT Equating Methods. While
explaining the need for re-calibration he writes(Page 3, Heading - Step
two:placing parameter estimates on a common scale) "Suppose the same set of
items is given to two groups of examinees, and the item parameters are
estimated twice, once in each group. Suppose, that the model of choice too
is 2PL model. Because the item characteristic curves are supposedly
independent of the groups used to derive them, the expectation would be that
the two sets of item parameters would be identical except for sampling
error. However this is not so. For the 2PL model stocking and Lord pointed
out that the expression for item characteristic curve is a function of
ai(theta1 - bi), where ai, theta1 and bi are the item discrimination,
ability and item difficulty parameters of the model"
I don't see an explanation for variation in item parameters
across the tests. In all the materials on IRT it is always said that test
parameters are non-variant with groups. I am thoroughly confused. Please
help. If there is a list of scenarios in which it changes, please list.
Please suggest me some document on this which gives insight into this.
Thanks in advance.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Rasch