ranganath.s at excelindia.com
Mon Aug 30 21:16:16 EST 2010
I have read that in the case of Computer Adaptability Test,
there will be a huge item bank which are tagged with 'a', 'b' and 'c'
value. If a new untagged item is to be introduced into the item bank, the
item for calibration is administered as dummy question to certain number of
examinees and their responses are collected. I have few qns based on this
1. I have the understanding that for a static test(Qns are same
across examinees in a test which had adequate sample size, say 1000) that
the probability of occurrence of certain pattern of response(for a test) is
calculated by feeding the response vector of an examinee to the software
like BILOG which uses MMLE and EM to estimate the abc values. This method
takes into consideration the responses of other examinees to the same Qn as
well as performance of the examinees in other items. Please comment if I am
2. In the above mentioned case of CAT, If the responses of many
examinees to the interested items are only selected, we will be missing the
responses to other items, and also in CAT which is dynamic unlike the
previous one does not have a fixed set of questions across examinees. In
this case how is the estimation process done.
3. In the above case, If the estimation is done just by taking the
responses to the interested item by examinees and fed to software like
BILOG. This as well provides the solution for small sample size, where-in
responses to the common item across different forms/tests are combined
together to get the required sample size to estimate the abc value for that
4. Please recommend some books on Linking.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Rasch