In
applying the adaptive testing equipped with the item response theory
to actual cases, it is crucial to know the appropriate number of
question items suffice for the specified accuracy.
In almost all the systems we have developed so far, we adopt small
number of items in a sequence of questions for adaptive testing because
too many questions will bore examinees or force to give up completing
the tests although estimation accuracy can be obtained. By experience,
we set the number of questions to be five. However, too small number
of questions will cause less accurate estimates for examinees' abilities.
In this paper, we show how the number of questions influences the
accuracy of the estimates.
For the sake of stable estimation, we made use of Bayes method, which
may make shrinkage estimation. In order to make sense to use the
estimates, we propose grouping, or classifying to examinees' abilities.
The optimal number of groups will be shown as a result.
By using simulation studies, we have found that five question items
are insufficient to estimate the examinee's true ability. At least
ten items are required even if the testing is adaptive. |
|
|
|
|
| number
of sufficient items, misclassification rate, confusion matrix,
item response theory, adaptive online testing. |
|
|