Acceso usuarios
Ying Lu, Ronald K. Hambleton
New test designs such as computer-adaptive testing have many advantages for candidates including considerable flexibility for scheduling their tests. But this increase in flexibility for candidates results in candidates seeing test items from the available bank of test items on a continuing basis. The risk is present that candidates may share their knowledge of the items they have seen with others, and thus, impact negatively on the validity of tests. The purposes of this papel are to propase two statistics for detecting disclosed items, to investigate their sampling distributions, and to determine their effectiveness under some cornmon testing conditions. Both statistics investigated (one based on a residual analysis, and the other based on an analysis of the likelihood function with and without item disclosure) appear to hold promise, and both appear worthy of considerably more research. Several suggestions for follow-up research are described in the paper.
