Detection of person misfit in computerized adaptive tests with polytomous items

One of the aims of a computerized adaptive test (CAT) is to construct an optimal test for each examinee. This is done by estimating the examinee’s ability-level (θ) after administration of each item and, based on this current ability estimate (�θ), the next item is selected using an item selection procedure. The θ-estimation procedure, the item selection procedure and the stopping rule of the CAT are all based on the assumption that the item scores of the examinee fit the assumed IRT model. It is questionable, however, whether the assumed IRT model gives a good description for each examinee’s test behavior. For those examinees for whom this is not the case, the current ability estimate as a measure of true θmay be inadequate, and as a result the construction of an optimal test may become difficult. There are all sorts of causes that may invalidate � θ; for example: knowledge of the correct answers due to test preview on achievement tests, faking on biodata questionnaires or personality tests, or randomly guessing on all items on the test in order to become familiar with the questions.