Meijer, R., University of Twente, The Netherlands
Recent attempts to introduce computerized adaptive testing (CAT) into practice have revealed a host of unexpected practical problems for which solutions are badly needed. For example, selection of test items from a pool only according to statistical principles may quickly lead to unbalance in test content between candidates and hence to threats to its validity. Also, examinees may produce item score patterns that are unexpected given the test model that underlies selection of items. As a result an examinee's proficiency level may be determined inadequately. Further, for high-stakes tests the danger of item compromise is present, and measures to prevent items from overexposure are required. Several of these problems are reviewed and recently developed solutions to some of them are introduced and evaluated.