Multi-stage Adaptive testing
Computer adaptive testing works with a calibrated item bank. Most of us will not be able to afford such luxury. Creating a large calibrated item bank costs a lot of time and money. It is difficult to collect responses under the right (high-stakes) circumstances, and you must trust the item parameters not to change over time or with item sequence (which of course they do). And we have not even mentioned the most troublesome problem in the context of summative assessment: item exposure during pre-calibration and actual testing. For all these reasons it is much more sensible and cost-efficient to do multi-stage testing. The modules comprising the multi-stage test can be constructed based on a limited pre-test of the items and calibration can be done after the test has been administered. All the advantages of adaptivity without item waste or exposure! We have developed the theory of (CML and Bayesian) calibration of multi-stage testing designs as well as an R package called dexterMST that works like a charm.
Wish to put this in practice? Contact us at firstname.lastname@example.org