Özet:

The aim of this paper, which is comprised of two studies, is to examine different classification criteria, item selection methods and ability estimation methods used in computerized classification testing applications in terms of average classification accuracy (ACA), average test length (ATL), and measurement precision under the constraint conditions with and without content balancing and item exposure control. In the first study, 48 simulation conditions for 1000 examinees and item pools with 300, 600 and 900 items were created. In the second one, 16 conditions were created by making use of 822 examinees’ responses to a real paper and pencil test with eighty items. In the first study, it was found that there were similar and higher values in terms of ACA, while the expected a posteriori (EAP) estimator had a slight advantage over the sequential probability ratio test (SPRT) in terms of ATL. Further, the extension of the item pool increased the test efficiency. Upon an examination of content balancing and item exposure control, test efficiency was more adversely affected when one of the methods from SPRT, the cutscore based maximum Fisher information (MFI-CB), and EAP or the pool with 300 items was used. On the other hand, in the second study it was found that test efficiency for particularly confidence interval CI: 98% was higher when the ability estimator was the weighted likelihood estimation (WLE). Moreover, the common finding by two studies was that SPRT was more useful in maximazing the classification accuracy compared to confidence interval classification criterion.

Anahtar Kelimeler: