Özet:

The aim of this study is to examine the performance of item selection algorithms according to the accuracy of measuring through different number of attributes, item quality and test lengths for DINA and DINO models in cognitive diagnostic computerized adaptive testing (CD-CAT). Within the scope of the study, the number of attributes was manipulated as 5 and 8, and each item was limited to measure at least one attribute and at most 4 attributes. In data generation, the g and s parameters were drawn from the uniform distribution of U(0.05-0.25) for high item quality level and U (0.10-0.30) for low item quality level. Cognitive patterns of 3000 examinees were generated so that each examinee had a 50% chance of having each attribute. Fixed test lengths of 8, 16 and 24 were used as termination rules. GDI, JSD, MI, PWCDI and PWKL were used as item selection algorithms in the study. The performances of item selection algorithms were evaluated according to their attribute and pattern recovery rates. Data generation and analysis in the study were carried out using R 3.6.3 software. As a result of the study, it was determined that the measurement accuracy values of all algorithms increased as the item quality and test length increased, and the measurement accuracy decreased as the number of attributes increased. It was found that the measurement accuracy of the JSD algorithm was the highest in all conditions, while the PWKL algorithm was the lowest. While the performance of the algorithms except for PWKL algorithm in DINA and DINO models was approximately the same, it was found that the measurement accuracy of the PWKL algorithm in the DINO model was lower than that of the DINA model.

Anahtar Kelimeler: