Comparison of general diagnostic classification model for multiple-choice and dichotomous diagnostic classification model

UNCG Author/Contributor (non-UNCG co-authors, if there are any, appear on document)
Yanyan Fu (Creator)
Institution
The University of North Carolina at Greensboro (UNCG )
Web Site: http://library.uncg.edu/
Advisor
Robert Henson

Abstract: A submodel of the general diagnostic classification models for multiple choice (GDCM-MC), the excluding guessing from the correct answer (EGCA) model, was first introduced because the submodel with kernel extended reparameterized unified model (ERUM) can be compared directly to the dichotomous reduced reparameterized unified model (RRUM) without model induced bias. A simulation study was used to demonstrate this equivalence of the EGCA parameters of the correct options and the RRUM item parameters. At the same time, the simulation study was also used to demonstrate the equivalence of the two models when there were no skills or misconceptions measured by the incorrect options, and show the improvement of the EGCA estimation when distractors are created to provide additional information. The results confirmed the equivalence of the EGCA parameters of the correct options and the RRUM item parameters. The results also show that the correct classification rates (CCRs) and test-level cognitive diagnostic index (??DI•) were the same for the two models when there was no informative distractor. Additionally, by including weakly informative distractors, the EGCA showed higher CCRs and ??DI• than the RRUM. When the distractors were strongly informative, the EGCA had much higher CCRs and ??DI•. The studies also showed that CCRs and ??DI• increased when the sample size, test length, and item quality increased, as well as when the number of measured test skills and misconceptions decreased. A real-world example was used to compare the classification differences and predictability of the classification on the selection of the options between the two models in a distractor-driven assessment. The results show that the profile classification agreement was 48%, and the classification based on the EGCA was more correlated with the students’ selection of the correct or the misconception-embedded options than the classification based on the RRUM. The results indicate that the EGCA provides more realistic classification than the RRUM. The results of both simulation and the real data studies suggest that the polytomous diagnostic classification models (DCMs), rather than the dichotomous DCMs, should be used when the multiple-choice items have informative distractors.

Additional Information

Publication
Dissertation
Language: English
Date: 2018
Keywords
Dichotomous DCM, ERUM, GDCM-MC, Polytomous DCM, RRUM
Subjects
Educational tests and measurements $x Evaluation
Classification

Email this document to