Değişen madde fonksiyonunun farklı koşullar altında dına ve dıno modellerinin sınıflandırma performansına etkisi

SEYHAN SARITAŞ AKYOL

Değişen madde fonksiyonunun farklı koşullar altında dına ve dıno modellerinin sınıflandırma performansına etkisi

Thesis Type: Doctorate

Institution Of The Thesis: Gazi University, Eğitim Bilimleri Enstitüsü, Eğitim Bilimleri Enstitüsü, Turkey

Approval Date: 2023

Thesis Language: Turkish

Student: SEYHAN SARITAŞ AKYOL

Supervisor: Mehtap Çakan

Open Archive Collection: AVESIS Open Access Collection

Abstract:

Since cognitive diagnostic models provide detailed feedback to students and practitioners, their use in the field of education is rapidly spreading. However, an important step to ensure appropriate use and interpretation from cognitive diagnostic models is to understand the impact of differential item functioning (DIF). In this study, within the scope of cognitive diagnosis models, it is aimed to examine the effects of model (DINA, DINO), number of attribute (4, 6), DIF magnitude (.075, .10, .15, .20), DIF type (uniform, non-uniform), the percentage of items with DIF (20%, 30%) and the number of attributes measured by items with DIF in the Q-matrix (1, 2, 3) on classification performance at the level of attribute and pattern. A Monte Carlo simulation study based on hypothetical cognitive diagnostic evaluation was conducted to create these test conditions that cannot be created in practice. In line with the aim of the study, 192 different data sets were created according to the factors and the levels of the factors determined based on the literature. In order to compare the conditions determined in the research in a more informative way, in addition to these conditions, data set was created for each model and number of attributes aforehand, for the absence of the item with DIF. Thereby, a total of 196 different conditions were examined. In the study, as the fixed factors for each group, the sample size of 1000, the correlation between the attributes p=0.5, the number of items in the test 30, and the distribution of slipping and guessing item parameters as a uniform distribution [.1, .3]. In the study, the factors were examined for their effects on the attribute and pattern level correct classification rate for the reference and focus group in the DINA and DINO models. According to the analysis, as the number of attributes increased from 4 to 6, the classification accuracy of the DINA and DINO models of both groups decreased. As the DIF magnitude increased, the focus group attribute level correct classification accuracy systematically decreased in the DINA model, while no systematic difference was observed in the DINO model. In both models, when the DIF type changed from uniform to non-uniform, in accordance with the expectation, classification accuracy decreased in the focus group and increased in the reference group. As the effects were examined according to the percentage of DIF, the classification accuracy of DINA and DINO models decreased when the percentage of items with DMF increased in the focus group. It was found that the item attribute relationship in the q matrix of the item with DIF had different effects on the classification accuracy at the levels of DINA and DINO models. As the number of attributes measured by the item with DMF increases, the classification accuracy of the attribute and pattern level increases in the DINA model, while it tends to decrease in the DINO model under the same conditions. As the number of attributes measured by the item with DIF increases, ACA and PCA tend to increase in the DINA model, while they tend to decrease in the DINO model under the same conditions. In addition, in general, the classification accuracies of the DINO model at the levels of all the factors discussed in the study were found to be higher than the DINA model. In addition, according to the findings, it was seen that the attribute level classification accuracy in different DIF conditions was robust in all conditions, and the pattern level classification accuracy was more negatively affected by the DIF effects.