Bayesian Ve Nonbayesian Kesti̇ri̇m Yöntemleri̇ne Dayali Olarak Siniflama İndeksleri̇ni̇n Timss-2015 Matemati̇k Testi̇ Üzeri̇nde İncelenmesi̇

Abstract

Purpose of this study is to compare the classification accuracy and consistency indices at different sample sizes in terms of Bayesian estimation methods with MAP, EAP and Nonbayesian estimation methods with MLE, WLE in the framework of IRT. In this direction, ability estimations based on MLE, WLE, MAP and EAP were obtained for each sample size. Then, for each condition of sample size, classification accuracy and consistency indices were calculated by using the Rudner' s appoach. According to the findings of study, it is seen that classification indices based on Nonbayesian methods are more accurate and consistent than the indices obtained based on Bayesian methods. Among Nonbayesian methods, it is concluded that MLE leads the more accurate and consistent classification indices than WLE. However, when the post hoc tests and effect sizes are investigated, it is seen that all pairs that results in significant difference have small effect in practice.