Data Fit Comparison of Mixture Item Response Theory Models and Traditional Models

Abstract

Thepurpose of this study is to determine the best IRT model [Rasch, 2PL, 3PL, 4PLand mixed IRT (2 and 3PL)] for the science and technology subtest of theTransition from Basic Education to Secondary Education (TEOG) exam, which is carriedout at national level, it is also aimed to predict the item parameters underthe best model. This study is a basic research as it contributes to theinformation production which is fundamental for test development theories. Thestudy group of the research is composed of 5000 students who were randomlyselected from students who participated in TEOG exam in 2015. The analyses werecarried out on 17 multiple choice items in TEOG science and technology subtest.When model fit indices were evaluated, the MixIRT model with two parameters and threelatent classes was found to fit the data best. According to this model, whenthe difficulties and discrimination averages of the items are taken intoaccount, it can be expressed that items are moderately difficult and discriminative for students in latentclass-1; the items are considerably easy and able to slightly distinguish the students in latent class-2; the items are difficult tothe students in the third latent class and they can slightly distinguish thestudents in this group.

Keywords

• item response theory
• latent class models
• mixed item response theory
• model-data fit
• item response theory
• latent class models
• mixed item response theory
• model-data fit