Bayesian Estimation Of The Logistic Testlet Response Model

With the economic development of the world,all the countries are paying more attention to persons of creative ability than ever and the developing education of learning promoted.People pay more and more attention to the system of examination,because the examination is everywhere in our social now.And the result after the exam is the focus of people's attention.The examination results analysis should take into account many aspects,the ability of the subjects and the state of the examination,the difficulty of the test,etc.So that the item response theory(IRT)model is one of the hot issues in order to make the estimated results practical.The testlet response theory model(TRT)with random effect parameters on the basis of the IRT model is more concerned by the researchers from various countries.In this paper,we will mainly demonstrate this point.This paper is divided into five chapters.The first chapter introduces the background of the IRT model and the TRT model.And the research method which used in this paper is Gibbs sampling.In the second chapter,we introduce the 2PLTRT model,the 3PLTRT model and the 4PLTRT model.In the third chapter,we introduce the 2PLTRT model and give the priori assumption.Then we use the Gibbs sampling method to expand the data of this TRT model.In the fourth chapter,firstly,we use the simulated data to estimate the 2PLTRT model and the 2PLIRT model respectively by the statistical software R.Then we compare their results.Secondly we use the real data to further evidence that the 2PLTRT model is more scientific and practical.Thirdly,the DIC of the model selection criterion is used to select the model and make the results more convincing.Finally,the 4PLTRT model and 4PLIRT model fit the real data.Sampling with the WinBUGS software.We observe whether the TRT model is still better than the IRT model.The paper summarizes the work done in the fifth chapter,and puts forward the further research and the next work.