irt模型參數估計的mcmc算法研究

1、南京師范大學碩士學位論文IRT 模型參數估計的MCMC 算法研究姓名：甘媛源申請學位級別：碩士專業(yè)：心理學;應用心理學指導教師：余嘉元2009-05-09南京師范大學碩士畢業(yè)論文 Abstract - ii - Abstract Simulation experiments on parameter estimation of item res

2、ponse theory (IRT) model were carried out with markov chain monte carlo (MCMC) algorithm in this dissertation. Firstly, parameter estimation veracity of 2-parameter logistic (2PL) model and 3-parameter logistic (3PL) mod

3、el were investigated with MCMC algorithm. Then the veracity of parameter estimation with MCMC algorithm was compared with EM algorithm, and the veracity of the parameter estimation with MCMC algorithm in 2PL model was co

4、mpared with 3PL model in the same sample size. Thirdly, the reason why the veracity of parameter estimation of 3PL model was worse was investigated, and the improvement of 3PL model was tried out. Finally, the parameter

5、estimation of 4-parameter logistic (4PL) model with MCMC algorithm was investigated. The details are as follows: (1) The MCMC algorithm was effective to estimate parameters of 2PL model, even the samples was limited. The

6、 veracity of MCMC algorithm would be increased with the increase of the items and examinees, which indicated that the algorithm proposed was valuable. (2) The veracity of parameter estimation with MCMC algorithm was bett

7、er than that with EM algorithm. The MCMC algorithm was also effective to estimate parameters of 3PL model, even if the veracity of parameter estimation of 3PL model was not as high as that of 2PL model. The key reason of

8、 the worse veracity of parameter estimation of 3PL model was the poor identifiability of its item parameters. (3) Simulation experiments on parameter estimation of the improved-3PL model were carried out with MCMC algor

9、ithm. The Improved-3PL model could increase the distinguishability of item parameters, and then the veracity of its item parameters estimation was better than standard 3PL model. (4) The MCMC algorithm could estimate par