MIRT补偿模型与非补偿模型的比较研究及其应用

发布时间：2019-02-19 17:33

【摘要】：本文通过使用BMIRT软件，设置不同的实验条件：被试样本量（1000和3000）×题目量（25和50）×能力相关（0.3和0.7），模拟生成多维三参数补偿数据和非补偿数据，并用多维三参数补偿模型和非补偿模型进行参数估计。通过比较项目参数和能力参数的RMSE值，，实现各种实验条件下的多维补偿模型与非补偿模型的参数返真性比较。结果发现，无论是估计多维补偿数据还是非补偿数据，三参数多维补偿模型的参数返真性都比三参数多维非补偿模型的参数返真性更好。尤其当估计多维补偿数据时，三参数多维补偿模型估计的能力参数RMSE值几乎是三参数多维非补偿模型的一半，显著优于三参数非补偿模型估计的能力参数返真性。本研究还将多维项目反应理论补偿模型与非补偿模型应用于瑞文高级推理测验中，发现多维补偿模型比多维非补偿模型拟合的更好。本研究使用多维项目反应理论补偿模型对高级瑞文推理测验进行深入分析，探究瑞文高级推理测验的各题目质量、难度及主要测量的认知成分，结果发现瑞文高级推理测验的整体区分度较好，并且项目难度几乎随着题序增大而增大。在五个能力维度上，瑞文高级推理测验试题的认知成分难度按A/S、CR、PP、D3和D2依次递增。最后，在多维补偿模型与非补偿模型对瑞文高级推理测验的被试能力参数估计的基础上，对被试在解决瑞文高级推理测验项目时能力间的相互作用进行了探索性分析，结果发现被试在解决瑞文高级推理测验项目时，CR、PP以及D3能力之间存在相互补偿关系，A/S与D2能力之间也存在相互补偿关系。最后，本文指出了该研究的不足，并对未来的研究提出展望。
[Abstract]:In this paper, by using BMIRT software, we set up different experimental conditions: sample size (1000 and 3000) 脳 subject quantity (25 and 50) 脳 ability correlation (0. 3 and 0. 7) to simulate the generation of multi dimensional three parameter compensation data and non compensation data. The multi-dimensional three parameter compensation model and the non-compensation model are used to estimate the parameters. By comparing the RMSE values of the project parameters and the capability parameters, the parameter fidelity comparison between the multi-dimensional compensation model and the non-compensation model under various experimental conditions is realized. The results show that the parametric fidelity of the three-parameter multi-dimensional compensation model is better than that of the three-parameter multi-dimensional non-compensation model, regardless of whether it is the estimation of the multi-dimensional compensation data or the non-compensated data. In particular, when estimating multidimensional compensation data, the capability parameter RMSE estimated by the three-parameter multi-dimensional compensation model is almost half of that of the three-parameter multi-dimensional non-compensation model, which is significantly better than the capability parameter fidelity of the three-parameter non-compensation model estimation. The multi-dimensional item response theory compensation model and the non-compensation model are also applied to the Raven advanced reasoning test. It is found that the multidimensional compensation model fits better than the multidimensional non-compensation model. In this study, the multidimensional item response theory compensation model was used to deeply analyze the advanced Raven reasoning test, and to explore the quality, difficulty and cognitive components of the Raven advanced reasoning test. The results show that the overall classification of Raven advanced reasoning test is good and the project difficulty increases with the increase of item order. In the five ability dimensions, the difficulty of cognitive components in the Raven Advanced reasoning Test was increased by A / S / C / PPD _ 3 and D _ 2 respectively. Finally, on the basis of multi-dimensional compensation model and non-compensation model to estimate the ability parameters of the Raven advanced reasoning test, the interaction between the ability of the participants in solving the Raven advanced reasoning test items is analyzed. The results show that there is a mutual compensation relationship between CR,PP and D3 ability and between A / S and D _ 2 ability in solving Raven advanced reasoning test items. Finally, this paper points out the deficiency of this research and puts forward the prospect of future research.
【学位授予单位】：江西师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：B841

【参考文献】