SP程序和DFTD策略应用于IRT取向下DIF检测方法的效应比较

发布时间：2018-05-09 17:12

本文选题：项目反映理论 + 项目功能差异　；参考：《江西师范大学》2014年硕士论文

【摘要】：本研究尝试对IRT取向下的三种方法：SIBTEST、IRT-LR和DFIT，设置三种模式：标准程序下（Standard模式，简称ST），加入Scale Purification程序的检测模式（简称SP模式）和加入DIF-free-then-DIF策略的检测模式（简称pure anchor，简称PA,），进而形成九种检测程序（SIB-ST，SIB-SP，SIB-PA，IRT-LR-ST,IRT-LR-SP、IRT-LR-PA、DFIT-ST，DFIT-SP，和DFIT-PA），在等级反应模式下以模拟实验方式，探讨三种模式和九种检测程序的检测效果比较。研究设计采用四个自变量（样本容量，DIF形态，DIF百分比以及DIF强度），因变量两个（I型错误率和统计检验力）。研究主要结论摘要如下：一、在不同样本容量下，九种程序的统计检验力都是是随着样本容量增大而逐步提高的，平均统计检验力和平均I型错误率亦如此。SP和PA检测模式的统计检验力分布与ST检测模式的分布基本相似，但I型错误率控制为较低。二、对于不同强度DIF检测，除了非一致性DIF题，一致性和混合型DIF的检测方面，各种程序对于强度为中度（0.6）的DIF题目检测效果都优于两种轻度DIF题目的。三、对于不同DIF比例（10%，20%，30%），9种程序的统计检验力和I型错误率随着DIF比例增加而提高。四，整体统计检验力而言，IRT LR法三种检测模式的DIF检测效果相对于其他方法较佳。DFIT次之，SIBTEST随后。五、不同检测模式而言，在低DIF比例和小样本时，ST模式统计检验力较好，而在高DIF比例和大样本时，，SP模式和PA模式表现较为接近，比ST模式要更好一些。SP和PA检测模式对控制I型错误率有积极作用。
[Abstract]:In this study, we try to set up three modes: standard program, standard program, IRT-LR and DFIT. for three methods:: SIBTESTT IRT-LR and DFIT. In short, the detection mode of joining Scale Purification program (SP mode) and the detection mode of adding DIF-free-then-DIF strategy (pure anchorm), and then forming nine detection programs SIB-STN SIB-SPN IRT-LR-STN IRT-LR-SPN IRT-LR-PADFIT-STDFIT-SPP, and DFIT-PAPX, and DFIT-PACU, in the hierarchical response mode. This paper discusses the comparison of the detection effects between the three modes and the nine detection programs. The design was designed with four independent variables (sample size, DIF form, DIF percentage and DIF strength), two dependent variables, type I error rate and statistical test power. The main findings of the study are summarized as follows: First, under different sample sizes, the statistical test power of the nine programs increases gradually with the increase of sample size. The distribution of statistical test power of the model of SP and PA is similar to that of the model of St detection, but the control of type I error rate is lower. Secondly, for DIF detection with different intensities, in addition to non-consistency DIF problem, consistency and mixed DIF detection, all kinds of programs are superior to two mild DIF problems for DIF subject detection with moderate strength of 0.6). Third, the statistical test power and type I error rate of 9 programs for different DIF ratios increase with the increase of DIF ratio. 4. The overall statistical test power of IRT / LR method was better than that of other methods in DIF detection, followed by SIBTEST. Fifthly, the statistical test power of St model is better in low DIF ratio and small sample, while in high DIF ratio and large sample, SP model and PA model are close to each other. Better than St mode. Sp and PA detection mode have positive effect on controlling type I error rate.
【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】