个人不拟合对IRT项目参数估计的影响及净化对策

Influence of Person Misfit on IRT Item Parameter Estimation and Data Purification

目的:探讨个人不拟合对IRT二参数模型项目参数估计的影响,并使用数据净化方法降低这种影响,提高个人拟合指标探测率。方法:基于二参数模型和lz指标进行分析。使用ICC面积法比较项目参数估计的变化,并使用数据净化的方法提高lz指标探测效果。结果:①不拟合被试比率越大,项目参数估计偏差越大;②增加测验长度可以降低个人不拟合对项目参数估计的影响;③加大样本量对降低个人不拟合对项目参数估计的影响没有作用;④数据净化方法可以有效的提高lz指标的探测效果。结论:个人不拟合会影响二参数模型的项目参数估计,数据净化方法可以校准项目参数估计,提高lz指标探测效果。

Objective： Investigate the impact of person-misfit on item parameter estimation and the effect of data purification procedure. Methods： The two-parameter logistic model （2PL） and lz statistic were used in this study. ICC theory was used to evaluate the bias of item parameter estimation. And in this paper an approach to data purification was de- scribed and applied. Results： （~）The more misfit a person, had the larger bias in model estimation; （~）The bias of item pa- rameter estimation would reduce, when a test contained more items; （~）Sample size had limited influence on the ability of lz; （~）The power of lz statistic was increased by data purification. Conclusion： Item parameters estimated from data in- cluding aberrants are biased and the power of lz, as a consequence, is underestimated. It is shown that by a data purification procedure, the recovery of the power of the lz is achieved.