题组测验中处理局部项目依赖性(LID)的模型发展

Model Development in Processing Local Item Dependency(LID) in Testlet-composed Test Analysis

题组是常用的测验题目组织形式,题组中的题目往往会违背局部独立性假设,产生局部项目依赖(Local Item dependence,LID),探讨如何处理题组中LID以提高参数估计精度具有重要意义。本文在评述题组测验中处理LID的标准IRT方法、多级评分IRT方法、题组随机效应方法和两因子方法的基础上,以参数估计精度作为准则,对这4种处理LID方法的效果和适用条件进行了归纳并作出相应的评析,为使用者选择合适的方法处理LID提供了参考依据。同时展望了今后的研究方向。

Abstract： Testlets-composed tests are commonly used in education and psychology fields, but items in the same testlet may violate the assumption of local item independence and lead to local item dependency （LID） Thus, in order to increase parameter estimation accuracy, probing into methods of handling LID is of significant importance. In this paper, the authors summarize and assess four methods in processing LID, including standard IRT model, polytomous IRT model, testlet response model and bi-factor model. Taking parameter estimation accuracy as the criterion, the authors further summarize the effects and suitable conditions of each method and provided corresponding comments. This paper will provide reference for method selection when processing LID. Future research directions are discussed as well. Key words： testlet; IRT model; local item dependency （LID）