应用IRT模型分析TEM4听力理解项目的质量

Application of IRT to Item Analysis of TEM4 Listening Comprehension

本研究应用项目反应理论,从单维性假设验证、模型选择、项目筛选等方面分析2007年TEM4听力理解项目的质量。分析显示,该项目难度中等,区分度非常理想,听力理解水平较低的考生仅凭猜测答对的概率很小。本文建议命题人员从提高题目区分度、增强题目对听力理解项目的贡献值等方面来改善部分质量较差的题目;对于猜测概率超出可控范围的项目,不仅需要研究考生的答题过程,而且需要分析项目的出现顺序。本文的研究方法对国内外语测试的题目质量分析和题库建设具有借鉴意义。

This paper reports on a study of application of Item Response Theory to item analysis of 2007 TEM4 （Test for English Major, Band 4） listening comprehension from three perspectives, namely, the test of unidimensionality, the selection of models and items. Structural Equation Modeling with software EQS. 6 and Exploratory Factor Analysis with SYSTAT 11. 0 are employed to test the unidimensionality, and BILOG-MG 3.0 for Windows is used for helping the selection of models and items. The data totaling 65,535 are randomly selected from the test population. It is revealed that theses items, on the whole, boast a good slop parameter, a moderate threshold parameter, and a desirable asymptote parameter. It is suggested that poor items can be improved by bettering the slop parameter and the factor loadings, and the test taking process, together with the item orders, should also be explored to yield insights into the items with Undesirable asymptote parameter. Moreover, the methods applied in the present study provide references for item analysis and item bank establishment in foreign language testing.