項目反應理論應用於學業性向測驗之可行性探討

Appliying IRT Models to a College Academic Aptitude Battery

黃盟惠(Meng-Hui Huang) ； 張淑慧(Shu-Hui Chang)

《中原學報》 24卷2期 (1996/04) Pp. 5-12

https://doi.org/10.6358/JCYU.199604.0005

摘要

本研究的主要研究目的在探討項目反應理論應用於大學系列學業性向測驗之可行性，以作為下一階段電腦適性測驗研究之基礎評估，主要探討的問題包括大學系列學業性向測驗是否符合項目反應理論的基本假設、項目反應理論哪一個參數模式（一、二或三個參數模式）比較適合應用於大學系列學業性向測驗、以及探討項目反應理論與傳統測驗理論之間的關係。本研究測驗材料為「大學系列學業性向測驗 」，包括語言推理和數量比較兩個分測驗各60題，而受試者共1 046位成人。茲將主要的發現摘述如下：一經因素分析的結果發現：語文推理及數理比較分測驗皆符合項自反應理論單一向度的基本假設。二、適合度考驗的結果發現：語文推理分測驗適用二個參數模式，數量比較分測驗適用三個參數模式。語文推理分測驗適用二個參數模式的題目有51題，而數量比較分測驗適用三個參數模式的題目有46題。三、在傳統測驗理論難度（P值）與項目反應理論難度（b參數）的相關方面，語言推理、數量比較分測驗相關係數分別為-0.9413及-0.8543。在傳統測驗理論鑑別度（二系列相關係數）與項目反應理論鑑別度（a參數）的相關方面，語言推理、數量比較分測驗的相關係數分別為0.7439及0.5975。在傳統測驗理論個人得分與項目反應理論能力參數的相關方面，語文推理、數量比較分測驗的相關係數分別為09064 及09385。從項目反應理論各項參數與傳統測驗理論各項統計數間的相關係數來看，其問的相關係數都不錯，表示項目反應理論與傳統測驗理論對同一測驗資料的分析結果一致。因此，本研究結論是項目反應理論適用於分析大學系列學業性向測驗。基於項目反應理論的優點，未來可以項目反應理論為基礎，進一步發展更簡便、 更快速的電腦化適性測驗。

關鍵字

項目反應理論 ； 一參數模式 ； 二參數模式 ； 三參數模式

並列摘要

The purpose of the study is to investigate the feasibility of applying item response theory (IRT) to the college academic aptitude battery in order to assess the feasibility of computerized adaptive testing (CAT) in the follow-up study. The major problems to investigate include: whether the college academic aptitude battery meets the assumption of IRT, which IRT model (1PL, 2PL, or 3PL) fits the college academic aptitude battery, and the relationship between IRT and CTT (classical test theory).
The major findings are as follows:
1. The result of factor analysis shows that two subtests meet the unidimensional assumption of IRT.
2. The results of the goodness-of-fit showed that: two-parameter model can be applied to the verbal reasoning subtest and three-parameter model can be applied to the mathematical comparison subtest. Fifty-one items are retained for the verbal subtest; and forty-six items are retained for the mathematical subtest. 3. Regarding the correlation between the difficulty (P value) of CTT, and difficulty (b parameter) of IRT, the correlation coefficients are -0.94 13 for the verbal subtest and -0.8543 for the mathematical subtest respectively. Regarding the correlation between the discrimination (biserial correlation coefficient) of CTT and the discrimination (a parameter) of IRT, the correlation coefficient are 0.7439 for the verbal subtest and 0.5975 for the mathematical subtest respectively. Regarding the correlation between the total score of CTT and the ability parameters of IRT, the correlation coefficient are 0.9064 for the verbal subtest and 0.9385 for the mathematical subtest respectively.
According to the correlation coefficients between IRT parameters and CTT statistics, the correlation of IRT parameters and CTT statistics are moderately high, which indicates the compatibility of IRT and CTT in terms of applying to the same set of test data. Thus the conclusion of the study is that IRT is applicable to the college academic aptitude battery. Then based on the advantage of IRT, we can develop the computerized adaptive test (CAT) to achieve more efficient and accurate assessment of examiner’s trait level in the future.

並列關鍵字

Item Response Theory (IRT) ； one-parameter-logistic-model (1PL) ； two-parameter-logistic-model (2PL) ； three-parameter-logistic-model (3PL)