著者

TSUCHIHIRA Taiko

出版者

日本言語テスト学会

雑誌

日本言語テスト学会誌

巻号頁・発行日

vol.17, pp.59-80, 2014

---

The purpose of the present study is to examine the appropriateness of item response theory (IRT) to language testing. Although it has been discussed by many researchers (e.g., Henning, 1992; Blais & Laurier, 1995) for a long time, its appropriateness has not been demonstrated using language testing data, especially listening test data. Therefore, the dimensionality of the listening test data was examined by several approaches in the present study. The 50 questions in the listening section of the TOEFL Sample Test (6th edition) were administered to 392 students as a part of their usual General English classes. The data were analyzed using two approaches: the factor-analytic approach and the principle component analytic approach. The analyses of the 30 questions in Part A showed less possibility of the existence of a second dimension. As for the analyses of the 20 questions in Part B and C, however, two out of the five analyses affirmed the possibility of a second dimension. These mixed results suggested that multidimensionality may be detected depending on the methods used. Moreover, it was found that the different tasks tended to measure different dimensions, even though they seem to measure the same language skill. In addition, there was a fairly large amount of unexplained variance in the data. It is possible to say that there was a great deal of noise in the data that could not be aligned along dimensions. An implication is that tests using IRT should be more construct-valid. Lastly, it is observed that unidimensionality is a continuum rather than one position of a binary variable. Its appearance depends on the methods used to seek it.