Relationships among IRT item discrimination and item fit indices in criterion-referenced language testing

Abstract

This study investigates relationships among the IRT one-parameter fit statistics, the two-parameter slope parameter and traditional biserial correla tions in terms of the role these indices play in criterion-referenced language test construction. It discusses the assumptions of the two models and how these assumptions can affect criterion-referenced test construction and interpreta tion. The study then specifically examines how the indices interrelate as indices of item discrimination. Examinees in Mexico, Saudi Arabia and Japan were administered one of two forms of a functional test (Form A n = 430, k = 94: Form B n = 400, k = 95). The data were analysed using the two IRT models and the results were compared. The results indicate strong relationships among biserial correlation, two-parameter slope, and one-parameter infit and outfit. These results indicate the need to employ the two-parameter model when con ditions allow, and to take item discrimination and item difficulty indices into account when conditions do not. Further implications for interpreting the strong relationships between the indices are discussed.