α系数与测验的同质性

Alpha Coefficient and Congeneric Test

本文从α系数与同质性测验、平行测验和基本τ-等价测验间的关系及三种测验间的关系入手,分析了α系数作为测验同质性信度估计的局限性;根据J reskog给出的信度定义(α系数),讨论了λ系数与α一致性信度、Guttman下限之间的关系,说明了在测验同质的前提下,λ系数在估计测验内部一致性时与α系数相比的优点。同时用模拟数据的方法就不同情景下测验的结构维度与α系数、Guttmanλ2下限和λ系数之间的关系进行了探讨。

The relationship of the Alpha coefficient with the congeneric test, the parallel test and the essential-τ-equivalent test, and the relationship among the three tests were reviewed, Firstly, the paper discussed the limitations of the Alpha coefficient when used to estimate consistency reliability. Second, the paper compared the advantage of the λ reliability coefficient introduced by joeresog（ 1971 ） to the Alpha coefficient and Guttman λ2 coefficient when used to estimate the consistency reliability when the test was congeneric. Third, the paper compared the advantage of the reliability λ′ coefficient to the Alpha λ′ coefficient and the Guttman λ2 coefficient when used to estimate the reliability when the test was muitidimensional, and the misuse and misinterpretation of the Alpha coefficient were discussed. Finally, a series of Monte Carlo Simulation studies were conducted to clarify the relationship of λ （or λ′）, the Alpha coefficient, Guttman λ2 coefficient and the structure of the test.