The Performance of Five Reliability Estimates in Multidimensional Test Situations

This paper investigates the estimation biases of five reliability indices, Cronbach’s α, Guttman’s λ2 and λ4, glb, and McDonald’s ω. The factors included are test dimensionality, population ability distribution, sample size, test length, and test discrimination. It was found that estimation biases of Alpha, λ4, and glb were correlated with the test’s true reliability, whereas λ2 and ω were not. Estimation biases were larger in two-dimensional tests than in unidimensional tests. Alpha overall had the largest estimation bias, but glb displayed similar bias in unidimensional tests. In light of the findings in the simulation study, we recommend McDonald’s ω because it had the smallest estimation bias in most test condition.