Investigating the Behavior of Reliability Coefficients for Different Test Score Distributions

• In Event: Reliability and Subscores

Abstract

Some recent literature criticized the use of coefficient alpha to estimate test reliability or test internal consistency (see Psychometrika, 74(1)). They criticize alpha as a lower bound to reliability and not a very good one unless stringent assumptions are satisfied; they also assert that the actual test attribute measured by alpha, often labeled internal consistency, unidimensionality, or item homogeneity is either poorly defined or poorly measured by alpha. The purpose of this study is to use actual and simulated educational test data to illustrate the robustness and usefulness of alpha, thereby supporting the use of alpha as a valid estimate of test reliability. Especially, the impact of different test score distributions on alpha’s sampling distribution is further investigated.

Authors

• Yi-Fang Wu, University of Iowa

• David J. Woodruff, ACT, Inc.