Misuses of KR-20 and cronbach's alpha reliability coefficients

Abstract

In this study, misuses of KR-20 and Cronbach's alpha reliability coefficients, used as internal consistency estimates, are illustrated. Two real data sets were used, dichotomously scored, KR-20, data set and polytomously scored, alpha, data set, to see variations in internal consistency coefficients when the unidimensionality assumption is violated. It is shown that a very high, but misleading internal consistency coefficient, KR-20 or alpha, can be obtained even when the unidimensionality assumption is violated. It is also shown that using the standard error of measurement as an indicator of reliability results in similar problems. Finally, misuses and the ways of avoiding those misuses of reporting internal consistency of a scale are suggested and briefly illustrated.