THE COMPARISON OF FOUR DIFFERENT COEFFICIENT ALPHAS FROM A PSYCHOMETRIC POINT OF VIEW


Yurdugul H.

HACETTEPE UNIVERSITESI EGITIM FAKULTESI DERGISI-HACETTEPE UNIVERSITY JOURNAL OF EDUCATION, sa.36, ss.327-339, 2009 (SSCI) identifier identifier

Özet

In this study, the properties of four different coefficient alphas in reliability were compared. Those were a) unstandardized alpha (alpha(VK)) based on variance-covariance matrix, b) standardized alpha (alpha(VK)) based on correlation matrix, c) maximum alpha (alpha(Max)) based on first eigenvalues by Armor (1974) and d) alpha(M) defined by McDonald (1985) and based on confirmatory factor analysis The comparisons made to determine the best coefficient conducted on Monte-Carlo simulations and received priority consideration were carried out in three stages: a) investigation of robust estimator properties according to root mean square error (RMSE) for four coefficient alphas, b) comparison of coefficient alphas' values in unidimensional data sets and multidimensional data sets, and c) comparison of coefficient alphas' behaviors in parallel, tau-equivalent, and congeneric data sets. As a result, the coefficient alpha(Max) is found to be the optimum coefficient among those three stages. The coefficient alpha(M) is found to be non-optimum in all the stages.