Determining sample size for a test of the equality of alpha coefficients when the number of part-tests is small