Independent random samples were generated from normal distributions having
zero means, under various combinations of population standard deviations and
sample sizes, for k=3 and 5. In each case, empirical test sizes of the original as
well as the modified B-F test were determined after 10,000 iterations, for nominal
levels of 1% and 5%. Results are displayed in Table I. Empirical test sizes are
enclosed in parentheses if they are more than three standard errors greater than the
nominal level.