1. Velleman, P. F., and Hoaglin, D. C., Applications, Basics, and Computing of Exploratory Data Analysis. Duxbury Press, Boston, MA, 1981.
2. Three different sets of values with an a priori mean of 0 and standard deviation of 1 were generated. The first set was multiplied with 0.6 (thereby reducing its a priori standard deviation by the same factor) and incremented by 3.78; the second set was just incremented by 4.05 and the third set by 4.25. The a posteriori means and standard deviations are listed in fig. 1. F-tests for comparing the variances resulted in following error probabilities: set 1 vs set 2: p=0.078; set 1 vs set 3: p=0.009; set 2 vs set 3: p=0.164.
3. Welch, B. L., Biometrika36 (1949) 293. This procedure for unequal variances converges to the one for equal variances when the standard deviations become equal. Therefore, it has been used for all non-graphical t-tests.
4. Andrews, H. P., Snee, R. S., and Sarner, M. H., Am. Statistn34 (1980) 195; Hochberg, Y., Weiss, G., Hart, S., J. Am. statist. Ass.77 (1982) 767; Godfrey, K., N. Engl. J. Med.313 (1985) 1450.
5. For ease of implementation and for clarity (context independence), only information from the considered individual data set should be included in the definition of an uncertainty measure. Therefore, unequal variances and sample-sizes, as well as comparison multiplicity were not taken into account for the definition of the significance limits. It is, however, important that the user of significance limits becomes aware of the imprecisions which may result from these simplifications.