where, yk and yl are means for two different treatments (k and l),
MSError is the mean square for error (obtained from ANOVA: single
factor), and nk and nl are number of observations in treatment k and
l. If a confidence interval is either completely positive or completely
negative, the pair means are significantly different. If the confi-
dence interval includes zero, no difference exists in that pair of
means. If the ANOVA indicates differences in pair means but Scheffe
post-hoc procedure does not pick out at least one pair being
significantly different, the pair that is most divergent should be
considered significantly different.