For a one-sided (SIDES=L or SIDES=U) or equivalence (TOST) test based on the normal ratio (TEST=RATIO DIST=NORMAL) with a paired design, two-sample design, or a crossover design with the IGNOREPERIOD option, if the "group 2" sample mean is negative, then the p-values reported are incorrect. The p-values are actually equal to one minus the displayed p-values. For the equivalence test, the Upper and Lower p-values should be adjusted in this manner first, then the Overall p-value is computed as the larger of these two adjusted values. "Group 2" here is the second member of a pair for paired designs, the second group for two-sample designs, and the second treatment for crossover designs with the IGNOREPERIOD option.

If all of the above criteria are met, you can work around this issue by negating the group 2 variable in data step code and rerunning the analysis.

Operating System and Release Information