Like Wine , the TTEST Procedure Improves with Age

PROC TTEST is an old reliable among SAS/STAT procedures, testing hypotheses about means of one and two normal samples. While this is a narrow class of analyses, it’s an important class, with key applications in all areas of science, including clinical trials. Recent updates to the TTEST procedure make it an even more powerful tool for the statistician. The enhancements include: • Equivalence tests, which answer not “Are the samples significantly different?” but rather “Are they significantly the same?” • AB/BA crossover studies, a common repeated measures design for clinical trials Perhaps the most useful new feature is the new plotting capability of PROC TTEST. It now produces univariate and comparative versions of histograms, box plots, and densities. This paper describes these new features of PROC TTEST in SAS 9.2 and illustrates their use in a number of examples. INTRODUCTION TTEST Nouveau The TTEST procedure is an old reliable in the SAS wine cellar, a bottle that satisfies even if it doesn’t surprise. In versions of SAS previous to 9.2, PROC TTEST performs t tests and computes confidence intervals for one sample, two samples, and paired samples of normal data. The TTEST procedure has matured in SAS 9.2, with graphics on the palate and new analytics in the nose. The most noticeable addition to PROC TTEST is the use of ODS Graphics to supplement the numerical results with visual displays. The graphs are automatically generated, including univariate and comparative versions of histograms, box plots, and densities. You can also customize them. An intriguing new note is the new SIDES= option for one-sided versions of tests and confidence intervals. This gives you higher statistical power when the directionality can be assumed in advance. The big, bold new TOST= option produces equivalence tests and intervals, popular in such situations as bioequivalence, where the typical goal is to establish whether the effects of two drugs are significantly similar rather than different. In this case, standard p-values from t tests are not appropriate for assessing similarity, since insignificant results may stem from low power rather than from the lack of a meaningful effect. Finally, in addition to one-sample, two-sample, and paired designs, PROC TTEST now gets new legs with the CROSSOVER= option. AB/BA crossover designs are a favorite in clinical trials and many other arenas. They reduce variability and increase the power of treatment comparisons by administering each of two treatments to each subject. Table 1 summarizes the major enhancements for PROC TTEST in SAS 9.2 and the required syntax. 1 Statistics and Data Analysis SUGI 31