Name: Kalyan Ghosh

Firstname: Kalyan

Title: Assoc. Director

Institution: Merck & Co.

Street: BL3-2, PO Box 4

City: West Point

Zip-Code: 19486

Country: USA

Phone: 484-344-7635

Fax: 484-344-2931

Email: ghoshk@merck.com

Authors: Kalyan Ghosh

Title: Multiple Comparison Procedures in Testing Qualitative Interaction

Abstract: Qualitative interaction is defined as the treatment-by-stratum interaction where the treatment differences among stratum levels are inconsistent in a way that in some stratum level the treatment difference is positive and in another level it is negative. Gail and Simon (Biometrics, 1985) have proposed a likelihood ratio test that is widely used for testing qualitative interaction. An alternative test proposed by Piantadosi and Gail (Statistics in Medicine, 1993), which was generalized by Pan and Wolfe (Statistics in Medicine, 1997), performs better than the Gail-Simon test when treatment-direction is different in only a few stratum levels. A closer look at this alternative test reveals that the qualitative interaction problem can be viewed as a multiplicity problem. Using this view, we propose some new tests that perform favorably in terms of power to detect a true qualitative interaction. These new test procedures also generalize the qualitative interaction test! s to non-normal and non-independent situations.