Resampling Tests in the Presence of Nuisance Parameters

• Activity Number: 192
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Ekkehard Glimm*
• Companies: Novartis
• Keywords: resampling tests ; confidence distribution ; bootstrap ; clinical trials

Abstract

Resampling and boostrap tests (Westfall and Young, 1993) are very useful tools for testing global null hypotheses about treatment effects in clinicals trials when trialists want to avoid making parametric distributional assumptions. However, their properties can become difficult to assess when multiple hypotheses are considered, or when additional covariates (other than treatment indicator) influence the response.

In such situations, one can attempt to (i) perform a permutation test on residuals (Korn and Freidlin, Biometrics, 2008) or generalize the permutation concept to (ii) a rotation test (Solari et al., Biometrics, 2014) or (iii) even further generalize to devise a confidence distribution for a suitable test statistic in the sense of Xie and Singh (International Statistical Review, 2013).

This talks explores these possibilities by applying corresponding tests to the comparison of responses to treatment in a clinical trial. The discussion will focus on the necessary properties of appropriate test statistics and on whether results hold exactly or only asymptotically. Simulations will be presented to compare the investigated methods.