Statistics Colloquium: Dr. Michael Baron
Dep. of Mathematics and Statistics, American University
Title: Weighted statistics, minimax error spending, and other cost reduction techniques in multiple hypothesis testing
Abstract:
Bonferroni, Holm, and Holm-type step-up and step-down approaches have been well developed for the simultaneous testing of multiple hypotheses. Methods exist for controlling familywise error rates and false discovery rates. Performance of these tests can often be substantially improved by accounting for the relative difficulty of tests and introducing suitable weights that result in optimal error spending. In sequential and non-sequential experiments, such an extension of classical tests provides a smaller required sample size while still controlling the error rates. Some of the proposed tools can be combined for greater efficiency.