Juliet Shaffer
UC Berkeley

New approaches to error control in multiple testing

Until about 15 years ago, the most widely-adopted criterion for control of errors in multiple testing was setting a limit on the familywise error rate (FWER): the probability of one or more Type I errors (rejections of true hypotheses) in a family of tests. With the advent of well-structured problems for testing massive numbers of hypotheses (e.g. thousands in genomics, millions in astronomy), this criterion was widely considered too conservative. The first competitor to achieve widespread acceptance was control of the false discovery rate (FDR) with a test proposed by Benjamini and Hochberg in 1995. Since then, other tests controlling this rate have been considered, and many other criteria have been and are still being proposed. This talk will review and critique some of the recently-proposed criteria, and note the renewed relevance of some early work of Erich Lehmann on this subject.