A Step-down FDR Procedure for Large Numbers of Hypotheses

Stepdown FDR procedures are developed for testing very large numbers of hypotheses when the expected number of false hypotheses is small. Tables of critical values are obtained for several values of m (number of hypotheses) ranging from 30 to 10,000. Critical values for m not in the range may be obtained by log-linear interpolation. Multivariate normality of the test statistics and a common value of the their correlation coefficients of 0 or.5 are assumed for the tables. The procedures can be used to control not only the expected false discovery rate, but also the actual number of false discoveries and proportion of false discoveries. Use of the tables is illustrated using a microarray dataset reported by Korn, Troendle et al (2003). 1