Step-down FDR procedures for large numbers of hypotheses
FALSE DISCOVERY RATE; MULTIPLE; Computer Science, Theory & Methods
Somerville (2004b) developed FDR step-down procedures which were particularly appropriate for cases where the number of false hypotheses was small. The test statistics were assumed to have a niultivariate-t distribution with common correlation. MCV's (minimum critical values) were chosen so that 8 unique critical values resulted. Tables were given for numbers of hypotheses in, ranging from 50 to 10,000, for rho = 0., 0.5, and nu = 1.5, infinity. In this paper we extend the results, using MCV's resulting in 31 critical values. Tables are given for the same values of m, for rho = 0: 0.1, 0.5 and nu = 15, infinity. Interpolation rules are given for m, rho and nu. Use of larger numbers of critical values increase both the power and the number of hypotheses falsely rejected. When the expected number of false hypotheses is small, use of the procedures of this paper results in a reduced number of false rejections with a negligible reduction in power.
Applied Parallel Computing: State of the Art in Scientific Computing
"Step-down FDR procedures for large numbers of hypotheses" (2006). Faculty Bibliography 2000s. 6608.