Generalized Augmentation to Control the False Discovery Exceedance in Multiple Testing

A new multiple testing procedure, the generalized augmentation procedure (GAUGE), is introduced. The procedure is shown to control the false discovery exceedance and to be competitive in terms of power. It is also shown how to apply the idea of GAUGE to achieve control of other error measures. Extensions to dependence are discussed, together with a modification valid under arbitrary dependence. We present an application to an original study on prostate cancer and on a benchmark data set on colon cancer. Copyright (c) 2009 Board of the Foundation of the Scandinavian Journal of Statistics.