Admissibility in partial conjunction testing
主 题: Admissibility in partial conjunction testing
报告人: Jingshu Wang (Stanford University)
时 间: 2015-09-16 10：30-11：30
地 点: 理科一号楼 1114
The partial conjunction hypothesis is to pool n p-values to test whether more than r of n hypotheses are non-null with r = 1 corresponding to a usual meta-analysis. It has many applications including replicability studies and cluster-wise analysis, for which partial conjunction is combined with multiple comparisons. Benjamini and Heller (2008) provided a valid combined p-value for the partial conjunction null by ignoring the r ? 1 smallest p-values and applying a valid meta-analysis p-value to the remaining n ? r + 1 p-values. We analysis a general form of their p-value (GBHPC p-value), and provide sufficient conditions for the admissibility of it among monotone combined p-values. We also show that admissible monotone combined p-values are necessarily GBHPC p-values. If one does not require monotonicity then GBHPC p-values is no longer admissible, but the dominating tests are too unreasonable to be used in practice. Bio of the speaker: Jingshu Wang is a Ph.D. student in Statistics at Stanford University. Before coming to Stanford, she received a B.S. degree in Mathematics and Applied Mathematics in 2011 from Peking University. Her research is focused on high dimensional factor models, multiple hypotheses testing and meta-analysis. She also has a broad interest in various other problems, including applications of statistical models to genetics, machine learning algorithms and parallel computing.
