项目名称: 多重比较中控制FDR的有效检验方法
项目编号: No.11471204
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 数理科学和化学
项目作者: 赵海兵
作者单位: 上海财经大学
项目金额: 60万元
中文摘要: 多重假设检验是对多个假设同时进行显著性检验的统计推断,其在实际中有着广泛的应用。随着信息时代的到来和大数据的产生,需要同时检验的假设个数达到成千上万,并且数据间存在复杂的相关性。这些迫使统计学家要不断发展新的多重检验方法来解决新出现的问题和挑战,也促使了多重检验的不断发展。对检验P值加权是一有效而广受欢迎的提高检验功效的途径,本项目研究目标之一为得到最优加权P值检验。Fan,Han and Gu (2012)在统计量服从正态分布且有复杂相关性下估计出fixed cut-off检验法的FDR。本项目旨在估计出他们估计法中的关键统计量,从而得到控制FDR的检验法,并把他们的结果推广到渐近正态及开方分布等情形。当需要在异方差下同时检验多个复杂原假设时,本项目首先致力于构造合理有效的P值检验法;其次还给出数据相关时的最优检验法。
中文关键词: 多重比较;加权P值;错误发现率;相关性;复杂原假设
英文摘要: Multiple comparison focuses on testing multiple hypotheses simultaneously. It has been widely used in practice. With the information era's coming and big data's emergence, thousaands of hypotheses need to be tested simultaneously. This, together with the complex correlations, requries new testing methods to be developed to solve the emerging challenges, which prompts the development of nultiple comaprison. An effective and widely used technieque is weighting p-values to get more powerful test methods. One of our tasks is to get optimally weighted p-values procedure. Fan, Han and Gu (2012) proposed an estmate of FDR for fixed cut-off method. When the statistics are correlated and normally distributed. WE will discuss how to estiamte some key parameters of their estimate, get the FDR control method, and generalize their estimate to the case where the statistics are asymptotically normally distributed or chi-squre distributed. For multiple composite null hypotheses, we compose a reasonable and effective p-values procedure and develope the optimal testing procedure for independed data.
英文关键词: multiple comparison;weighted p values;FDR;correlation;composite null hypotheses