项目名称: 不可忽略缺失机制下的广义矩方法和调整经验似然方法研究
项目编号: No.11501208
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 数理科学和化学
项目作者: 王磊
作者单位: 华东师范大学
项目金额: 18万元
中文摘要: 不可忽略缺失数据,作为近几年缺失数据领域里一个理论新颖且具有重要应用价值的问题,广泛存在于科学研究和生产生活的各个领域。由于不可忽略缺失数据的缺失机制与缺失数据本身有关,导致在大多数情况下参数是不可识别的,现有的统计方法无法满足理论研究和实际应用的需要。本项目致力于解决不可忽略缺失数据的难点和热点问题,主要包括:(1)借助工具变量解决指数倾斜模型参数不可识别的问题,建立具有双重稳健性和多重稳健性的估计方法;(2)利用重抽样的高阶调整经验方法建立参数的有效估计,基于空间深度统计量提出稳健的调整经验似然方法;(3)利用带惩罚项的变量选择方法和充分降维方法处理不可忽略缺失的高维数据问题,研究在参数维数固定以及维数随样本量增大时估计方法的理论性质;(4)在不可忽略缺失数据下研究快速变点检测和控制图的应用。四个部分既相互独立,又相互支撑,把理论研究与实际应用紧密联系。
中文关键词: 不可忽略缺失;可识别性;经验似然;变量选择;充分降维
英文摘要: Nonignorable missing data is now attracting increasing attention and becoming an important research topic in the missing data studies. However, identifiability and estimation of parameters with nonignorable missing response data are challenging problems in statistical theory and applications. In this project, we will first address the question about how to overcome the non-identifiability issue by utilizing a nonresponse instrument variable under the exponential tilting model. The double robust and multiply robust estimators will be provided by using generalized method of moments. Second, inference on parameters will be developed by employing resampling calibrated adjusted empirical likelihood and robust adjusted empirical likelihood, with or without auxiliary information. Third, in order to handle the high-dimensional data with nonignorable missing values, variable selection based on SCAD penalty and sufficient dimension reduction methods will be proposed. In addition, the asymptotic properties of the proposed estimators will be investigated, when the dimension is fixed or diverge as the sample size approaches infinity. Finally, we will establish the quickest change detection methods and control charts with regard to nonignorable missing data, and their properties will be studied.
英文关键词: Nonignorable Missing ;Identifiability;Empirical Likelihood;Variable Selection;Sufficient Dimension Reduction