项目名称: 带有随机效应的广义空间自回归模型的统计推断
项目编号: No.11301463
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 数理科学和化学
项目作者: 喻达磊
作者单位: 云南财经大学
项目金额: 23万元
中文摘要: 正态假定下的空间自回归模型无法分析非正态型、离散型或者具有偏态分布,且拥有复杂相关结构的空间数据。依托惩罚准对数似然函数,本项目拟将空间权重矩阵引入广义线性混合模型的表达式,从而构造一种带有随机效应的广义空间自回归模型,以研究上述拥有复杂结构的空间数据。因引入了空间权重矩阵,这类模型可充分利用已存在于地理空间中的先验信息,提高统计推断之效率。针对此模型中固定/随机效应以及自回归参数之估计理论(包括估计方法、大样本性质和渐近分布)的研究也将展开。基于这些结果,将推广条件赤池信息量的定义到该模型,并给出其渐近无偏估计即条件赤池信息准则,同时研究该准则用于模型选择时的表现。最终,将依托稳健化惩罚准对数似然函数,推导出相应的稳健估计并研究其在异常点存在时的统计性质。上述工作的完成将大大推动空间数据分析、广义线性混合模型统计推断理论的发展和应用。
中文关键词: 模型平均;模型选择;纠偏;随机效应模型;空间自回归模型
英文摘要: The spatial autoregressive model under normal specification cannot be adopted directly to analyze the non-normal, discrete or skewed spatial data with complex correlation structure. Based on the penalized quasi-log-likelihood function, the project will introduce the spatial weights matrix into the expression of generalized linear mixed model and propose a new generalized spatial autoregressive model with random effects. This model can be adopted to study the spatial data with complicated structure. Due to the embedded spatial weight matrix, the generalized spatial autoregressive model can extract the underlying prior information from the geographic space and improve the efficiency of statistical inference procedure. The estimation problems (including estimation method, large sample proterties and asymptotic distribution) of the fixed/random effects and autoregressive parameter of the proposed model will then be studied. Based on these, the definition of conditional Akaike information will be extended to the current setting and its asymptotically unbiased estimator (i.e. conditional Akaike information criterion) will be derived. The model selection performance of the derived criterion will also be assessed. Finally, based on the robustified penalized quasi-log-likelihood function, the corresponding robust estima
英文关键词: model averaging;model selection;bias correction;random effects model;spatial autoregressive model