Statistical modelling to treat data with missing value

用于处理缺失值数据的统计模型

基本信息

批准号：
15500187
负责人：
KIKUCHI Yasuki
金额：
$ 2.05万
依托单位：
NAGASAKI UNIVERSITY
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (C)
财政年份：
2003
资助国家：
日本
起止时间：
2003 至 2005
项目状态：
已结题

来源：
https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-15500187/
关键词：
incompelete information EM-algorithm normal mixture model hidden Markov normal model estimation of survival function Gamma-Weibull distribution observed information matrix Bootstrap method 不完全データ BootStrap法 FM-algorithm 一般ガンマ分布潜在マ

项目摘要

The purpose of this research was to construct the effective statistical models and to develop the effective method of estimation about the error of the estimates in statistical inference based on incomplete information such as data containing the missing value.We compared the estimates of the variance obtained by the observed information matrix and Bootstrap method for the estimates obtained by EM-algorithm for Poisson mixture model. Kikuchi and Nomakuchi presented this result at the 17th meeting of Japanese Society of Computational Statistics (2003.05).We studied about the estimation of the parameters for normal mixture model and hidden Markov normal model by EM-algorithm and the variance of the estimates by the observed information matrix. Kikuchi and Nomakuchi will present this result at the 20th meeting of Japanese Society of Computational Statistics (2006.06).Kikuchi became the research partaker of the research project of Ministry of Health, Labour and Welfare since 2003. We added a new theme relevant to the application of EM-algorithm to the clinical trial. We developed the method of the estimation of the survival function assuming the Gamma-Weibull distribution and the generalized Gamma distribution. Kikuchi and Nomakuchi presented this result at the 18th meeting of Japanese Society of Computational Statistics (2004.05), and Kikuchi, Nomakuchi and Anraku presented a paper about this result entitled "Estimation of the survival function by EM-algorithm (in Japanese)" on the collected papers of the above research project.

本研究的目的是构建有效的统计模型，并开发基于不完整信息（例如包含缺失值的数据）的统计推断中估计误差的有效估计方法。我们比较了通过以下方式获得的方差估计：观测信息矩阵和 Bootstrap 方法用于通过泊松混合模型的 EM 算法获得的估计值。 Kikuchi 和 Nomakuchi 在日本计算统计学会第 17 次会议（2003.05）上发表了这一结果。我们研究了用 EM 算法对正态混合模型和隐马尔可夫正态模型的参数估计以及观测到的估计方差信息矩阵。菊池和野间口将在日本计算统计学会第20次会议（2006.06）上公布这一成果。菊池自2003年起成为厚生劳动省研究项目的研究参与者。我们增加了一个与应用相关的新主题EM算法的临床试验。我们开发了假设伽玛-威布尔分布和广义伽玛分布的生存函数估计方法。 Kikuchi 和 Nomakuchi 在日本计算统计学会第 18 次会议（2004.05）上发表了这个结果，Kikuchi、Nomakuchi 和 Anraku 在上述研究项目的论文集。