回归系数变点估计的快速非迭代抽样算法  

Fast Non-iterative Sampling Algorithm for Change-point Estimation of Regression Coefficients

作　　者：杨丰凯 袁海静 

机构地区：[1]山东大学(威海)数学与统计学院,山东威海264209

出　　处：《统计与决策》2017年第24期10-13,共4页Statistics & Decision

基　　金：山东大学(威海)教学改革研究项目(B201623)

摘　　要：文章讨论了线性回归模型中回归系数变点位置估计的非迭代抽样算法。在贝叶斯框架下,分别采取无信息先验和共轭先验,利用逆贝叶斯公式,得到来自变点位置后验分布的独立同分布的样本,可直接用于变点位置的统计推断。避免了Gibbs抽样算法中的收敛性诊断问题以及样本的相依性问题。This paper discusses a non-iterative sampling algorithm for the estimation of the change-point position of regres- sion coefficients. In the Bayesian framework, the paper uses an informative priori, a non-informative priori respectively, and the inverse Bayesian formula to obtain the independently and identically distributed samples from the posterior distribution of the change-point position. The samples can be directly used for statistical inferences of the change-point position, which avoids the problems of diagnosis of convergence and the dependency of samples in Gibbs sampling.

关 键 词：线性回归 系数变点 逆贝叶斯公式 GIBBS抽样 

分 类 号：O212.8[理学—概率论与数理统计]