二元相依生存数据的贝叶斯统计诊断  

Bayesian statistical diagnosis of bivariate dependent survival data

作　　者：佘思琪 唐安民[1] SHE Si-qi;TANG An-min(School of Mathematics and Statistics,Yunnan University,Kunming 650500,Yunnan,China)

机构地区：[1]云南大学数学与统计学院,云南昆明650500

出　　处：《云南大学学报（自然科学版）》2022年第3期453-463,共11页Journal of Yunnan University(Natural Sciences Edition)

基　　金：国家基金自然科学基金(11961079).

摘　　要：基于FGM Copula函数刻画二元相依生存数据的相关关系,B样条函数拟合边际基本危险函数,建议半参数二元相依生存模型,相比较于全参数模型,其更为灵活柔性,适应范围更广.在贝叶斯理论框架下,基于Gibbs抽样和MH抽样的混合算法,对模型未知参数进行估计同时,建议了贝叶斯数据删除影响分析和贝叶斯局部影响分析的统计诊断方法.数据模拟和实例分析例证了所提出的半参数模型及其算法的可行性,所建议的统计诊断方法能检测出潜在异常点或强影响点.Based on the FGM Copula function,we describe the correlation between bivariate dependent survival data,and fit the marginal basic risk function by the B-spline function.We suggest a semiparametric bivariate dependent survival model,which is more flexible and has a wider range of adaptation than the full-parameter model.Based on the hybrid algorithm together with the Metropolis-Hastings algorithm within the Gibbs sampler,we propose a feasible Bayesian method simultaneously to estimate unknown parameters of interest,and to fit baseline survival functions.Meanwhile,we propose statistical diagnosis methods,including Bayesian data deletion influence analysis and Bayesian local influence analysis.Data simulation and example analysis demonstrate the feasibility of the proposed semi-parametric model,and the potential outliers or significantly influence can be detected by the proposed statistical diagnosis method.

关 键 词：贝叶斯统计诊断 FGM Copula 二元相依生存数据 B样条 数据删除影响分析 局部影响分析 

分 类 号：O110.6740[理学—数学]