缺失数据假设下的灵活贝叶斯分位数回归分析  

Flexible Bayesian Quantile Regression Analysis under Missing Data Hypothesis

作　　者：谢正粉 戴琳[1] 付英姿[1] XIE Zhengfen;DAI Lin;FU Yingzi(Faculty of Science,Kunming University of Science and Technology,Kunming 650500,China)

机构地区：[1]昆明理工大学理学院,昆明650500

出　　处：《河南科学》2019年第2期157-164,共8页Henan Science

基　　金：国家自然科学基金资助项目(11201200;11561035)

摘　　要：考虑缺失数据假设下,响应变量为有序分类变量的灵活贝叶斯分位数回归分析.对有序分类变量添加一潜在变量,并假设潜在变量服从广义偏斜拉普拉斯分布,由此建立起有效的贝叶斯后验推断程序.与传统的贝叶斯分位数回归方法相比,本方法的主要优势在于广义非对称拉普拉斯分布的尾部灵活,具有高度的适应性.最终,以云南省某医院生殖遗传中心的试管婴儿数据为例来说明本方法的有效性和适用性.This paper introduces the flexible Bayesian quantile regression model to analyze the response variables as an ordered categorical variables under non-ignorable missing mechanism. Specifically,a latent variable is added to the ordered categorical variables,and the latent variable is assumed to obey the generalized skewed Laplace distribution,thereby establishing an effective Bayesian posteriori inference procedure. Compared with the traditional Bayesian quantile regression method,the main advantage of this method is that the tail of the generalized asymmetric Laplace distribution is flexible and highly adaptable. Finally,the paper considers the real data to illustrate the effectiveness and applicability of our proposed method.

关 键 词：二维光滑 B-样条基函数 广义非对称拉普拉斯分布 缺失数据 贝叶斯分位数回归 

分 类 号：O212.8[理学—概率论与数理统计]