比例数据基于Tobit分位数回归模型的贝叶斯变量选择

Variable Selection of Proportional Data Based on Tobit Quantile Regression Model

针对家庭商业健康保险参保比例在[0,1]闭区间上取值的特点,本文基于Tobit模型给出了比例响应数据的贝叶斯分位数回归建模方法。通过引入回归系数的“Spike-and-slab”先验分布,应用EM算法我们提出了基于门限规则的贝叶斯变量选择方法。大量数值模拟研究验证了所提的贝叶斯变量选择方法的有效性,且具有易操作、计算量小等优点。最后,将此方法应用到家庭商业健康保险数据的实证分析,研究不同分位数水平下家庭健康保险参保比例的影响因素,得到了许多有意义的研究结果。

According to the features of the proportional data taken values on the closed interval[0,1]for the household coverage rate of commercial health insurance,in this paper,the Bayesian quantile regression modeling method is investigated for proportional response data based on the Tobit model.By introducing the“spike-and-slab”prior distributions of coefficients,the threshold rule-based Bayesian variable selection method is proposed by applying the EM algorithm.Extensive numerical simulation studies are used to verify the effectiveness of the proposed Bayesian variable selection method,and it has many merits such as easy implementation and cheap computation.Finally,the newly proposed method is applied to analyze the household coverage rate of commercial health insurance data,the influence factors of the coverage rate of commercial health insurance are uncovered at different quantile levels and some interesting results are gotten.