有限样本条件下重复抽样建模方法研究  

Modeling Method of Repeated Sampling Under Finite Sample Condition

作　　者：高艳平[1] 王晶[1] Gao Yanping;Wang Jing(School of Statistics, Shanxi University of Finance & Economics, Taiyuan 030006, China)

机构地区：[1]山西财经大学统计学院,太原030006

出　　处：《统计与决策》2019年第8期22-25,共4页Statistics & Decision

基　　金：国家社会科学基金资助项目(16BTJ016)

摘　　要：文章从数学及模拟仿真的角度证明特征样本重复抽样回归方法(FSR)的优越性。从数学角度证明了对于连续型变量,FSR中抽样方法可充分利用非原采样点的数据分布信息,从而提高估计精度;由FSR方法得到的统计量变异系数大小与抽样次数m成反比。通过对小样本参数估计的传统回归、Bayes Bootstrap和特征样本重复抽样方法的仿真模拟比较,验证了上述结论。This paper proves the superiority of the Features Sample Re-sampling Regression (FSR) method from the perspective of mathematics and simulation. Firstly the paper mathematically proves that for continuous variables, the sampling method in FSR can make full use of the data distribution information of non-original sampling points to improve the estimation accuracy, and that the statistical coefficient of variation obtained by FSR method is inversely proportional to the number of samples [m]. And then the paper makes simulation and comparison of small sample parameter estimation' s traditional regression, Bayes Bootstrap and the method of features sample repeated sampling to verify the above conclusions.

关 键 词：有限样本 特征样本重复抽样回归 先验分布 

分 类 号：O212[理学—概率论与数理统计]