非独立数据内部相关系数的抽样分布  

Sampling Distribution of Intra-unit Correlation Coefficient for Non-independent Data

作　　者：顾海雁[1] 陈峰[1] 任仕泉[2] 

机构地区：[1]南通医学院医学统计学教研室,226001 [2]四川大学管理科学与工程学院,610065

出　　处：《中国卫生统计》2002年第6期325-327,共3页Chinese Journal of Health Statistics

基　　金：国家自然科学基金资助课题 ( 3950 0 12 7)

摘　　要：目的　研究非独立数据组内相关系数 ρ的抽样分布。 方法　用MonteCarlo方法从已知组内相关系数的假定总体中重复抽样 ,用RIGLS法拟合二水平模型 ,分别计算 ρ值以观察其分布类型。 结果　组内相关系数抽样分布类型随 ρ的数值大小而变化 ,小于 0 5时为正偏态 ,大于 0 5时为负偏态 ,接近 0 5时为近似正态分布。其估计误差与 1水平和 2水平的样本含量有关 ,总样本含量相同时 ,2水平的样本含量越大 ,估计误差越小。结论　组内相关系数的大小反映非独立数据的共变关系大小 。Objective To declare the distribution of inta unit correlation coefficient (ρ) of non independent data.Methods Monte Carlo method is used to get random samples from the populations with assumed parameters.Restricted iterative generalized least squares (RIGLS) is used to estimated the parameters of the level 2 model.The intra unit correlation coefficient is estimated based on the model.Results The type of sampling distribution varies from the ρ. When ρ is less than 0\^5,ρ is distributed which has positive skewness,when large than 0\^5,ρ is distributed which has negative skewness,and when about 0\^5,then distribution is normal.The standard error of ρ is proportion to the sample size.With the same total sample size,the larger the number of level 2 unit is,the smaller the SE of intra unit correlation coefficient will be.Conclusion The intra unit correlation coefficient of non independent data always mirrors the dependence of the data.The results of interested can act as a guide in the further analysis of non independent data.

关 键 词：组内相关系数 非独立数据 MonteCrrlo方法 多水平模型 限制迭代广义最小二乘 卫生统计 

分 类 号：R195[医药卫生—卫生统计学]