观察次数不等的SURS中回归系数的两步估计  

Two stage Estimators of the Regression Coefficients in SURS with Unequal Numbers of Observations

作　　者：丁树良[1] 

机构地区：[1]江西师范大学计算机系

出　　处：《江西师范大学学报（自然科学版）》1998年第1期24-32,共9页Journal of Jiangxi Normal University(Natural Science Edition)

基　　金：江西省自然科学基金

摘　　要：该文着重讨论协方差阵未知时观察次数不等的半相依回归系统（ＳＵＲＳ（ＵＮＯ））中回归系数β的两步估计问题，对３个因素的４８种不同水平组合进行ＭｏｎｔｅＣａｒｌｏ模拟，并对模拟结果进行统计检验，发现在我们给出的ｓｄ准则下：１．使用Ｓｃｈｍｉｄｔ推荐的估计量导出的两步Ｚｅｌｎｅｒ估计（ＴＺＥ）性质不好；特别是模拟中相关系数很高（＞０．９４８）时表现更差；２．一般情况下，两步广义协方差改进估计的行为比ＴＺＥ更好；３．β或的不同估计方法、设计阵的不同选取、模拟中对的选择不同，都对两步估计有影响．The focus of this paper is to discuss the statistical properties of several two stage estimators of the regression coefficients (β) in seemingly unrelated regression system with unequal numbers of observations(SURS(UNO)) when the covariance matrix unknown.For m=2,the Mntor Carlo study with 3 factors 48 levels combinations and the statistical test are considered.The following facts are discorved: (1)Though Zellner's estimator of β is BLUE when  is known,two stage Zellner's estimator(TZE) is not so good when  is unknown and using ^ which was recommended by Schmidt,as a substitute for ,especially,when the correlation coefficient,which is defined by  in Monte Carlo study,is very high(>0.948). (2)Usually,the behavior of two stage generalized covariance improvement estimate(TGCIE) of β is better than that of TZE under some criterion. (3)The following factors,such as the methods of estimating β or ,the choices of design matrices or the covariance matrix  in the expe-riments of Monte Carlo study,effect the statistical properties of two stage estimator in some extent.

关 键 词：半相依回归系统 回归系数 两步估计 

分 类 号：O212.1[理学—概率论与数理统计]