First , this paper gives an introduction of some methods of unequal probability sampling , their estimators and variance estimators , including sampling with pps with replacement , methods of sampling without replacement suggested by brewer , durbin , sampford , des raj , murthy , rao - hartley - cochran . then , at the basis of rao and bayless ' s study , we consider that population can be splited two random subpopulations , which are respectively drawn from different infinite super - populations , and compare the stabilities of estimators of the methods that given above . we find that the minor difference between two super - populations has great effect on the efficiency of the estimators for the population with moderately large coefficient of variation ( c . v . ) 本文首先从理论上介绍了若干种不等概率抽样方法，它们的估计量、估计量的方差及其估计，其中包括有放回ppz及pps抽样，不放回不等概率抽样中的brewer 、 durbin 、 sampford 、 desraj ， murthy 、 rao - hartley - cochran等人的方法；其次，在rao和bayless两人就样本单元数n = 2的情形对上述抽样方法进行比较的基础上，将总体随机地分成两个子总体，视每个子总体取自不同的线性超总体，在文中，我们利用计算机实现随机分组，并通过画图比较各方法估计量的稳定性，结果表明，对变异系数c . v . ( x )较大的总体而言，两个超总体之间的微小差异将对估计量的稳定性产生很大的影响，从而说明rao和bayless的比较结果还不够完善。
Methods : 3312 people in year of 2001 , 4990 in year of 2002 , who took part in the aat for government functionary in gd province ; aii of the datas split into two parts . one part is in use of efa . the other is in use of cfa . results : ( 1 ) aat is valid in testing administrative aptitudes such as " mathematical aptitude " , " language aptitude " and " reasoning aptitude " 方法： 2001年4月和2002年11月广东省公务员考试中使用的两套《行政职业能力倾向测验》 ； 2001年4月测验的有效答卷3312份和2002年11月测验的有效答卷4990份；所有数据随机分组，一半用于efa探索模型，另一半用于cfa验证模型。
To observe the curative effect of central serous chorioretinopathy ( csc ) treated with modified wulingsan and laser coagulation , the patients were randomly divided into three groups , the results showed that modified wulingsan and laser coagulation is a better treatment for csc 摘要观察五苓散加减合激光治疗中心性浆液性脉络膜视网膜病变，采用随机分组比较的方法，中药合激光治疗组与激光治疗组、中药组相比具有显著性差异，提示五苓散加减合激光是治疗中心性浆液性脉络膜视网膜病变的较好方法。
The best qualitative results indicated that rvpcs model with md grouping scheme reached a discrimination rate of 100 % with 1 % accuracy in the randomly assigned background constituents of the same prescription , and rvpcs model with rs grouping scheme gave a discrimination rate of 86 % ~ 96 % with 2 % accuracy in the randomly assigned background constituents of 4 mixed prescriptions 四种处方之不同背景混合处方分析中，以主成份空间之残差变异定性模式搭配随机分组方式，可达到86 % ~ 96 %之辨识成功率，并具有2 %之微量检测精度。
