域的样本量分配方法研究

The Research on Sample Size Allocation of Domain

域估计是抽样调查的研究热点之一,其核心问题是样本量问题。在实践中,域包含计划域和非计划域,若为计划域,兼顾总体和域的估计精度的抽样设计方案更有效,其中关键是域样本量分配问题。域样本量的研究方法主要有5种,包含传统的比例、最优、内曼、等量分配方法和Power分配方法。通过使用中国家庭追踪调查数据对5种方法进行比较研究。研究表明,比例、最优、内曼分配方法有助于提高总体的估计精度,当总样本量较小时,无法兼顾域的估计精度;等量分配方法能提高域的估计精度,便于域之间的比较,但降低了总体的估计精度;Power分配方法虽然略微降低了总体的估计精度,却显著提高了域的估计精度,是兼顾总体和域的估计精度的样本分配方法。

The domain estimation is one of the research hotspots in sampling survey,the key problem is the sample size.In practical survey,there are planned domains and unplanned domains.If the domains are planned domains,the sample design is more effective which taken into account the accuracy of the population and domain estimation,and the key problem is the allocation of domain sample size.There are five kinds of sample size allocation methods,including proportional allocation,optimal allocation,Neyman allocation,equal allocation and power allocation.Then,we use the CFPS2010 data to compare the five sample size allocation methods.The study further illustrate,the proportional allocation,optimal allocation,Neyman allocation are helpful to improve the estimation accuracy of population,but not estimation accuracy of domain.Equal allocation improves the estimation accuracy of the domain,is helpful in the comparison between domains when the variance coefficients of each domain have little difference,but it will reduce the estimation accuracy of population.Although the power sample allocation method reduces the estimation accuracy of population slightly,it improves the estimation accuracy of the domain significantly.It is a better sample allocation method considering both the estimation accuracy of population and domain.