基于轮换样本连续性调查的校准组合估计方法研究

The Study of Calibration Composite Estimation Based on Rotating Panel Successive Sampling Survey

校准估计是抽样调查中比较常用的一种利用辅助信息提高估计量精度的方法。回归组合估计量作为轮换样本连续性调查中使用的一种有效的估计量,是可以通过校准程序得到的。基于回归组合估计量和校准程序之间的关系,本文提出了轮换样本连续性抽样调查条件下的不同校准组合估计量及其方差估计。校准组合估计量的主要思想是在校准估计程序中将拼配样本和非拼配样本的辅助信息进行不同的组合利用。本文利用美国现时人口调查的微观数据进行数值模拟,来比较不同校准组合估计量的估计效率,模拟结果表明两步校准组合估计量和两步校准双组合估计量的表现相似,且估计精度都高于H-T估计量及回归组合估计量;而两步校准组合估计量由于其简便性更适合应用于实践中。最后以我国农村住户连续性抽样调查为例,设计一套符合我国实际的轮换样本连续性调查方案,并将提出的校准组合估计量运用于估计阶段,为中国政府统计调查提供一定的借鉴和参考.

Calibration is commonly used in survey sampling by utilizing auxiliary information at the estimation stage of a research variable to improve the precision of the estimation. As an efficient estimator applied in successive rotating panel survey, regression composite estimator can be derived through a calibration procedure. Based on the relationship between regression composite estimator and calibration procedure, different calibration composite estimators and their variance estimators under the framework of successive rotating panel survey sampling are launched. The main idea of calibration composite estimators is to composite auxiliary information of matched and unmatched samples in the calibration estimation process in different ways. A numerical simulation using the microdata of American Current Population Survey is proposed to get insight into the efficiency of different estimators. It is shown that two-step calibration composite estimator has a similar performance as two-step calibration double composite estimator, and both of them are better than H-T estimator or regression composite estimator. In practice, two-step calibration composite estimator is more suitable due to its simplicity compared with two-step calibration double composite estimator. At last, Chinese rural household successive sampling survey is taken as an example, and a rotating panel successive sample survey scheme is proposed with the calibration composite estimator launched using in the estimation phase, which could be a good reference for our governmental statistical survey.