Generalized Regression Calibration Estimation Method And Its Extension Research Under Successive Sampling

In recent years,how to improve the precision of sampling estimators with given auxiliary information has become an important research issue.Traditional sampling methods focus more on the usage of auxiliary information in sampling design instead of their usage in sampling estimation.The generalized regression estimator is an estimator that makes use of auxiliary information in the stage of sampling estimation.By building super-population regression model and making model-assisted sampling estimation,this estimator has higher estimation precision.The calibration estimator makes use of auxiliary information to calibrate the design weights of the samples based on post-stratification,which not only greatly reduces the difference between structures of the samples and population,but also lowers the sampling error and improves estimation precision.Focusing on how to improve estimation precision by using auxiliary information,the generalized regression calibration estimator is built based on the calibration estimator and the generalized regression estimator.Furthermore,the generalized regression calibration estimator is extended to the successive sampling of rotating sampling survey to make use of the auxiliary information of matched and unmatched samples.Good properties of the estimator,including the design unbiasedness and asymptotic consistency,are proved theoretically.Numerical analysis is also conducted under traditional and successive sampling settings,and the proposed estimator is assessed using the root mean square error,correlation deviation,related efficiency and error rate,under different survey scales and correlations.By comparing the simulation results under different settings,the following conclusions are shown.Under the traditional survey sampling,the generalized regression calibration estimator behaves better than the traditional generalized regression estimator and calibration estimator.Under the successive sampling setting,the proposed estimator has smaller error rate than the other estimators.The proposed estimator can be used in survey sampling in government statistics to obtain higher precision of the estimators.