Research On Calibration Estimation And Its Application Based On Semi-parametric Model

The accuracy of sampling estimation can be improved by using the auxiliary information which is closely related to the study variables.With the development of big data technology,many daily data can be recorded completely.Using these completely recorded data can make the result of sampling survey more accurate.It is called the auxiliary variable,when the auxiliary information related to the study variable enters the sample There are different functional relationships between the study variables and the auxiliary variables,so the corresponding model should be built in the sampling estimation stage.Model calibration estimation is a model aid method that uses auxiliary information to make the weights of the study variables closer to the population.For the different relationships between the auxiliary variables and the study variables,the corresponding superpopulation model is established to improve the estimation accuracy.In this paper,on the basis of the traditional model calibration estimation method,the semi-parametric regression method is introduced to construct the super-population model,and the semi-parametric model calibration estimator is constructed,and the variances of the estimators are given.The model calibration estimation is extended to two-stage sampling,and the expressions of the model calibration estimator under second-order sampling are given according to different conditions.Numerical simulation is carried out.Using correlation deviation(RB)and correlation efficiency(RE)as evaluation criteria,the performance of different calibration estimators was compared when the auxiliary variables had different functional relationships with the study variables.For the model calibration estimator under second order sampling,the error rate of the estimator is used to evaluate its properties,and the advantages of the model calibration estimator are proved.Finally,using the data of the employed population in Guangdong Province to make an empirical analysis,the total value of the population is estimated by simple random sampling and second order sampling respectively,and the estimated results are compared and analyzed.According to the numerical simulation results,In terms of correlation efficiency,no matter whether the relationship between the study variable and the auxiliary variable is linear or semi-linear,the semi-parameter calibration estimator is far superior to other estimators,so it has a wider application range and is more flexible to use.For the model calibration estimation under second-order sampling,the estimation accuracy of the model calibration estimator using the auxiliary information of the primary element and the estimator using the auxiliary information of the total element are improved,and the estimator using the auxiliary information of the total element for calibration is better.