Research On Modelling Rotation Group Bias In Successive Sampling Estimation

How to produce accurate and reliable time series data using scientific successive sampling estimation method has always been a research focus at the forefront of the research of successive sampling theory and its application.Evidence is available from many empirical analyses that repeated interviewing of the same people can frequently change response patterns,which is the root of the problem:rotation group bias.A series of studies have been carried out in the classical approach field,but no ideal modified estimator has been founded.Meanwhile,with the rapid development of the time series approach,its concept of model-based rather than design-based provides a new way to eliminate rotation group bias.Therefore,choosing time series approach and rotation group bias as the research objects,related research results have been overviewed systematically at first to lay a solid foundation for the following study in this paper.Then,under the assumption that rotation group bias is the only systematic bias,the estimator of the population total can be decomposed into the population total,rotation group bias and sampling error so that a time series model accounting for rotation group bias has been established.In order to guarantee its applicability under various rotation sample survey,a generalized model has also been given in this paper.Finally,taking specific 2~2~2~1rotation sample survey as an example,the population total and 4 rotation groups'estimators of the population total has been produced by simulation.Then,the time series sampling estimation model ignoring and accounting for rotation group bias are compared using simulation data.Results show that the time series estimation model accounting for rotation group bias has two advantages:Firstly,the model can estimate most of the rotation group bias from rotation groups'estimators of the population total,significantly decreasing the negative effect of rotation group bias.Secondly,the population total estimator given by the model is closer to the population total simulated previously in the paper.That is,the model can obtain more accurate estimator of the population total.