基于广义回归估计的校准加权设计效应模型研究

The Research of Design Effect Model for Calibration Weighting Based on Generalized Regression Estimating

校准是最常用的加权调整方法,然而传统加权调整设计效应模型只考虑有差异权数导致的精度损失,忽略使用辅助信息后的精度改进,因此应用于设计效应计算时存在一定的缺陷。本文在Spencer模型的基础上进行拓展,引入反映辅助变量和调查变量相关关系的广义回归估计量,构建了校准加权设计效应的一般模型。数值分析结果显示,校准加权设计效应模型的效果优于传统加权调整设计效应模型;尤其在调查变量与辅助变量高度相关的情形下,校准加权设计效应模型能够准确地估计出不等概率抽样设计和校准调整的综合效率。

Calibration is the most commonly used weighting adjustment method.However,the traditional models for weighting design effects only consider loss in precision due to unequal weights,and ignore precision improvement with the use of auxiliary information.Therefore,the models may not apply well to calculating design effects.This paper extends Spencer model and uses a generalized regression estimator reflecting the correlation between auxiliary variables and survey variables to develop a general model for design effect of calibration weighting.The results of numerical analysis show that the effect of calibration weighting design effect model is better than that of traditional weighting adjustment design effect model.Especially when the survey variables are highly correlated with the auxiliary variables,the calibration weighting model can accurately estimate the joint efficiency of unequal probability sampling design and calibration adjustment.