scientific article; zbMATH DE number 3868411

From MaRDI portal
Publication:3334778





Consider the class of estimators of \(T=\sum^{N}_{i=1}a_ iY_ i\) given by \(\hat T=\sum^{N}_{i=1}a_ i\hat Y_ i+\sum^{N}_{i=1}r_ ia_ i\hat e_ i\), where \(\hat Y_ i=X^ t_ i{\hat \beta}\), \(\hat e{}_ i=Y_ i-\hat Y_ i\), \({\hat \beta}=(\sum_{s}q_ ix_ ix^ t_ i)^{-1}(\sum_{s}q_ ix_ iY_ i)\) and \(x_ 1,x_ 2,...,x_ N\) are known auxiliary p-vectors. This class contains the simple projection estimator (SPRO) when \(r_ i=0\), the linear prediction estimator (LPRE) when \(r_ i=1\) and the generalized regression estimator (GREG) when \(r_ i=\pi_ i^{-1}\). The forms of SPRO and LPRE subclasses are simple and realistic and appeal to the clients and hence they are called 'cosmetic'. In this paper, the authors prove the asymptotically design-unbiasedness property of these estimators and obtain their design-based variance using an extension of the Yates-Grundy expression. Finally, these estimation techniques are illustrated by an example taken from an electric power load research study.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3334778)