Utilizing auxiliary information in sample survey estimation and analysis
Utilizing auxiliary information in sample survey estimation and analysis
The use of auxiliary population information to improve estimation and analysis in sample surveys is considered in this thesis. Regression estimation is emphasized throughout because it provides a flexible, integrating, efficient and yet simple approach to incorporate auxiliary population information at the estimation stage of sample surveys. Three important issues regarding regression estimation are examined.
The sample-based choice of a 'best subset' of auxiliary variables to use in combination with regression estimators is considered first. This becomes important when the number of potential auxiliary variables is large, as is the case in some recent applications. It also highlights the role of model selection preceding regression estimation. While gains in efficiency can be achieved with regression estimation following variable selection based on specialized variance estimators, variance estimation itself becomes more difficult in this case.
The estimation of finite population distribution functions is considered next. A poststratified estimator of the distribution function is proposed. It was found to perform satisfactorily, providing useful gains in efficiency, while keeping computations fairly simple, when compared to a number of other distribution function estimators which incorporate auxiliary information at the estimation stage.
A pseudo maximum likelihood procedure using regression weights is developed to fit regular parametric superpopulation models. It is subsequently applied to the case of linear regression modelling of sample survey data. Some theoretical results suggest that gains in asymptotic design variance are possible when the regression estimator of the slope coefficients is compared to ordinary least squares and Pearson adjustment estimators. However, some limited simulation evidence indicates that these gains are minor when the target models are well specified. This evidence also suggests that using regression weights for model fitting via pseudo maximum likelihood provides similar results as when the standard design weights are considered.
University of Southampton
Nascimento Silva, Pedro Luis do
1996
Nascimento Silva, Pedro Luis do
Nascimento Silva, Pedro Luis do
(1996)
Utilizing auxiliary information in sample survey estimation and analysis.
University of Southampton, Doctoral Thesis.
Record type:
Thesis
(Doctoral)
Abstract
The use of auxiliary population information to improve estimation and analysis in sample surveys is considered in this thesis. Regression estimation is emphasized throughout because it provides a flexible, integrating, efficient and yet simple approach to incorporate auxiliary population information at the estimation stage of sample surveys. Three important issues regarding regression estimation are examined.
The sample-based choice of a 'best subset' of auxiliary variables to use in combination with regression estimators is considered first. This becomes important when the number of potential auxiliary variables is large, as is the case in some recent applications. It also highlights the role of model selection preceding regression estimation. While gains in efficiency can be achieved with regression estimation following variable selection based on specialized variance estimators, variance estimation itself becomes more difficult in this case.
The estimation of finite population distribution functions is considered next. A poststratified estimator of the distribution function is proposed. It was found to perform satisfactorily, providing useful gains in efficiency, while keeping computations fairly simple, when compared to a number of other distribution function estimators which incorporate auxiliary information at the estimation stage.
A pseudo maximum likelihood procedure using regression weights is developed to fit regular parametric superpopulation models. It is subsequently applied to the case of linear regression modelling of sample survey data. Some theoretical results suggest that gains in asymptotic design variance are possible when the regression estimator of the slope coefficients is compared to ordinary least squares and Pearson adjustment estimators. However, some limited simulation evidence indicates that these gains are minor when the target models are well specified. This evidence also suggests that using regression weights for model fitting via pseudo maximum likelihood provides similar results as when the standard design weights are considered.
This record has no associated files available for download.
More information
Published date: 1996
Identifiers
Local EPrints ID: 459801
URI: http://eprints.soton.ac.uk/id/eprint/459801
PURE UUID: cbb6a876-2948-40ee-9f20-07fca709c940
Catalogue record
Date deposited: 04 Jul 2022 17:18
Last modified: 04 Jul 2022 17:18
Export record
Contributors
Author:
Pedro Luis do Nascimento Silva
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics