University of Southampton
Empirical likelihood approach for estimation from multiple sources
Abstract
dc:description.abstractEmpirical likelihood is a non-parametric, likelihood-based inference approach. In the design-based empirical likelihood approach introduced by Berger and De La Riva Torres (2016), the parameter of interest is expressed as a solution to an estimating equation. The maximum empirical likelihood point estimator is obtained by maximising the empirical likelihood function under a system of constraints. A single vector of weights, which can be used to estimate various parameters, is created. Design-based empirical likelihood confidence intervals are based on the χ<sup>2</sup> approximation of the empirical likelihood ratio function. The confidence intervals are range-preserving and asymmetric, with the shape driven by the distribution of the data.<br/><br/>In this thesis we focus on the extension and application of design-based empirical likelihood methods to various problems occurring in survey inference. First, a design-based empirical likelihood methodology for parameter estimation in two surveys context, in presence of alignment and benchmark constraints, is developed. Second, a design-based empirical likelihood multiplicity adjusted estimator for multiple frame surveys is proposed. Third, design-based empirical likelihood is applied to a practical problem of census coverage estimation.<br/><br/>The main contribution of this thesis is defining the empirical likelihood methodology for the studied problems and showing that the aligned and multiplicity adjusted empirical likelihood estimators are √n-design-consistent. We also discuss how the original proofs presented by Berger and De La Riva Torres (2016) can be adjusted to show that the empirical likelihood ratio statistic is pivotal and follows a χ<sup>2</sup> distribution under alignment constraints and when the multiplicity adjustments are used.<br/><br/>We evaluate the asymptotic performance of the empirical likelihood estimators in a series of simulations on real and artificial data. We also discuss the computational aspects of the calculations necessary to obtain empirical likelihood point estimates and confidence intervals and propose a practical way to obtain empirical likelihood confidence intervals in situations when they might be difficult to obtain using standard approaches.
Degree
thesis:*- Name dc:type.qualificationname
- Ph.D.
- Level dc:type.qualificationlevel
- doctoral
- Grantor dc:publisher.institution
- University of Southampton
- Year dc:date.issued
- 2017
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kabzinska, Ewa Joanna
- Advisors dc:contributor.advisor
-
- Berger, Yves
- Zhang, Li-Chun
- ORCID person:identifier.orcid
Rights
- Language dc:language
- English, eng