Abstract
dc:description.abstractRegression analysis is usually based on the method of least squares . However, the least squares estimator is highly sensitive to outliers and can be far from optimal when the errors are non-normal in distribution. As an alternative to the least squares estimator, the least absolute error estimator has attracted much attention during the past thirty years with the development of several efficient algorithms based on its connections to linear programming, least absolute error regression can be put into practical use. In this thesis, we review some basic theoretical properties of least absolute error regression, examine the asymptotic efficiency of the least absolute error estimator relative to the least squares estimator for a general family of error distribution, and demonstrate its use for practical data analysis in a variety of situations using computer routines written in S-Plus .
Degree
thesis:*- Level thesis:degree_level
- Master of Science M.Sc.
- Department thesis:degree_department
- Department of Mathematics and Statistics
- Year dc:date.issued
- 1998
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Tung, Yi-Yuan
Rights
dc:rights- Statement dc:rights
-
- Available to the World Wide Web
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1828/19946