Comparison of robust regression estimators

Loading...
Thumbnail Image
Date
2014
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This study evaluated the performance of the Ordinary Least Squares Estimator (OLSE) method of estimating regression parameters and some robust regression methods. The Least-Trimmed Squares Estimator (LTSE), Huber Maximum like-lihood Estimator (HME), S-Estimator (SE) and Modi ed Maximum likelihood Estimator (MME) were considered in this study. Criteria for the comparison were: coe cients and their standard errors, relative e ciencies, Root Mean Square Er-rors, coe cients of determination and the power of the test. The sensitivity of these robust methods were considered using Anthropometric data from Komfo Anokye Teaching Hospital. The dataset was on Total Body fat and Body Mass Index, Triceps skin-fold, Arm Fat as percent composition of the body and Height as predictors. Leverages were introduced rst into two variables, and into all predictors. The percentages were 5%, 10% and 15 % leverages. Also, 10%, 20% and 30% outliers were introduced in addition to 20% error contamination and contamination with data from non-normal distribution were considered. Results showed that robust methods are as e cient as the OLSE if the assumptions of OLSE are met. OLSE was a ected by leverages, outliers, contaminants and non-normality. HME broke-down with leverages in data, and was slightly a ected by outliers, contaminants and non-normality; whilst SE and MME were robust to all aberrations. LTSE was a ected by contaminants, non-normality, high outliers perturbation and was slightly a ected by leverages and low outliers perturbation.
Description
A thesis submitted to the Department of Mathematics, Kwame Nkrumah University of Science and Technology in partial fufillment of the requirement for the degree of Mphil Mathematical Statistics
Keywords
Citation