Advances and Applications in Statistics
Volume 40, Issue 2, Pages 109 - 132
(June 2014)
|
|
ROBUST REGRESSION MODELING VIA PARAMETRIC LIKELIHOODS: AN ALTERNATIVE TO USING THE SKEW t DISTRIBUTION AND SEVERAL RANK-BASED METHODS
Tsung-Shan Tsou and Wei-Cheng Hsiao
|
Abstract: The skew tdistribution considered by Azzalini and Genton [2] has parameters believed to be able to properly capture the skewness and kurtosis possessed in data. We scrutinize the alleged robustness property of the skew tdistribution and some semi-parametric methods and show that they are robust only when the extent of asymmetry is mild and the magnitude of kurtosis is small. We recommend using an existing parametric robust likelihood approach to analyze data when one is uncertain about the distribution underlying the data. |
Keywords and phrases: skew tdistribution, skewness, kurtosis, robust likelihood. |
|
Number of Downloads: 361 | Number of Views: 1140 |
|