|
Activity Number:
|
339
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #306265 |
|
Title:
|
A Novel Statistical Approach to Identifying and Limiting the Effect of Influential Observations
|
|
Author(s):
|
Tamekia Jones*+ and David Redden
|
|
Companies:
|
The University of Alabama at Birmingham and The University of Alabama at Birmingham
|
|
Address:
|
327 Ryals Public Health Building, Birmingham, AL, 35294,
|
|
Keywords:
|
outliers ; leverage ; influential observations ; robust regression ; MCD
|
|
Abstract:
|
Outliers are observations with extreme standardized deviations between the observed dependent variables and their predicted values. Outliers can be detected using studentized residuals. Leverage is a measure of the standardized deviation of an observation's independent variable from the mean of the independent variables and is assessed using the diagonal of the hat matrix. An influential observation (outlier and leverage point) affects estimation of regression parameters. Detection of these points is difficult due to the masking effect which occurs when influential points are hard to identify using regression diagnostics. We present a robust regression method that extends Rousseeuw's concept using the minimum covariance determinant. Results utilizing our proposed method illustrate that our approach overcomes the masking effect by properly identifying influential observations.
|
