|
Activity Number:
|
37
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 2, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #305205 |
|
Title:
|
Bayesian Methods for Detecting Influential Observations When Using the Box-Cox Transformation
|
|
Author(s):
|
Lawrence I. Pettit*+ and Nalaiyini Sothinathan
|
|
Companies:
|
Queen Mary University of London and Queen Mary University of London
|
|
Address:
|
School of Mathematical Sciences, London, E1 4NS, United Kingdom
|
|
Keywords:
|
Bayesian methods ; Box-Cox transformation ; Influential observation ; Bayes factor ; Masked outlier
|
|
Abstract:
|
Riani and Atkinson (Technometrics, 2000) discuss the use of the forward search method for detecting influential observations when a Box-Cox transformation is to be used in a linear model. using one of the data sets from the original paper by Box and Cox they adapt it by adding a group of masked outliers and suggest that these observations cannot be detected by single deletion diagnostics. We show that by using a combination of the $k_d$ diagnostic of Pettit and Young (Biometrika, 1990) and the conditional predictive ordinate we can indeed unmask these particular observations.
|
