|
Activity Number:
|
529
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #303356 |
|
Title:
|
A Power Transformation for Minimizing Heteroscedasticity in the General Linear Model
|
|
Author(s):
|
Mitchell J. Rosen*+
|
|
Companies:
|
ICON Clinical Research
|
|
Address:
|
1700 Pennbrook Parkway, North Wales, PA, 19454,
|
|
Keywords:
|
Power transformation ; Box-Cox transformation ; delta method ; multiplicative heteroscedasticity ; multiplicative heteroscedasticity ; multiplicative heteroscedasticity
|
|
Abstract:
|
Box and Cox pioneered the use of power transformations to normalize residuals in a linear model. However, when the aim is to minimize heteroscedasticity the Box-Cox transformation and its analogues may not be effective. The power transformation proposed in this study is designed to reduce heteroscedasticity by minimizing the correlation between the residuals and predicted values. Maximum likelihood estimates for a multiplicative heteroscedastic model and a generalized heteroscedastic model that permits negative values are readily obtained. The hypothesis that the model residuals and predicted values are correlated can be tested using a Lagrange multiplier statistic.
|
