|
Activity Number:
|
379
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #308187 |
|
Title:
|
A Multiplicative Heteroscedasticity-Based Transformation To Reduce Heteroscedasticity in General Linear Models
|
|
Author(s):
|
Mitchell Rosen*+
|
|
Companies:
|
Omnicare Clinical Research
|
|
Address:
|
630 Allendale Rd., King of Prussia, PA, 19406,
|
|
Keywords:
|
transformation ; Box-Cox ; multiplicative heteroscedasticity ; homogeneity of variance ; delta method ; general linear model
|
|
Abstract:
|
Data analysts commonly use transformations to achieve homogeneity of variance, linearity, additivity, and/or normality. Box and Cox (1960) proposed estimating a single parameter power transformation model using maximum likelihood methods. Their method is optimum when normality of residuals is desired. In this paper the power transformation model is estimated using Harvey's (1976) multiplicative heteroscedasticity model. This method is optimum when homoscedasticity is desired. Estimation is accomplished using generalized linear model methods. Performance of the transformation vis-à-vis Box-Cox is assessed using Monte Carlo methods.
|
