|
Activity Number:
|
320
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 14, 2002 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biopharmaceutical Section*
|
| Abstract - #300956 |
|
Title:
|
Which Method to Use: Direct Modeling of Heteroscedasticity or Modeling Based on Transformed Data
|
|
Author(s):
|
Hui Zhi*+ and Ashwini Mathur and Lisa Ying
|
|
Affiliation(s):
|
GlaxoSmithKline, Inc. and GlaxoSmithKline, Inc. and Merck & Company, Inc.
|
|
Address:
|
One Franklin Plaza, FP1375, PO BOX 7929, Philadelphia, Pennsylvania, 19101-7929, USA
|
|
Keywords:
|
heteroscedasticity ; weighted least squares ; clinical pharmacology study ; transformation ; confidence interval
|
|
Abstract:
|
The problem of non-homogeneous variance occurs often in our relevant projects. In most of these projects, we are interested in estimating the mean differences and related confidence intervals among treatment groups. Two methods for approaching this kind of problem are discussed and evaluated under various assumptions. One is the direct modeling of heteroscedasticity using weighted least square regression, where the responses are weighted according to the variance of the corresponding treatment group. The other one is the modeling based on data transformation, where a certain transformation is applied so that the transformed responses have constant variance. Especiall, in clinical pharmacology studies, sample sizes are relatively small, usually less than twenty subjects per treatment group. Under such circumstance, simulation studies are performed to compare the two methods. Based on the simulations, point estimates, corresponding confidence intervals and coverage probabilities are calculated and compared. Recommendations and suggestions on which modeling method to be used are provided.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2002 program |
