|
Activity Number:
|
198
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 30, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #308589 |
|
Title:
|
Something Might Be Missed in Missing Data Analysis
|
|
Author(s):
|
Xiang Guo*+ and Guangrui Zhu
|
|
Companies:
|
sanofi-aventis and Eisai Medical Research Inc.
|
|
Address:
|
35 River Dive South, Jersey City, NJ, 07310,
|
|
Keywords:
|
Fictitious assumption ; Minimum Homogeneous Group ; Missing data ; Randomization
|
|
Abstract:
|
To make inference on missing data, an essential step is to posit assumptions about missing values. Those assumptions can dramatically affect the analysis result. It is hard to tell which result is unbiased since the methods are based on unverifiable assumptions. This core difficulty is generally not highlighted by the modelers, and the results often be mistakenly weighted by decision makers. We introduce a concept of fictitious assumption and distinguish it from the traditional concept of statistical modeling assumptions where observed data can verify the assumptions to certain degrees. Viewing the missing data handling methods from fictitious inference point of view can make the dependency of the inferences transparent to the decision makers. For available missing data methods, we use fictitious formulation to demonstrate the impact of the fictitious assumptions on the inference.
|
