|
Activity Number:
|
75
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, August 9, 2004 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistical Education
|
| Abstract - #300813 |
|
Title:
|
Influence of Binary Confounders on Associations Using a Noninteractive Model
|
|
Author(s):
|
Milo Schield*+ and Thomas Burnham
|
|
Companies:
|
Augsburg College and Cognitive Consulting
|
|
Address:
|
2211 Riverside Drive, Minneapolis, MN, 55410,
|
|
Keywords:
|
epidemiology ; causation ; significance
|
|
Abstract:
|
The defining and sufficient conditions for a binary confounder to nullify or reverse an association involving two binary factors have been previously identified using a noninteractive model. This paper investigates various models or summaries of these binary confounder conditions or families using confounder size. The relationship between confounder size and the association that can be nullified or reversed thereby is determined. An association that is confounder-resistant to nullification or reversal by confounders below a given size can be viewed as "confounder significant" against confounders up to this size. This approach allows researchers, editors, and journalists to set a minimum standard for the confounder significance of associations in the same way that alpha is used to set a minimum standard for statistical significance. Being confounder significant is argued to have an increasing importance as the size of datasets increase and the effect sizes that are statistically significant decrease. This paper also extends the analysis of outcomes from binary to continuous.
|
