2001 Fiscal Year Final Research Report Summary
On Confounding Biases in Observational Studies
| Project/Area Number |
12680320
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Statistical science
|
|---|
| Research Institution | Okayama University of science |
|---|
Principal Investigator |
YAMAMOTO Eiji Okayama Univ. of Science,Dept. of Infomatics, Prof., 総合情報学部, 教授 (50068920)
|
|---|
| Project Period (FY) |
2000 – 2001
|
| Keywords | ODDS RATIO / CONFOUNDING / NONOCOLLAPSIBILITY / CAUSAL INFERENCE / PROBABILITY CONTUOUR METHOD / CAUSAL GRAPH / STRATIFIED ANALYSIS / BACHDOOR CRITERION |
|---|
| Research Abstract |
An association measure of two binary categorical variables, odds ratio could be strongly effected by potential confounders. An ordinal method for controlling confounding is stratification of a contingency table by each level of the confounder. It is known that odds ratio has noncollapsibility with or without confounding. In this study, we succeeded to explain geometrically and analytically mechanism of the noncollapsibility by a probability contour method. The difference between a crude odds ratio and a common stratified odds ratio is separated in a noncollapsibility part and a confounding part. The above discussion is extended to a multilevel stratification. The sufficient conditional condition of collapsibility of odds ratio is shown to be a point on the probability contgur. The backdoor criterion, a sufficient estimable condition of causal effects in the causal diagram is fixed. Forth case studies are conducted in statistical analyses of causal inference including controlling potential confounders.
|
Research Products
(10 results)
