under construction; last updated August 18, 2003
> Home > Epidemiology Vade Mecum under construction
Simpson's Paradox
When an exposure and a disease are associated with a third variable, the
association between the exposure and the disease ignoring the third variable
may differ from the association in strata of the third variable.
Here is a table at level 1 of the third variable. There is no association
with the factor and the disease:
|
|
Disease
|
No Disease
|
Total
|
|
Factor
|
100
|
500
|
600
|
|
No Factor
|
100
|
500
|
600
|
|
Total
|
200
|
1000
|
1200
|
Here is the table at level 2 of the third variable...again showing
no association between the risk factor and the disease.
|
|
Disease
|
No Disease
|
Total
|
|
Factor
|
2
|
118
|
120
|
|
No Factor
|
18
|
1062
|
1080
|
|
Total
|
20
|
1180
|
1200
|
But aggregating these two tables together, we find an odds ratio of
2.18.
|
|
Disease
|
No Disease
|
Total
|
|
Factor
|
102
|
618
|
720
|
|
No Factor
|
118
|
1562
|
1680
|
|
Total
|
220
|
2180
|
2400
|
All content © 2003 Mathepi.Com (except R and Rweb).
About us.