The Simpson paradox [Le paradoxe de Simpson].
The Simpson Paradox is based on the following phenomenon:
Let {n1,d1,n2,d2,n3,d3,n4,d4} be eight integer numbers such that:
n1 n2
---- < ----
d1 d2
n3 n4
---- < ----
d3 d4
Then:
\
n1 n2 |
---- < ---- |
d1 d2 | n1 n3 n2 n4
| ==> ---- + ---- < ---- + ----
n3 n4 | d1 d3 d2 d4
---- < ---- |
d3 d4 |
/
But unfortunately (and obviously) nothing can be said about:
n1 + n3 n2 + n4
--------- ? ---------
d1 + d3 d2 + d4
where '?' means '<', '=' or again '>'.
Nota: this wrong way for adding two rational numbers is the so-called Median-Mean.
The {n1,d1,n2,d2,n3,d3,n4,d4} space is a eight-dimensional one
and then difficult to visualize. In order to use a bidimensional space, for
{n1,d1,n2,d2,n3,d3,n4,d4} inside [1,4], the two
following coordinates are computed:
X = (((n1-1)*4 + (n3-1))*4 + (d1-1))*4 + (d3-1)
Y = (((n2-1)*4 + (n4-1))*4 + (d2-1))*4 + (d4-1)
Then, the white points {X,Y} display the case where:
n1 + n3 n2 + n4
--------- > ---------
d1 + d3 d2 + d4
the so called paradoxal case...
For example:
1 2
--- < ---
3 4
4 2
--- < ---
3 1
when:
1 + 4 2 + 2
------- > -------
3 + 3 4 + 1
By the way {Black, Red, Green} colors are used for the 3x3x3-2=25 other cases...
See some related pictures (including this one):
More information on this subject
is
available in the july 2013 issue of PLS (Pour La Science)
with the Jean-Paul Delahaye's paper L'embarrassant Paradoxe de Simpson.
See my own discovery of the "double" Simpson Paradox.
(CMAP28 WWW site: this page was created on 04/18/2013 and last updated on 09/20/2022 12:16:36 -CEST-)
[See all related pictures (including this one) [Voir toutes les images associées (incluant celle-ci)]]
[Go back to AVirtualMachineForExploringSpaceTimeAndBeyond [Retour à AVirtualMachineForExploringSpaceTimeAndBeyond]]
[The Y2K Bug [Le bug de l'an 2000]]
[Site Map, Help and Search [Plan du Site, Aide et Recherche]]
[Mail [Courrier]]
[About Pictures and Animations [A Propos des Images et des Animations]]
Copyright © Jean-François Colonna, 2013-2022.
Copyright © CMAP (Centre de Mathématiques APpliquées) UMR CNRS 7641 / Ecole Polytechnique, 2013-2022.