Author: Frank Schubert
Date: 00:40:35 01/18/00
Hi everybody,
there were some confusions produced by the fact that the Elo calculation in
Fritz 6 of Enriques Cadaqués tournament produces different results compared to
my own calculations. I looked on the problem and came to the conclusion that
the Elo calculation in Fritz 6 doesn't work correctly. Here is a very
instructive example why:
I took Enriques tournament after round 6 (120 games). At this time all programs
played 40 games besides Hiarcs and Tiger which only finished 20 games. So these
two programs only played against one opponent so far:
Hiarcs against Shredder (reaches 57.5 %, i.e. +52.5 Elo points after official
FIDE formula), and
Tiger against Century (reaches 60 %, i.e. +70.4 Elo points)
Now if the Elo calculation is working correctly the Elo differences in the
rating list between Hiarcs/Shredder and Tiger/Century, respectively, should be
equal to the Elo differences given above. Let's check it:
This is the result of the Fritz 6 Elo calculation ("generate Elo start list",
offset = +250):
1 Junior 6.0 2661 40
2 Hiarcs 7.32 2648 20
3 Fritz 6a 2643 40
4 Shredder 4 2601 40
5 Nimzo 7.32 2586 40
6 Rebel-Tiger 12.0e 2573 20
7 Century 2486 40
Elo difference between Hiarcs and Shredder = 47 (should be 52.5 points).
Elo difference between Tiger and Century = 87 (should be 70.4 points).
Both results are obviously not correct.
Here are the results of my own calculations (ELOStat 1.0):
Program Elo + - Games Score Av.Op. Draws
1 Hiarcs 7.32 : 2720 157 138 20 57.5 % 2667 25.0 %
2 Junior 6.0 : 2701 101 74 40 60.0 % 2630 45.0 %
3 Shredder 4 : 2667 80 106 40 43.8 % 2710 37.5 %
4 Fritz 6a : 2607 96 86 40 63.7 % 2509 37.5 %
5 Nimzo 7.32 : 2593 76 102 40 41.2 % 2654 42.5 %
6 Rebel-Tiger 12.0e : 2496 153 79 20 60.0 % 2425 60.0 %
7 Century : 2425 78 92 40 32.5 % 2551 45.0 %
Elo difference between Hiarcs and Shredder = 53 (should be 52.5 points).
Elo difference between Tiger and Century = 71 (should be 70.4 points).
You can see that these results are in a good agreement with the expected
values (the small differences are caused by numerical roundings).
Consequently the results of Fritz seem to be incorrect and all I wanted to show
with this comparison is that you can still trust my future rating lists :-)
Hope you will do so,
Thanks
Frank
P.S. There are some indications that the ChessBase7 Elo calculation
is working fine. Maybe we can find a user of CB7 who can test it. But be
sure that you take only the first 120 games of the database. Thanks in advance.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.