Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: time for another ssdf rating adjustment.

Author: Maurizio De Leo

Date: 15:36:46 07/13/03

Go up one level in this thread


>>ok maybe just -100 for all.
>
>That was done one time already, and cheapened ssdf in my eyes, because the more
>truthful way would have been to have reduced every rating with seperate
>calculation for each program, by a certain (same for each) percentage.
>
>That would have been much more honest as well as accurate, and it's surprising
>that chess players aren't intelligent enough to see this!
>S.Taylor

Maybe you should reconsider calling stupid the chess players and think about
what you said.

Elo rating are calculated so that what matter is the DIFFERENCE between them.
The difference between two ratings is related to the probability of the outcome
of the match between two players.

You seem to think that the ABSOLUTE VALUE of elo is a misure of "strenght" of
the player.
In fact it is not so :

1) a 2000 elo player isn't 2 times stronger than a 1000 player
2) a 3000 elo player isn't 1.5 times stronger than a 2000 player.

Actually what elo points mean is that in both case 1 e 2 the player rated higher
will have a winning expectancy of 99.x (I don't know the exact number).


Bottom line :

The correct way of reducing the rating in an elo list is EXACTLY to reduce all
the elo by a fixed amount and NOT to scale all the rating with a certain
percentage.

If in case 1 the two player become 1900 and 900 the winning expectancy between
them remains the same. If they become 1900 and 950 as you proposed this is not
longer true.


Thanks to whom read until here :-)

Maurizio



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.