Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: A theory of ratings drift for the SSDF

Author: Sune Fischer

Date: 01:54:08 04/13/02

Go up one level in this thread


On April 12, 2002 at 23:50:17, Robert Hyatt wrote:

>On April 12, 2002 at 06:27:29, Sune Fischer wrote:
>
>>On April 12, 2002 at 00:16:35, Robert Hyatt wrote:
>>
>>>It _must_ affect both the rating of other players, and the "spread" to some
>>>small degree.  A drop of water hitting the pond changes the depth everywhere
>>>as the ripples propogate around and reflect...  Remember that the spread between
>>>two player's ratings is a statistical average of how the two players do against
>>>all other players in the pool.  Adding a new player can change this.  IE a
>>>player joined our local club back in 1970, and he was rated 300 points above
>>>me, yet I won the majority of games against him because our "styles" gave me
>>>an advantage...  that obviously changed the spread between me and other players
>>>in the club, yet I did no better (or worse) against them after the new player
>>>arrived...
>>>
>>>
>>
>>Well as a short term effect this will send ripples up and down the ranks, but
>>when they have decayed and you have gone back to playing games with the other
>>players, you will share some of your won rating with them, or they will also
>>beat the new guy.
>>The lasting effect is just an indication of the usual uncertainties that applies
>>to ratings, i.e. you where actually stronger than you thought if only you were
>>able to increase your rating.
>>
>>The overall spread can't really change if the same formula is used, the pool
>>will expand or subtract to fit the formula eventually.
>>
>>-S.
>
>\
>If chess were perfectly transitive that would be true, but it isn't.  I've
>seen many cases where A beats B regularly, B beats C regularly, and C beats
>A regularly.

Of cause, this happens quite often actually, e.g. you can win
a tournament even tough you lost a game.

>When you change the pool, you can change _everything_.  Because
>a rating within a pool factors in _all_ players within that pool.  That's why
>absolute ratings mean nothing, and the "spread" between players in two different
>pools means even less...  And finally, if you transplant player A and B
>from one pool to another both their raw ratings and the spread between them
>can change, easily.

Everything will change, but the correlation is high enough to convince me that
the changes are only noise, besides humans sometimes have a bad day and do not
perform at a constant level.

-S.



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.