Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Inflationary Effects?

Author: Robert Hyatt

Date: 10:30:49 07/15/03

Go up one level in this thread


On July 15, 2003 at 09:28:23, Mike Hood wrote:

>On July 14, 2003 at 16:10:09, Robert Hyatt wrote:
>
>>On July 14, 2003 at 13:57:18, Sune Fischer wrote:
>>
>>>On July 14, 2003 at 13:33:26, Robert Hyatt wrote:
>>>
>>>>On July 14, 2003 at 05:14:06, Sune Fischer wrote:
>>>>
>>>>>On July 14, 2003 at 00:00:53, Robert Hyatt wrote:
>>>>>
>>>>>>If you let a new, strong player start at the top, he can establish a higher
>>>>>>rating than if he started at the bottom and stumbled a few times, all the while
>>>>>>dragging everyone's rating above him downward, rather than just jumping on the
>>>>>>top few and leaping over them.
>>>>>>
>>>>>
>>>>>Yes, but only because the Elo formula is flawed!
>>>>
>>>>I don't think it is _flawed_ at all.  It does its job pretty well.  The
>>>>only problem is that the number (rating) is not an absolute measure of
>>>>strength, it is simply relative to the current players in the population.
>>>
>>>I believe it is, for the reason I stated below.
>>>
>>>>>If the formula was working accurately it wouldn't matter where he played, he
>>>>>should end up getting the same 2800 rating whether he played 2100 opponents or
>>>>>2750 opponents.
>>>>
>>>>
>>>>NO, and that is a basic misunderstanding.  In the Elo system, X and Y
>>>>(those being rating numbers) are _meaningless.  X-Y is the only thing
>>>>that carries any information content.  IE the relative difference, not
>>>>the absolute number.  Elo _never_ claimed that the absolute rating of a
>>>>player meant anything, it is only useful when compared to another rating
>>>>to predict the outcome between those two players.
>>>>
>>>
>>>I think it is you who is misunderstanding here, I'm not talking about the
>>>absoluteness at all. The 2800 was _relative_ to the group.
>>
>>Then I don't understand your complaint.  If everybody's rating shifts up by
>>250 points, so what?  That is only a problem if you think the absolute number
>>means something.  Which it doesn't.  I'd just like to see the top number stop
>>increasing year by year, indicating more improvement in playing skill than is
>>actually happening.  IE the top programs are _not_ a Kasparov-like player yet.
>>
>>>
>>>>>It seems to me you want to subject him to a flawed formula by pitting him
>>>>>against lower rated playes so he gets a deflated rating, I don't think that is a
>>>>>good idea.
>>>>
>>>>I simply suggested a way to prevent ratings of 3200 in a couple of years.
>>>>Because so many want the absolute rating to mean something when it doesn't.
>>>
>>>Ahem...
>>>
>>>Are we on wavelength here?
>>
>>I'm not sure.  :)
>>
>>My posts were about the inflation at the top.  And a possible way to prevent
>>or reduce it.
>>
>>>
>>>-S.
>
>One possibility to prevent inflation in the Elo system is to introduce a
>fictional character into the rating pool, "God", with an arbitrarily chosen
>unchangable Elo. Maybe 3000, maybe 5000, who cares? This character would be
>introduced into the rating pool playing games against the top relative player
>and never losing. Maybe alternate wins/draws, or maybe 75%-25%-0% Win-Draw-Loss.
>
>Of course, this doesn't solve the problem of compatibility between a
>computer-only pool and a human-only pool. The pools would still have to be mixed
>to have any overall validity.


That's not half-bad.  Every so often you hold a "mega-tournament" where everyone
plays "god" and loses a high percentage of the games...

:)




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.