Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Kasparov vs Deep Blue

Author: Amir Ban

Date: 01:53:18 05/31/02

Go up one level in this thread


On May 30, 2002 at 20:16:47, Robert Hyatt wrote:

>On May 30, 2002 at 17:59:35, Amir Ban wrote:
>
>>On May 30, 2002 at 13:34:25, Robert Hyatt wrote:
>>
>>>On May 30, 2002 at 13:19:45, Dann Corbit wrote:
>>>
>>>>On May 30, 2002 at 13:15:59, Jerry Jones wrote:
>>>>
>>>>>Does anybody know what the highest official ELO rating according to FIDE is that
>>>>>was ever attained by a human, Kasparov that is.
>>>>>Is it possible that a few years ago his rating was a few points higher ?
>>>>>If Kasparov had declined to play Deep Blue, would this have influenced his
>>>>>rating ?
>>>>
>>>>You can add one million points to his ELO rating if you like.  Or subtract them.
>>>> Just be sure to do it to everyone else and it is perfectly valid.
>>>>
>>>>ELO figures are only valuable as differences within a pool of players who have
>>>>had many competitions against each other.  The absolute numbers mean absolutely
>>>>nothing.
>>>
>>>
>>>This is a continual problem.  :)  32 degrees F means one thing.  32 degrees C
>>>means another thing.  32 degrees K means another thing.  No way to compare
>>>today's 2850 rating to the ratings of players 40 years ago.
>>
>>It is perfectly sensible to compare ratings of 40 years ago and even more to
>>today's. That's because at no point in time did the pool of players change, with
>>an old group completely replaced by another. The ratings are measured against
>>the field, which changes continuously, and provides continuity of the ratings.
>
>statisitically this is not true.  First, there is the spectre of rating
>inflation caused by various problems.  It has been seen in USCF more than
>once, on the chess servers, etc.  It also happens within FIDE.
>
>Second, taking two populations and producing ratings for them works well.  If
>you try to say that the two populations are "joined" because a few from one
>group played a few from the other group, statistically that claim is invalid.
>
>You can't just have a few "common people".  You need them _all_ to be common
>to both pools to be able to compare the ratings.
>

This is a reductio ad absurdum, as it follows that ratings in Germany cannot be
compared to ratings in France, and the whole rating notion is not useful.


>>
>>So, even if Kasparov and Fischer never met (certainly Kasparov 2001 never met
>>Fischer 1972), they had many common opponents, whose ratings where themselves
>>determined by common opponents, etc.
>
>
>Actually they didn't.  Since they weren't playing in the same "era" the age
>of those "common opponents" was a classic variable that influences the rating
>difference.
>
>
>> There's no more reason to assume that
>>ratings in time are incomparable than to assume that ratings in the US and in
>>Europe are incomparable, for, although most games are in one region, there are
>>enough interregional games to give the ratings worldwide meaning.
>
>Again, you can believe what you want, but sampling theory finds problems
>with this sort of sampling approach.  Particularly when one geographic region
>has a higher than usual number of top GM players, while another region does
>not.  That causes significant rating variances...
>

You can also believe what you want ...


>
>
>>
>>There are random fluctuations in the rating standard, because it's all
>>statistics, but the numbers are large, and I'm not aware of anything that would
>>cause ratings to systematically drift in any direction (actually this can be
>>simulated effectively, by creating a random population of players and slowly
>>change the pool over time and see if averages drift).
>
>This has been done.  It does...  it depends on where you start off "brand new
>players".
>
>
>>
>>Most strong players agree that the level of play is higher than 30 years ago,
>>and that's a good enough reason why today top ratings are higher.
>
>Wrong.  If the level is uniformly higher, then the ratings should be _unchanged_
>since everybody is equally better.  And the difference between two ratings is
>used to predict outcomes, not the absolute value of a single rating.
>

There's nothing to keep ratings constant if the field improves. A new and
stronger generation of players would get higher ratings vs. the older
generation, and as the older generation gradually drops out of competition,
averages go up.

Amir


>>
>>Fischer, Alekhine, Capablanca are of course classics, but so are Johnnie
>>Weissmuller and Jessie Owens, who would be today's also-rans. It is tempting to
>>say that this is because today our clocks run slower than in their time, but
>>they don't.
>>
>>Amir



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.