Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: I just don't get this ...

Author: Bob Durrett

Date: 16:46:23 01/05/04

Go up one level in this thread


On January 05, 2004 at 19:17:04, Christophe Theron wrote:

>On January 05, 2004 at 16:50:03, Bob Durrett wrote:
>
>>On January 04, 2004 at 23:04:55, Christophe Theron wrote:
>>
>>>On January 04, 2004 at 11:10:41, Bob Durrett wrote:
>>>
>>>>On January 04, 2004 at 11:00:31, Dan Andersson wrote:
>>>>
>>>>> I admire your persistance. I guess most of us that have a mathematical
>>>>>statistics education got tired explaining things after the first thread or so.
>>>>>
>>>>>MvH Dan Andersson
>>>>
>>>>I, too, have a "mathematical statistics education."
>>>>
>>>>What bugs me is that all of the CCC bulletins seem to suggest that those who run
>>>>and evaluate tournaments look only at the win/loss statistics.  There is
>>>>considerably more information in a game score than just the final game result.
>>>>
>>>>Throwing away useful information is what I call "blind adherence to statistics."
>>>> One needs to rise above one's formal education and supplement it with good
>>>>thinking.
>>>>
>>>>: )
>>>>
>>>>Bob D.
>>>
>>>
>>>
>>>The games themselves do not contain more information about the relative strength
>>>of the opponents than the bare winning percentage of the winner.
>>
>>I would like to try to offer a counter-example:
>>
>>Suppose there is a match between two chessplayers, A and B. [human or
>>otherwise].  Suppose also that N games are played in the match.
>>
>>In this example, endgame of type #1 will occur in the match a finite number of
>>times, assuming that each game in the match has a finite number of moves.
>>[Adjudicate after 60 moves].
>>
>>If endgame of type #1 occurs rarely in practice [as seen in a much larger
>>collection of games such as Megabase 2004], then the best estimate of the number
>>of times that type of endgame would occur in this match would be small, likely
>>much much smaller than N.  However, suppose an uncharacteristically large number
>>of occurrences of endgame type #1 occurred in this particular match.
>>
>>Suppose also [since I'm the one dreaming up this example] that every game in
>>which endgame of type #1 occurred, chessplayer A handled that endgame properly
>>but chessplayer B obviously didn't have a clue about how to play that endgame.
>>
>>Then post-mortem analysis would have to conclude that in a more normal sample,
>>where the number of occurrences of endgame type #1 were more typical, the
>>percentages of wins and losses might be different, becoming more favorable to
>>chessplayer B.  [Chessplayer B was handicapped by the abnormal number of
>>occurrences of endgame type #1 which chessplayer B cannot play well.]
>
>
>
>There is no reason to assume that...
>
>If you start to assume this or that, we are not talking about a measure anymore.
>
>A measure of the elo difference between two players does not involve anything
>else than the number of wins, draws and losses. Or the winning percentage.
>
>The information contained in the games themselves can be used for many things,
>but not to fix the measure of the elo difference...

I am desperately trying to understand what it is you are saying.  Let me say
what I feel you are saying and then please correct me.

If one were to go to the FIDE website and look up a rating for someone, like
Garry Kasparov for example, one would find a number.  We know how that number
was computed by FIDE.  The method for computation is published and based on
Arpad Elo's pioneering work.  The computations use wins, draws and losses and
nothing else as you say.  That is true.

Perhaps your point is that the number, called "rating," found at the FIDE
website is a measure of wins, draws, and losses and nothing else.  That makes
sense inasmuch as the computations used only data on wins, draws and losses.

Perhaps we have a symantic [or language] problem which is making it difficult
for us to communicate.

Some people use ratings as a measure of something else.  They use ratings as a
measure [or estimate] of the individual's future chess performance.  The measure
of win/draw/loss performance is itself a measure of something else as well and
that "something else" is the chessplayer's future.  We seek to estimate his
future [as yet unknown] performances.

We may wish to predict the average win/draw/loss performance of Garry Kasparov
in his next tournament or match.  ANY INFORMATION AT ALL which would help us to
make that prediction should be used if the very best prediction is desired.

It is true that a record of Garry Kasparov's past wins, draws, and losses would
help us in our endeavor to make a good prediction of Garry Kasparov's future
win/draw/loss performance.

However, it does not follow logically that NO other information [other than a
record of his wins, draws, and losses] could help to improve the prediction, or
the estimate of future performance.

Consider a trivial case.  Suppose it were known that Garry had come down with a
particularly virulent form of SARS.  [Let's hope that never happens.]  In that
case one could say that our best estimate of his performance is dismal, and that
he may not even be able to play at all.

This is an example of how "other information," other than his win/draw/loss
record, can be used to improve our estimate of his future performance.

I see no reason why estimates of future performance should be based solely on
win/draw/loss records.

Bob D.







>
>
>
>    Christophe
>
>
>
>
>>Since "ratings" are merely estimates of performance, then the information
>>relating to endgame type #1 in the match could be used to obtain a better
>>estimate of the ratings or, in this case, the relative ratings.
>>
>>Generally, the more information that is available, the better can be the
>>estimates.  Obviously, the information must be used properly.
>>
>>- - - - - -
>>
>>Incidentally:  During post-mortem analysis, tens, hundreds or even thousands of
>>opening, middlegame, and endgame "types" could be identified in the games of the
>>match and the same comparisons made with a large population.  Doing so would
>>permit a better estimate of the ratings or rating difference.
>>
>>- - - - - -
>>
>>
>>Bob D.
>>
>>
>>>
>>>That should not be forgotten.
>>>
>>>
>>>
>>>    Christophe



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.