On May 1, 7:13 am, David Richerby
wrote:
RookHouse wrote:
I've only played on GameKnot, so that is the only one that I can
speak to.

I've played 287 games over the last few years on there and have a
1724 rating. Just to correlate that figure for you a little bit, I
am definitely NOT a 1724 player. I've had a National Master play
through my games and he evaluated me to be about a 1400 caliber of
player.

That tells you nothing about the strength of the players at GameKnot.
It just suggests that ratings at that site are bigger numbers than
USCF ratings. The rating system only looks at the difference between
ratings so GameKnot could subtract 324 from everyone's rating tonight
(or add 1000 or 1000000) without making any substantive difference to
anything.

Elo-style rating systems are essentially defined by three constants:

1. the rating of an `average' player;
2. the rating difference corresponding
to a given expected score; and
3. `k', which governs how quickly ratings change.

Actually, I have proposed some other essential
parameters, so that the same site or (still better)
the same rating agency would provide you with several
ratings. For instance, the re could be a rating for
everybody but characteristic for expert strength, which
would be especially meanningful for experts. Or on the
highest level: Morozevich is perhaps stronger than
Kramnik against the ordinary granmasters, while Kramnik
is stronger than Morozevich against super-grandmasters.
The universal rating would have Morozevich ahead of
Kramnik, while the super-gm rating would have Kramnik
ahead of Morozevich. Let me add that in the addition
to the universal rating, there would be a class rating
for each parameter, i.e. a continuous spectru, of
ratings, but in practice you would use only ten of them
or so.

Furthermore, when you rate a game, the weight of the
game, as I ahve proposed, would depend on the "rusting
coefficient" of both players, and on the difference
between their skills. The more similar the opponents
the higher the weight of the game. This way flukes
would have less influence on the rating. Unestablished
(new or rusty) players would at first get more classified
among themselves than against the players with the solid,
presently established rating. Indeed, it's hard to tell
the meaning of a game, which involves a Lasker or
Botvinnik or Fischer or Kamsky, when they come back from
their consecutive retirmemnt. But the same goes for
ordinary players, e.g. for older players making
a comeback. The other end of this untried spectrum
are young, rapidly improving players.

It would, perhaps be a good thing if sites chose
dramatically different values for 1. and 2., so
that ratings on different sites would be dramatically
different.

That's artificial. let them do whatever they like.
However, I'd like to see some superb rating
agencies which would provide rating service for
sites and chess organizations (clubs too). You'd
higher one (or more) of them or not, it'd be up
to you. These agencies would do nothing but ratings
and rankings.

I would rank even all sports and activities together:
chessplayers and tennis players etc., so you'd say
that Kasparov was ahead of Agassi but not ahead of
Tiger, etc. It'd be a separate ranking/rating.

However, it's impossible to meaningfuly rank individuals
in the team sports. Playing for a good team, in a team
oriented way (as you should) can make your individual
stats worse than if you played for a weaker team
(not always, it's complicated). However in some stable
sports you may rank/rate the whole team just like
it were an individual: tennis pairs (doubles), bridge...
it is still ify, since some teams are stable and some
arfe not.


That way, people wouldn't be tempted
to try to compare their 1700 rating on
one site with their 1400 rating on another.

I wish you to have no worse worries.

Best regards,

Wlod

Dave.

--
David Richerby Evil Happy Whisky (TM): it's like awww.chiark.greenend.org.uk/~davidr/ single-malt whisky that makes your
troubles melt away but it's genuinely
evil!