"David Richerby" wrote in message
...
Wlodzimierz Holsztynski (wlod) wrote:
Ralf Callenberg wrote:
Most players, including amateurs don't regard ratings as fun but
take it quite seriously.

Serious rating has to cost money. Amateurs who want a serious
rating would have to pay for rating quite a bit, say $10 per game
instead of 1 cent per game

Why? The `serious rating' that I have from the ECF (formerly BCF)
costs something like 0.40ukp/game to cover the administrative costs.
Why should I have to pay $10 per game just to have some guy enter data
into a computer and have it calculate a number for me?


Why all this is important? Because it'd make the open system
cheaper, something like a penny per game, per player.

What exactly will make it cheaper?

A multipurpose rating, which had to serve also professionals, would
need all kind of constructions and precautions which amateur system
does not need. E.g. in a purely amateur system I would not worry
that someone is using a PC program on his/her 2nd computer.

Are you proposing extending the rating system to every game ever
played? I don't bring a `first computer' to my tournament games, let
alone a second. I don't want my off-hand games rated: apart from
anything else, I want to be able to drink beer and chat with my
friends while playing.


On the other hand, the professionally rated games would have to be
played under well understood and controlled circumstances.

Principally this is already the case with the Elo-rating.

It's completely independent of the formulae (Elo's or otherwise) that
are used to compute the ratings.


As you see, for any reasonable rating function, the total amount of
$$ won by any subgroup of pplayers depends mainly of their
performance against other players and almost does not depend on the
results between the conspiring players (or else the rating function
is useless and should not be used).

This makes no sense whatsoever. The rating function and the prize
money allocation are completely separate. Whether or not Elo's
formulae (or anyone else's) for computing ratings are `reasonable'
does not depend on how prize money is allocated. Are you saying that
Elo's formulae would somehow become `unreasonable' if all the prize
money were given to the player with the most Z's in his name?


Elo's definition of performance might
be "reasonable" for many situations,
and certainly has *some* general predictive
value, but it has not been demonstrated
optimal.

Example. Some tournaments, to reduce
draws, have been played with the
experimental "0-1-2-3 BAP" scoring
system. (W draw, B draw, W win, B win).
It is quite possible to have a good "BAP"
performance and a poor "ELO" performance.
Don't you think, using your knowledge
of math, you could come up with a more
predictive rating function for games played
under these conditions? I know I could.

This is probably peripheral to Ralf's point.
I await his full exposition.