Re: Fw: [pbmserv-dev] ratings in multiplayer games

[Lyman Hurd <lhurd@yahoo.com>]

Richard informed me a while ago that he was trying to
overhaul the ratings system and that the primary flaw
people found with the current system was just this
behavior (ratings going down because of a win). 
Although, I personally found it more curious when I
started playing Zertz and lost seven straight games to
find my rating higher each time :-)...

Please note that this is a different phenomenon from
the across-the-board correction to re-center the norm
of the distribution.

The rationale behind the current behavior was that
they did not want a player to get one lucky win and
then lock in a high provisional rating by playing a
bunch of beginners.  My solution takes care of that
(as these games would not count in any way including
towards the twenty games you need to become
established).  As the drop in ratings is predictable,
the alternative is for one player not to have played
the game at all.  There is no statistical
justification to use a win as evidence that your
rating should go down.

Cheers,

Lyman


--- David J Bush <twixt@cstone.net> wrote:
> 
> Sorry, I should have directed this to pbmserv-dev:
> 
> | | By the way, has there been any consideration to
> my
> | | suggestion (in two player games) to simply make
> any
> | | game in which the effect of a win would be to
> lower
> | | the winning players rating, simply unrated? 
> This
> | | would allow people to play teaching games
> without fear
> | | (though I believe this only affects provisional
> | | players, as it has been noted some games are a
> lot
> | | harder to complete 20 of than others).
> | 
> | In order for the ratings to have value, they
> should
> | represent an accurate statistical measure of the
> player's
> | performance. eliminating the effect of provisional
> player
> | wins against lower rated players, would arguably
> lead to
> | ratings inflation.
> | 
> | However, as has been pointed out, under the
> current
> | system, provisional players can lose rating points
> even
> | when they defeat HIGHER rated players. This is
> because
> | of a bias towards some median value, 1720 I think.
> | I'm not sure what the statistical justification of
> this
> | bias is. It applies only to provisional players.
> | 
> | There are other rating systems; Glicko for
> example, which
> | does not have a separate provisional system. I
> don't
> | think a player can lose rating points for a win
> with
> | Glicko. Please correct me, if I'm wrong!
> | 
> | David
> 
> 
> To unsubscribe, send a message to esquire@gamerz.net
> with
> 	unsubscribe pbmserv-dev@gamerz.net
> as the BODY of the message.  The SUBJECT is ignored.
> 


__________________________________________________
Do you Yahoo!?
Y! Web Hosting - Let the expert host your web site
http://webhosting.yahoo.com/