Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Rating swings on ICC

Author: Bruce Moreland

Date: 15:09:45 08/02/98

Go up one level in this thread



On August 02, 1998 at 14:32:55, Don Dailey wrote:

>I think the effects of inflation or deflation would balance out
>in the long run.  But if rapidly improving players tended to
>prefer high K values then the pool would tend to inflate.

If you allowed people to select their own K, within one week you'd have cases
where some over-rated bozo helped his accurately-rated bozo friend get an
inflated rating by setting his own K low, his friend's K high, and just playing
chess.

If you give people more ways to manipulate the system, they'll do it like crazy.

>  I think
>the right way to control inflation or deflation is to pick out
>a few active players who have well established Fide ratings and
>make very minor adjustments over a period of time to keep these
>in line.

A GM can be 3000 one day and 2650 the next day.  You get some guy who eeks
himself up a hundred points by playing 3 0 against over-rated people, and the
same guy drops 250 points in a "just let me win one game against this stupid
computer so I can disconnect and go to bed" death-match.

So pinning the ratings of others on a few of these guys would be like trying to
measure financial health in the US by making an index composed of the likes of
Iomega, Netscape, and Amazon.com.

>  For instance when anyone wins, they can be given slightly
>less points (like a fraction of a point) until the the pool was
>normalized if you wanted to deflate the rating pool.   How do
>they determine the ratings anyway?  Did they start with ratings
>of known players?   There is bound to be inflation or deflation
>over time anyway, in any system that is open like this.

They use new players as a balancing mechanism.  The goal is to keep the average
rating at some value.  They do this by fudging the ratings of new players while
they are in the provisional period.  If the average rating is a bit above the
desired value, someone who plays their first game against a 1600 and wins might
find themselves with a rating of 1990 rather than the expected 2000, but if the
average rating is below the desired value, they might end up at 2010.

This sounds like it wouldn't work but it does, the average rating on ICC is
pretty stable.

For some reason though, we have gotten a lot of expansion at the high end.  It
used to be that 2600 was an incredible rating, achievable only by grandmasters
after a heck of a lot of eeking (playing down a huge amount).

Now a rating of 2600 will put you at 165th on the server, amongst active
players.

bruce



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.