Ranking system for chess clubs

Factual Questions
1

It doesn’t have to be chess. But I’m looking for the details of a ranking system that would take into account playing agianst players of varying skill.

For example, Let’s say Abe and Bob are equally ranked… perhaps they both have 1000 points. They play a game of chess (or Go, or checkers, or dominos - whatever). The winner might get 100 point added to his total, and the loser would lose 100 points.

Now comes Charlie, a master player with 2300 points. He plays Abe. He wins, and only gets 10 points added to his score. If Abe should happen to win, though, he’d get a huge chunk of points because he beat someone with such a higher total.

See what I’m asking? The system would keep track of wins and losses, but allow for the fact that beating a better player is more meaningful than beating a worse player… and it would let two people in the same system compare their own relative standings even if they never played each other, or even if they never played common opponents.

What would the specific rules/formulae of such a system be?

2

I don’t pretend to know all the math, but is that really any different from the Elo rating system?

3

See the Elo Rating System.

4

Let’s see, I beat you by one minute. Taking into account your roughly 1500 posts, and my 800ish… Do the math… Carry the one…

Ok. I now should have about 1067 posts. Could a mod please take care of that for me?

5

I understand the ELO system. (I have run international chess tournaments, which included calculating the required score to qualify for titles such as Grandmaster.)
Do ask me if you want details.

There is also the English Chess Federation system, but which is only designed for assessing an entire season’s results.
An experienced player has a grade.
Roughly speaking:
50 = beginnner
100 = club player
150 = county player
200 = national player
250 = world class player

You take your season’s results as a percentage. (So 10 wins, 5 draws and 5 losses would be 62.5%.)
You take the average of your opponent’s grades. (Suppose it’s 145.5)
The player’s new grade would be the % difference from 50, plus the average of his opponent’s grades.
In this case, it would be 62.5 - 50 + 145.5 = 158.

There are simple adjustments when players rated more than 40 points apart meet.

6

For an N-player generalization of ELO (more or less), check out Microsoft Research’s TrueSkill system.

7

I’ve read - or at least skimmed - the links.

What I’m not really seeing is the precise method.

I start out with six players, each with 1000 rating points.

A plays B and wins.
A plays C and wins.
A plays D and wins.

B plays C and wins.
B plays D and loses.
C plays D and loses.

E plays D and wins.

F plays D and loses.
F plays B and wins.

Who has what points now?

8

Looks like the Trueskill page (**Taran ** provided a link) has formulas and such and at the bottom there is a link to a ranking calculator. (http://research.microsoft.com/mlp/trueskill/RankCalculator.aspx)

I won’t pretend to understand much of it. But there it is.

9

Bricker,

Here’s a simple model from what I remember back in my chess playing days. This might not be entirely accurate and I welcome corrections, but hopefully it serves to answer your question

Let’s say you come into a tournament unrated. For the first ten matches they’ll rate you using an initial simple computation which then produces your provisional rating.
If you win against someone, you get a score of 400 above his rating.
If you lose against someone, you get a score of 400 below his rating.
If you tie, you get exactly their rating.

So say you enter an 8 round tournament and win matches against opponents with established ratings of: 1267, 1375, 1589, and 1601

You lose matches against opponents with established ratings of: 1567, 1703, 1476, and 1698

You’ll end the tournament with a provisional rating of 1534.5. Notice that after the first four rounds, your rating was at 1858. Provisional ratings swing about more wildly than a blind samurai.

It’s also a general flaw with provisional ratings: they’re not always entirely accurate. Imagine you’re unrated and play in a tournament full of players with ratings of 2200 and above. No matter how badly you do, even if you forget what color the queen should start on or what the “horsey” does, you’re going to come out of the tournament with a pretty respectible rating of at least 1800, maybe more.

But if you play in a tournament with all different skill levels, the provisional rating will do a pretty good job of narrowing down the window to your true rating. Then what?
Well now it switches over to a 32 point scale and the number of points you can win or lose is determined by the ranking of your opponent. You’ll never win or lose more than 32 however.

If you’re at 1800 and you win against a 1200 that’s no big deal. You were expected to win. You’re not going to gain more than a handful of points. If you lose, or even tie, however, you’ll lose a considerable amount of points, maybe upwards of 25.
If you and your opponent are both at 1500, then winning gives you 16 points, losing loses 16 points from your rating and ties don’t change anything.

It’s why grandmasters don’t like playing players of a considerably lower skill level. The points they win are negligable (1-2 points), but losing is a huge blow to their overall rating (31-32 points).

Overall, the more you play, the more accurate your rating becomes.

10

The two rating systems I’ve mentioned both operate best where most players already have a rating, and so new players can be added more easily, because there already is a baseline to assess their results.

If you intend to start from scratch, as per above…

I’ll have a go with the English rating system (though as I said, it’s really intended for a season’s worth of games). Note that the ratings do not change after every game.

A has scored 100% against an average of 1000.
New rating is (100-50) + 1000 = 1050.

B has scored 25% against an average of 1000.
New rating is (25-50) + 1000 = 975.

C has scored 0% against an average of 1000.
New rating is (0-50) + 1000 = 950.

D has scored 60% against an average of 1000.
New rating is (60-50) + 1000 = 1010.

E has scored 100% against an average of 1000.
New rating is (100-50) + 1000 = 1050.

F has scored 50% against an average of 1000.
New rating is (50-50) + 1000 = 1000.

You are welcome to note that A has played 3 games to E’s 1, although they have the same rating. With such small numbers of games, I think any rating is going to be a rough estimate.

11

First you have to find such a tournament!
Many events have rating categories, so that the 2200+ play amongst themselves, then there are events for 1800 - 2199, 1400 - 1799, and anotehr for everyone below 1400, or unrated.

I’ve played a lot of grandmasters over the years, with an ELO rating hovering around 2300. Grandmasters expect to beat players of my strength 99.9% of the time.
In turn I would have no fear facing a 2000 player. Obviously you take them seriously (that’s what top chess players do!), but there will be something you have they don’t (e.g. more opening knowledge / faster calculation / reliable endgame technique) that sees you home.