Item # 17 - Modifications to the Can. Rating System / Report of the Rating Auditor

[Christopher Mallon]

That was one of the problems I saw with it Paul, but a way around that might be either a) most recent 24 games or b) all games in previous 12 or 18 months, whichever is the lower number of games.

[Paul Leblanc]

Chris, I think you have found the solution. It is common sense and easy to apply. I hope the Rating Auditor is tuned in.

[Bob Armstrong]

Hi Chris and Paul:

If you think you have an improvement to the system, I'd suggest you e-mail the proposition to Bill - he has not yet attended this meeting.

Bob

[Fred McKim]

This is a variation on changing the k (memory) factor.

The basic principle is that the players likely to be improving fast, will have their ratings move faster.

For example junior gains 8 points from a game adult opponent loses only 4 points.

So an easier solution would be to base juniors memory on 15 games instead of 25 games. That means each win against an evenly rated player would be 26.7 points instead of the present 16.

Once juniors got to say 2000, they could drop back to a normal k value.

[Stuart Brammall]

As many of you are probably aware there is currently a contest being held to see who can develop a rating system to out predict the Elo system--- perhaps changes to the rating system should wait for the results.

[Bob Gillanders]

Okay, so the problem is that some juniors are underrated. And until their rating catches up to their real strength, their opponent is getting screwed.

How about this idea.

George (old guy) - rated 1800
Johnny (hot shot kid) - rated 1300 (but really much stronger)

George and Johnny play each other in tournament. Johnny wins!
Johnny's tournament performance rating = 1800

For their game:

George only loses 16 points, not 30 points. Calculated using Johnny's performance rating, not his pre-tournament rating.

Johnny gains 30 points! (same as now)

[Christopher Mallon]

Bob, that's similar to the other ideas proposed, except it's setting his provisional game limit to however many rounds were in the current tournament, sortof.

The problem is, what if Johnny is a typical youth player and is wildly inconsistent? Round 1 he crushes George, and the following four rounds he plays well but ends up making big mistakes and losing - bad timing/luck. His performance for the tournament - 900. So George loses 32 points now, and those who beat Johnny don't gain much at all despite him actually being rated 1300.

Better to use a larger (but still recent) sample size.

On another note, should we consider following FIDE's lead and NOT rating games played against unrated players? (And possibly Provisional players also)? It would be rated for one person (the unrated) but not the other.

[Paul Leblanc]

Chris, just speaking for myself I'd prefer to have all my games rated. I don't mind playing the hot-shot kid if I know that his recent provisional rating is going to be used to indicate how many rating points I'm going to win or lose.
The game doesn't seem serious if it's not rated for one player.

[Christopher Mallon]

The problem is you don't know - provisionals are calculated first, so you actually don't know what his rating will be when your game gets rated. Also, what about unrated opponents?

[Bob Gillanders]

Chris, going back to my example.

If the kid loses his other 4 games with blunders etc. and has a performance rating of 900, well then is he really stronger than his 1300 rating?

To calculate George's rating, I would use the higher of Johnny's current rating of 1300 and his performance rating of 900.
Either way, poor George loses 30 points.

After all, the system needs some credibility. You can't just assume every kid is better than his/her rating!