KGS Ranking adjustment?

[gennan]

Yes, KGS rankings seem to be pretty far off, and have not been closely tied to appropriate anchors in a very long time. I am trying to gradually tweak the system, without making major shockwaves. Our goal would be for KGS ranks to align more closely with AGA ranks. There are so many servers, and such wide rank variations among them, that picking a standard is very hard. The merit of AGA ranks is that we have extensive records of in person play for hundreds of individuals.

If you find your rank, or someone else's has changed in a way that is totally irrational, let me know here. I can't promise to fix anything, but my actions may cause unintended fluctuations, and knowing about them can help me to make better decisions in the future.

As you're working on aligning the KGS and AGA rating systems, I'd like to point out some critiques to the AGA rating system and other rating systems: see https://wismuth.com/elo/calculator.html in section "Elo per stone in go". I think the author has some valid points there.
(I'm currently in a commission that scrutinizes the EGF rating system and I happened to stumble on that page.)

[jlt]

Do I understand correctly from the graph of the above link that, around EGF GoR 1500,

• 1 EGF rank difference corresponds to a strength difference of about 2.5 stones
• 1 AGA rank difference corresponds to a strength difference of about 5 stones ??

That's hard to believe.

[gennan]

Do I understand correctly from the graph of the above link that, around EGF GoR 1500,

• 1 EGF rank difference corresponds to a strength difference of about 2.5 stones
• 1 AGA rank difference corresponds to a strength difference of about 5 stones ??

That's hard to believe.

That's not really what it says.
Elo is only about winrate, not about ranks or handicap.
GoR is also about winrate, but intended to be adjusted for handicaps (that's why the green line is curved).

So it says around 6k EGF (= 1500 GoR)

• In reality (multicolored lines, blue line), 1 rank difference corresponds to about 55 Elo difference = 58% observed winrate
• In EGF system (green line), 1 rank difference corresponds to 100 Elo difference = 64% expected winrate
• In AGA system (red line), 1 rank difference corresponds to 250 Elo difference = 80% expected winrate
• In KGS system (table below the chart), 1 rank difference corresponds to 50 Elo difference = 57% expected winrate

So if you compare the different systems:

• around 6k EGF (= 1500 GoR), the KGS expected winrates are closer to observed winrates than than EGF and AGA.
• around 8d EGF (= 2800 Gor), the EGF and AGA expected winrates are closer to observed winrates than KGS

So KGS is best for kyu rank gaps, AGA is worst. For top amateurs, EGF and AGA rank gaps are better than KGS rank gaps.

[jlt]

[*]In reality (multicolored lines, blue line), 1 rank difference corresponds to about 55 Elo difference = 58% observed winrate

(...)

[*] In AGA system (red line), 1 rank difference corresponds to 250 Elo difference = 80% expected winrate

250 is about 5 times 55, so doesn't it imply that 1 AGA rank difference amounts to 5 stones?

[gennan]

[*]In reality (multicolored lines, blue line), 1 rank difference corresponds to about 55 Elo difference = 58% observed winrate

(...)

[*] In AGA system (red line), 1 rank difference corresponds to 250 Elo difference = 80% expected winrate

250 is about 5 times 55, so doesn't it imply that 1 AGA rank difference amounts to 5 stones?

No, 100 Elo means 64% winrate, not 1 stone handicap, except around EGD 1d (= 2100 GoR) where the observed winrate for 1 rank difference happens to be about 64% = 100 Elo.

You cannot add or multiply Elo rating differences to add or multiply winrates. It's not a lineair relationship (it's an s-curve as shown in the first chart on that page).

[jlt]

Sorry, I am obviously misunderstanding something. Consider six players

• Player A: about 6 kyu EGF
• Player B: 50 Elo stronger than A
• Player C: 100 Elo stronger than A
• Player D: 150 Elo stronger than A
• Player E: 200 Elo stronger than A
• Player F: 250 Elo stronger than A

According to what you say, one or several of the following statements are false. Which one(s)?

(1) B is about 1 stone stronger than A
(2) C is about 2 stones stronger than A
(3) D is about 3 stones stronger than A
(4) E is about 4 stones stronger than A
(5) F is about 5 stones stronger than A
(6) F is about 1 AGA rank stronger than A.

[gennan]

The consequence of these discrepancies between rating systems and reality is this:

• Promotion of EGF kyu and low dan ranks is too though (so you get rank deflation in that range)
• Promotion of AGA kyu and low dan ranks is even tougher than EGF (so you get even worse rank deflation in that range)
• Promotion of KGS high dan ranks is too easy (so you get rank inflation in that range)

[Bill Spight]

It goes against the grain, I suppose, but I have long believed that ranking in terms of handicap stones, despite the fact that skill in giving or taking handicap stones is variable, it better than ELO ratings or score differences in differentiating skill levels. The main reason for my belief is that, over a large range of skill levels, handicap stones have a roughly linear relationship. E. g., if player A is three ranks stronger than player B, that means that a handicap of 3 stones with White giving komi normally gives each player around a 50-50 chance of winning the game. If player B is four ranks stronger than player C, then if A gives C a handicap of 7 stones, White giving komi, that will also normally give each player around a 50-50 chance of winning. Elo ratings, OTOH, will tell us that A will crush C in an even game.

[gennan]

Sorry, I am obviously misunderstanding something. Consider six players

• Player A: about 6 kyu EGF
• Player B: 50 Elo stronger than A
• Player C: 100 Elo stronger than A
• Player D: 150 Elo stronger than A
• Player E: 200 Elo stronger than A
• Player F: 250 Elo stronger than A

According to what you say, one or several of the following statements are false. Which one(s)?

(1) B is about 1 stone stronger than A
(2) C is about 2 stones stronger than A
(3) D is about 3 stones stronger than A
(4) E is about 4 stones stronger than A
(5) F is about 5 stones stronger than A
(6) F is about 1 AGA rank stronger than A.

I suppose you are asking about true Elo and not mixing it up with rating points of the EGF, AGA or KGS systems.
So first I translate your question to winrates:

• Player A: about 6 kyu EGF
• Player B: 57% winrate in even games against A
• Player C: 64% winrate in even games against A
• Player D: 70% winrate in even games against A
• Player E: 76% winrate in even games against A
• Player F: 80% winrate in even games against A

Also, the data on https://wismuth.com/elo/calculator.html#elo_diff=250 does not contain enough data to answer each of those questions. So I have to consult some other pages, like https://senseis.xmp.net/?EGFRatingSystem and my own pages from 2017 at http://goratings.eu/Probabilities/P_ObservedEGD.
Also, I prefer to talk about ranks rather than stones (the 1st handicap stone is only half a move advantage, so I want to avoid confusion here)

So I would say that everything below is true:

Predicted by KGS
observed winrates form EGF historical data (1 million tournament games over a period of 20 years):
B is about 1.0 rank stronger than A.
C is about 2.0 ranks stronger than A.
D is about 3.0 ranks stronger than A.
E is about 4.0 ranks stronger than A.
F is about 5.0 ranks stronger than A.

Predicted by EGF:
B is about 0.25 ranks stronger than A.
C is about 0.75 ranks stronger than A.
D is about 1.0 rank stronger than A.
E is about 1.5 ranks stronger than A.
F is about 1.75 ranks stronger than A.

Predicted by AGA:
B is about 0.2 ranks stronger than A.
C is about 0.4 ranks stronger than A.
D is about 0.6 rank stronger than A.
E is about 0.8 ranks stronger than A.
F is about 1.0 rank stronger than A.

Now if you ask which system's predictions match observations best, I would say it's KGS (at least around 6k EGF). EGF is pretty far off and AGA is even worse.

[jlt]

Predicted by KGS and observed winrates form EGF historical data:
(...)
F is about 5.0 ranks stronger than A.
(...)
Predicted by AGA:
(...)
F is about 1.0 rank stronger than A.

So you seem to say that

5 KGS ranks = 1 AGA rank

which is hard to believe...

[gennan]

Predicted by KGS and observed winrates form EGF historical data:
(...)
F is about 5.0 ranks stronger than A.
(...)
Predicted by AGA:
(...)
F is about 1.0 rank stronger than A.

Now if you ask which system's predictions match observations best, I would say it's KGS.

So you seem to say that

5 KGS ranks = 1 AGA rank

which is hard to believe...

What exactly is hard to believe?
The AGA system expects a 5k to win 80% of even games against a 6k. I think that's more or less a fact.
In reality, a 6k wins about 57% in even games against a 6k. I think that's also more or less a fact.

The conclusion that 5KGS ranks = 1 AGA rank is only a secondary effect and it's not really working like that in practice.
But it does show that the rating systems that are in use nowadays are not perfect.

[gennan]

Also, the EGF and AGA rating systems have mechanisms that kind of reduce the problem: new players entering the system at mid kyu range and rating resets will bring rating points into the sytem, basically correcting the system somewhat to align it better with reality.

[jlt]

But that doesn't fit observations in https://idex.github.io/go-rank-survey/g ... mean-ranks

which say that for each n between 1 and 13,

n kyu KGS = (n or n+1 or n+2) kyu AGA.

In other words, KGS ranks and AGA ranks are already quite close to each other throughout the kyu range.

Edit: I hadn't read your last message before posting.

[gennan]

It's not exactly that 5 AGA ranks = 1 KGS rank. But it is true that the AGA system has too high expectations on winrates (80% instead of 57% at 5k against 6k over a long time period). So players will tend to be stuck at their rank and promotions in the kyu range are overly delayed by the system, which will lead to deflation of kyu ranks. I would expect improving kyu players' AGA ranks to lag very much behind their KGS ranks, unless there are other promotion mechanisms in the AGA system.

[gennan]

Another effect that can result from this is that strong player won't be able to give the nominal handicap to a weaker player. I don't live in the US, but from the AGA rating system characteristics, I would predict that a 3d AGA rated player will have a hard time giving a 6 stone handicap to a 3k-4k AGA rated player. This trend was mentioned as occuring in France. Do you happen to observe such a trend in the US?