- Blind mode tutorial
lichess.org
Donate

All about the Elo Rating System

#2

FIDE should probably adopt Glicko-2 like lichess does.
Glicko-2 requires more calculations, but converges more quickly.
Elo was simplified to enable manual calculations around 1970 when computers were not yet widely available.
Elo assumes a constant rating deviation RD, the same for all players and uses only 3 different K-factors: 40, 20, or 10.
Glicko-2 calculates the variable RD and K for each player at all times.

FIDE should probably adopt Glicko-2 like lichess does. Glicko-2 requires more calculations, but converges more quickly. Elo was simplified to enable manual calculations around 1970 when computers were not yet widely available. Elo assumes a constant rating deviation RD, the same for all players and uses only 3 different K-factors: 40, 20, or 10. Glicko-2 calculates the variable RD and K for each player at all times.
#3

The reason the 400 point rule existed was for high rated players who happened to face an extremely lower rated player on rarer occasions (e.g. in open tournaments). This would reduce the rating loss in the case of a draw or loss, and also increase rating earned.

I don't think the rating loss on loss or draw was very important factor here, rather the gain on a win. In practice, we are talking about players with K=10, i.e. with the 400 point rule the deltas are +0.8 / - 4.2 / -9.2 while at the extreme you can only get to 0.0 / -5.0 / -10.0. From human point of view, losing 9 or 10 points of rating is not that different; on the other hand, it's quite frustrating to play a game, knowing that you don't gain anything if you win and only risk losing your rating. Always getting at least 0.8 for a win was IMHO the important part.

What I find very questionable is the effect of 400 point rule from the point of view of (much) weaker players: you play a game where you stand almost no chance and know that when you lose (which is almost sure), you still lose 1.6 points of rating (or even 3.2 with K=40). And with current version of the rating regulations, it's even worse: when a 2700 rated player faces a 1900 rated one, the stronger player does not get anything for the win but the weaker one still loses 1.6 or 3.2.

> The reason the 400 point rule existed was for high rated players who happened to face an extremely lower rated player on rarer occasions (e.g. in open tournaments). This would reduce the rating loss in the case of a draw or loss, and also increase rating earned. I don't think the rating loss on loss or draw was very important factor here, rather the gain on a win. In practice, we are talking about players with K=10, i.e. with the 400 point rule the deltas are +0.8 / - 4.2 / -9.2 while at the extreme you can only get to 0.0 / -5.0 / -10.0. From human point of view, losing 9 or 10 points of rating is not that different; on the other hand, it's quite frustrating to play a game, knowing that you don't gain anything if you win and only risk losing your rating. Always getting at least 0.8 for a win was IMHO the important part. What I find very questionable is the effect of 400 point rule from the point of view of (much) weaker players: you play a game where you stand almost no chance and know that when you lose (which is almost sure), you still lose 1.6 points of rating (or even 3.2 with K=40). And with current version of the rating regulations, it's even worse: when a 2700 rated player faces a 1900 rated one, the stronger player does not get anything for the win but the weaker one still loses 1.6 or 3.2.
#4

A small nitpick: the K-factor does not drop to 20 after 30 rated games but after completing events with at least 30 rated games, i.e. if you play your 30th rated game in the middle of a tournament, your update is still calculated with K=40 for the rest of it and K=20 is only applied after that.

A small nitpick: the K-factor does not drop to 20 after 30 rated games but after completing **events with** at least 30 rated games, i.e. if you play your 30th rated game in the middle of a tournament, your update is still calculated with K=40 for the rest of it and K=20 is only applied after that.
#5

@tpr said in #2:

FIDE should probably adopt Glicko-2 like lichess does.

Sadly, they won't - or at least not in the near future. Reading between lines in the Sonos' analysis, it seems obvious that he is well aware that a more modern rating system with dynamic reliability evaluation would be much better solution than ad-hoc hacks piled on top of the old one. But FIDE is afraid that people would not like a more complex system with what they call "hidden parameters" (RD and volatility); in other words, they prefer a system which provides worse results but is simple enough for common folk to be able to calculate (and check) their rating updates by hand. And I'm afraid any system providing significantly better properties than Elo would also be too complex to fulfill that requirement.

@tpr said in #2: > FIDE should probably adopt Glicko-2 like lichess does. Sadly, they won't - or at least not in the near future. Reading between lines in the Sonos' analysis, it seems obvious that he is well aware that a more modern rating system with dynamic reliability evaluation would be much better solution than ad-hoc hacks piled on top of the old one. But FIDE is afraid that people would not like a more complex system with what they call "hidden parameters" (RD and volatility); in other words, they prefer a system which provides worse results but is simple enough for common folk to be able to calculate (and check) their rating updates by hand. And I'm afraid any system providing significantly better properties than Elo would also be too complex to fulfill that requirement.
#6

@mkubecek said in #4:

A small nitpick: the K-factor does not drop to 20 after 30 rated games but after completing events with at least 30 rated games, i.e. if you play your 30th rated game in the middle of a tournament, your update is still calculated with K=40 for the rest of it and K=20 is only applied after that.

Thanks for pointing that out, I edited the blog to reflect that. I also clarified that the 400 point rule isn't cancelled, but is only applied to players below 2,650. 2,650 and above will not have the 400 point rule.

@mkubecek said in #4: > A small nitpick: the K-factor does not drop to 20 after 30 rated games but after completing **events with** at least 30 rated games, i.e. if you play your 30th rated game in the middle of a tournament, your update is still calculated with K=40 for the rest of it and K=20 is only applied after that. Thanks for pointing that out, I edited the blog to reflect that. I also clarified that the 400 point rule isn't cancelled, but is only applied to players below 2,650. 2,650 and above will not have the 400 point rule.
#7

lichess is RIGGED if I win a rated game I don't get rating and if I lose I get less rating WHAT'S WRONG WITH THIS

lichess is RIGGED if I win a rated game I don't get rating and if I lose I get less rating WHAT'S WRONG WITH THIS
#8

@mkubecek said in #3:

The reason the 400 point rule existed was for high rated players who happened to face an extremely lower rated player on rarer occasions (e.g. in open tournaments). This would reduce the rating loss in the case of a draw or loss, and also increase rating earned.

I don't think the rating loss on loss or draw was very important factor here, rather the gain on a win. In practice, we are talking about players with K=10, i.e. with the 400 point rule the deltas are +0.8 / - 4.2 / -9.2 while at the extreme you can only get to 0.0 / -5.0 / -10.0. From human point of view, losing 9 or 10 points of rating is not that different; on the other hand, it's quite frustrating to play a game, knowing that you don't gain anything if you win and only risk losing your rating. Always getting at least 0.8 for a win was IMHO the important part.

What I find very questionable is the effect of 400 point rule from the point of view of (much) weaker players: you play a game where you stand almost no chance and know that when you lose (which is almost sure), you still lose 1.6 points of rating (or even 3.2 with K=40). And with current version of the rating regulations, it's even worse: when a 2700 rated player faces a 1900 rated one, the stronger player does not get anything for the win but the weaker one still loses 1.6 or 3.2.

Yes

@mkubecek said in #3: > > The reason the 400 point rule existed was for high rated players who happened to face an extremely lower rated player on rarer occasions (e.g. in open tournaments). This would reduce the rating loss in the case of a draw or loss, and also increase rating earned. > > I don't think the rating loss on loss or draw was very important factor here, rather the gain on a win. In practice, we are talking about players with K=10, i.e. with the 400 point rule the deltas are +0.8 / - 4.2 / -9.2 while at the extreme you can only get to 0.0 / -5.0 / -10.0. From human point of view, losing 9 or 10 points of rating is not that different; on the other hand, it's quite frustrating to play a game, knowing that you don't gain anything if you win and only risk losing your rating. Always getting at least 0.8 for a win was IMHO the important part. > > What I find very questionable is the effect of 400 point rule from the point of view of (much) weaker players: you play a game where you stand almost no chance and know that when you lose (which is almost sure), you still lose 1.6 points of rating (or even 3.2 with K=40). And with current version of the rating regulations, it's even worse: when a 2700 rated player faces a 1900 rated one, the stronger player does not get anything for the win but the weaker one still loses 1.6 or 3.2. Yes
#9

@mkubecek said in #3:

I don't think the rating loss on loss or draw was very important factor here, rather the gain on a win. In practice, we are talking about players with K=10, i.e. with the 400 point rule the deltas are +0.8 / - 4.2 / -9.2 while at the extreme you can only get to 0.0 / -5.0 / -10.0. From human point of view, losing 9 or 10 points of rating is not that different; on the other hand, it's quite frustrating to play a game, knowing that you don't gain anything if you win and only risk losing your rating. Always getting at least 0.8 for a win was IMHO the important part.

Good point. I agree with you now. Edited the blog to reflect that it is actually the possibility of not gaining any points which was the problem.

@mkubecek said in #5:

FIDE should probably adopt Glicko-2 like lichess does.

Sadly, they won't - or at least not in the near future. Reading between lines in the Sonos' analysis, it seems obvious that he is well aware that a more modern rating system with dynamic reliability evaluation would be much better solution than ad-hoc hacks piled on top of the old one.

If top players ratings start sinking to a noticeable level they'd have to change it. The question is when it could happen.

Ken Regan says "Measuring the deflation directly, via my Intrinsic Performance Rating (IPR) measurements which are grounded in games played from 2010 to 2019, is complicated by several factors. One is that the guesstimated 25—50 Elo effect at elite levels is close to natural uncertainty in ratings themselves. "

We don't know if the rating deflation has reached the top yet as the 25-50 Elo difference could be natural uncertainty. Also, there is a bubble where some top players either don't play much or only play amongst themselves, shielding their rating. So it may take some time for a change.

But FIDE is afraid that people would not like a more complex system with what they call "hidden parameters" (RD and volatility); in other words, they prefer a system which provides worse results but is simple enough for common folk to be able to calculate (and check) their rating updates by hand. And I'm afraid any system providing significantly better properties than Elo would also be too complex to fulfill that requirement.

When did FIDE say they were afraid that people would not like a more complex system. I'd like to see.

@mkubecek said in #3: > I don't think the rating loss on loss or draw was very important factor here, rather the gain on a win. In practice, we are talking about players with K=10, i.e. with the 400 point rule the deltas are +0.8 / - 4.2 / -9.2 while at the extreme you can only get to 0.0 / -5.0 / -10.0. From human point of view, losing 9 or 10 points of rating is not that different; on the other hand, it's quite frustrating to play a game, knowing that you don't gain anything if you win and only risk losing your rating. Always getting at least 0.8 for a win was IMHO the important part. Good point. I agree with you now. Edited the blog to reflect that it is actually the possibility of not gaining any points which was the problem. @mkubecek said in #5: > > FIDE should probably adopt Glicko-2 like lichess does. > > Sadly, they won't - or at least not in the near future. Reading between lines in the Sonos' analysis, it seems obvious that he is well aware that a more modern rating system with dynamic reliability evaluation would be much better solution than ad-hoc hacks piled on top of the old one. If top players ratings start sinking to a noticeable level they'd have to change it. The question is when it could happen. Ken Regan says "Measuring the deflation directly, via my Intrinsic Performance Rating (IPR) measurements which are grounded in games played from 2010 to 2019, is complicated by several factors. One is that the guesstimated 25—50 Elo effect at elite levels is close to natural uncertainty in ratings themselves. " We don't know if the rating deflation has reached the top yet as the 25-50 Elo difference could be natural uncertainty. Also, there is a bubble where some top players either don't play much or only play amongst themselves, shielding their rating. So it may take some time for a change. >But FIDE is afraid that people would not like a more complex system with what they call "hidden parameters" (RD and volatility); in other words, they prefer a system which provides worse results but is simple enough for common folk to be able to calculate (and check) their rating updates by hand. And I'm afraid any system providing significantly better properties than Elo would also be too complex to fulfill that requirement. When did FIDE say they were afraid that people would not like a more complex system. I'd like to see.
#10

I attended the FIDE QC meeting last Friday and can tell you that Moiseenko and Sutovsky are particularly opposed to the idea of overhauling the rating system and switching from Elo to Glicko. As I had voting rights in the open meeting, and per my earlier article, quoted here, I was fully in favor of modernizing the system. The commission members were mostly split on the issue. There will be no massive change before 2028, and the results of the rating decay questionnaire haven't been made public yet, as far as I know.

I attended the FIDE QC meeting last Friday and can tell you that Moiseenko and Sutovsky are particularly opposed to the idea of overhauling the rating system and switching from Elo to Glicko. As I had voting rights in the open meeting, and per my earlier article, quoted here, I was fully in favor of modernizing the system. The commission members were mostly split on the issue. There will be no massive change before 2028, and the results of the rating decay questionnaire haven't been made public yet, as far as I know.