EGF Rating System Commission Report 2020

[jlt]

If (say) 50% of people improve by 1 stone and 50% don't improve, the former won't have their rating reset because 1 stone difference is not a clear evidence of improvement, so the whole rating system will be deflated by one half stone.

[gennan]

I think it's very optimistic to assume that 50% of tournament players improve by a stone during the corona crisis (even while they can only play online). A large majority of tournament players is at a plateau (for years) and that won't suddenly change during the corona crisis.

[gennan]

It is possible to fix this by changing the rating bonus mechanism. Now it works by giving a bonus for every tournament game. If this is changed to give a bonus over time (regardles if you played tournament games during that time), then it would be possible to compensate for average player improvement even during a pandemic.

[jlt]

Some people are in favor of developing online tournaments and rate them with a full coefficient, but this solution is not ideal because of cheating or disconnection risks.

[jlt]

I think it's very optimistic to assume that 50% of tournament players improve by a stone during the corona crisis (even while they can only play online). A large majority of tournament players is at a plateau (for years) and that won't suddenly change during the corona crisis.

Looking more closely at the EGF rating system, I think you are right. Around 5 kyu, the expected gain after one match is con*epsilon/2 which is a bit less than 0.4 rating point. If we estimate that 60000 rated games are played each year, and that 6000 people are EGF rated, then each rated player plays about 10 rated games/year, meaning that the system assumes that a player gains on average 0.04 stone/year.

The rating system does not necessarily correspond to reality, and my calculation is very rough, but after 1 year or a bit more of corona crisis, the drift will probably be less than 0.1 stone, which looks fine and could easily be corrected by adjusting the parameter epsilon at least for a while.

(I was imagining that many sandbaggers would appear all around but this probably won't happen.)

[gennan]

I think those estimations are roughly correct.

In the upcoming changes to the rating system, the bonus will be increased for kyu players (it will be about 1 rating point around 3k, 5 rating points around 15k and 8 rating points around 30k, which is about 2 to 3 times the current value). We think this larger bonus is closer to the real average player improvement per tournament game.

Plugging that increased bonus into your rough estimations for the average tournament player improvement per year, the predicted values are still fairly small (0.1 stone per year around 3k, 0.5 stone around 15k and 0.8 stone around 30k).

[gennan]

The announced EGF rating system update has taken place now:

(email sent to EGF Board and EGF Members on 5-4-2021)

Dear all,

in the recent years it showed that the calculation of the european go ratings needs a revision. It could be observed that there was a deflation in the ratings and consequently also in the european go ranks. It's natural that the ranks in Europe are bit higher than the ones in the United States or Japan but this gap even increased by the inherent deflation of the system itself. Some players did some research on this and finally Dave de Vos put up a webpage on goratings.eu to show the effect and propose a new formula. This new calculation algorithm has now been implemented. For stronger dan players there is often almost no difference, for the big bulk of players around 1 dan there is a change of bit less than 1 rank upwards, so for example a 1 kyu with 1992 rating points (GoR) has now 2068 GoR and can play as 1 dan, a difference of about 60 to 100 points. For a single or double digit kyus this change can make up to 140 rating points, it depends bit how many tournaments / games he has played.

We also expanded the rank floor from 20 kyu to 30 kyu. The intention for this change is to give beginners an easier start into the world of rated games as the usual case till now was that a beginner not only lost most of his games but also stayed for a long time at 20 kyu.

It's now more than a year that me and Julien Corcessin took over the european go database. At the moment it's bit more quiet than usual but still there is a lot of daily work validating the tournaments results and doing small improvements as the ones you see now. We still have a lot of work in lifting up the whole system to current web standards. If you are knowledged in the areas of PHP programming, databases, web design or simply want to help update the documentation or help validating tournaments please don't hesitate to contact us! There is also room for new projects like e.g. having an integrated web application for doing the tournament pairings instead of entering them in an offline java program.

Please, if you know some programmers let them know that we are very interested to have more people in our team!

I would like to say thanks to all members of the rating commission making this upgrade possible and a special thanks to Dave for all the work he has done.

Cheers,
Lorenz (EGF Secretary and EGD rating manager)

https://europeangodatabase.eu/EGD/

PS: Until the AGM in 2022 online tournaments can be entered as class C if they are well organized with proper anti cheating measures and national online championships can be entered as class B.

[Adin]

I've noticed an important change, points gain/loss seem to be much lower than before. For example some years ago I had a nice tournament where i gained 109 points rising from 2k to 1k. But in the new system I only gained 59 points.

It was mentioned before in this thread that on average a player has roughly 10 rated games/year, which means roughly 2 tournaments. Now that it's become more difficult to gain/lose points and with players having 2 tournaments/year on average, doesn't it mean that the gap between real strength and EGF rating risks getting bigger? Essentially having a lot of underrated players (due to slowed down rating increase not being able to match improvement) and also having more overrated players.

[jlt]

Edit: I misread your post, ignore the italicized part.


=== begin comment to misread post ===
The problem of the EGD rating not keeping up with the player's progress during the corona crisis was discussed a few messages earlier:

viewtopic.php?p=260066#p260066

In a nutshell, the majority of players are not improving anymore, and those who are improving quickly can ask their rating to be reset.

On the other hand, you can't say there are no tournaments anymore. We recently had two Corona cups, the End Polio Tournament, the Paris Tournament and many others.

=== end comment to misread post ===


In the new rating system I managed to gain 49 points with the "End Polio" Tournament which is class C. So I think the rating can keep up with your progress even with a couple of tournaments/year.

[gennan]

The formula at https://europeangodatabase.eu/EGD/EGF_r ... php#System does not mention handicaps. I looked around that page and this thread but couldn't find anything.

How are the rating changes affected if the player gives or takes a handicap?

Handling of handicap stayed mostly the same: handicap increases the rating going into the expected winrate formula by 100 * (H - 0.5), so it assumes handicap elevates the level of the handicap reciever by the size of the handicap.

[gennan]

I've noticed an important change, points gain/loss seem to be much lower than before. For example some years ago I had a nice tournament where i gained 109 points rising from 2k to 1k. But in the new system I only gained 59 points.

It was mentioned before in this thread that on average a player has roughly 10 rated games/year, which means roughly 2 tournaments. Now that it's become more difficult to gain/lose points and with players having 2 tournaments/year on average, doesn't it mean that the gap between real strength and EGF rating risks getting bigger? Essentially having a lot of underrated players (due to slowed down rating increase not being able to match improvement) and also having more overrated players.

1) We felt that ratings of lower rated players were overly volatile in the original system, so their con/K factor was reduced.

2) The system is not clairvoyant. It has no way of knowing if a player has improved 3 stones between 2 tournaments 1 year apart. When players improve so quickly that their calculated rating doesn't keep up (because they play few tournaments relative to their improvement rate), they should just declare their new rank in the next tournament instead of sandbagging with their obsolete rank. When their new rank is 2+ ranks higher than their previous highest declared rank, this triggers an automatic rating reset in the system, so their rating is just updated to their new rank before updating it with the tournament game results.
This rating reset mechanism already existed before the rating system update. However, some countries have quite conservative policies in this regard and don't allow it, even for low rated players, effectively forcing those players to sandbag until the rating system catches up.

[Adin]

some countries have quite conservative policies in this regard and don't allow it, even for low rated players, effectively forcing those players to sandbag until the rating system catches up.

This is a very good reason to not make it even harder for those players by lowering the rating gains.

In real life players are much more likely to overrate themselves. They will happily think they are stronger than they truly are. The new system will reinforce the idea that "I'm actually much stronger and I can't go up fast enough by playing so I need to declare a stronger rank". Then as the rating losses are also significantly reduced, the overrated player will take a longer time to reach its real rating.

Essentially instead of rewarding or punishing players based on objective results in games the new system promotes setting your own rating based on subjective opinions.

[jlt]

In real life players are much more likely to overrate themselves. They will happily think they are stronger than they truly are. The new system will reinforce the idea that "I'm actually much stronger and I can't go up fast enough by playing so I need to declare a stronger rank".

Not sure about that. Even though my online ranks are still improving, and even though my EGD rank is a bit underrated, the improvement rate is not spectacular so I keep thinking that I am stuck, especially after losing against a weaker player. So I wouldn't declare a stronger rank at the next tournament even if it is allowed.

[gennan]

some countries have quite conservative policies in this regard and don't allow it, even for low rated players, effectively forcing those players to sandbag until the rating system catches up.

This is a very good reason to not make it even harder for those players by lowering the rating gains.

In real life players are much more likely to overrate themselves. They will happily think they are stronger than they truly are. The new system will reinforce the idea that "I'm actually much stronger and I can't go up fast enough by playing so I need to declare a stronger rank". Then as the rating losses are also significantly reduced, the overrated player will take a longer time to reach its real rating.

Essentially instead of rewarding or punishing players based on objective results in games the new system promotes setting your own rating based on subjective opinions.

There are both optimistic and pessimistic players when it comes to self-promotions (this was also the case before the rating system existed). The rating system has a rating bonus to compensate for any imbalance between those types of players.
Your gut feeling may be that players tend to be too optimistic on average, but the overall statistics of the rating system indicate that it needs a positive rating bonus instead of a negative rating bonus to minimize long term overall deflation/inflation of the system. This supports the view that on average, European players tend to be (slightly more) pessimistic rather than optimistic in regard to self-promotions.

[Adin]

Trust me, as a KGS admin for almost 20 years I've seen how much some players are willing to do just to increase in rank. From all kinds of dishonest stuff to things like specifically choosing their opponents based on profiles that they think give then better chance of winning. I'm not saying that most players are like that but certainly most players want to have a better rank. I've also seen a lot of offline/EGF rank special promotions that made no sense approved by my national federation (Romanian one). I even made an analysis shared with Romanian players community that showed how the majority of special EGF rank promotions have proven to be overrated over the following years. It's human nature to want to appear better to others.

Ideally ranks should be based just on game results. That's not always possible in offline ranks, but in my opinion lowering the impact that games have on rating is a big step in the wrong direction.