How berserking influences local ratings?

@mcgoves said in #8:

It is the main question, although one which would be hard to reply, given that many factors influence the overall rating distribution.
I do not expect to get to 2650 in Chess960 any time soon, I was overrated before the titled arena. That said, opening this topic before losing a plenty of points still felt better than after it.
My question was motivated by an experience when I had lost around 100 points of my relatively stable blitz rating on my alternate account and then struggled to get even halfway back in spite of playing well. Scoring 7.5/9 against the opponents rated about 200 points below me only brought me 10 (Lichess) rating points, although it was a very good result. Of course this topic is not about my online ratings, I am just wondering if this (a significant decrease of Glicko-2 through berserking, which could lead to a rating deflation) is not a relevant general phenomenon.

@RealDavidNavara said in #11:

I do not expect to get to 2650 in Chess960 any time soon, I was overrated before the titled arena.

One could also argue that your fellow GMs are underrated, it felt a bit strange to see that the top players at yesterday's event had so low ratings compared to blitz and bullet.

Of course this topic is not about my online ratings, I am just wondering if this (a significant decrease of Glicko-2 through berserking, which could lead to a rating deflation) is not a relevant general phenomenon.

The berserk feature distributes points from the top players to lower rated players and therefore caps the maximum ratings at a lower number than they would be without it, but I do not see how it would deflate the rating average. After all, the non-berserking opponents experience the opposite effect, their ratings get boosted temporarily and due to decreased deviation will only go down slowly after the event.

@RealDavidNavara said in #11:

Scoring 7.5/9 against the opponents rated about 200 points below me only brought me 10 (Lichess) rating points, although it was a very good result.
IMHO the problem here rather is that with 200 points of rating difference, the expected result is ~0.75 points per game for the stronger player. Therefore your expected result in 9 games would be 6.75 points and you scored 0.75 more than that, i.e. better than expectation but not too much better.

For comparison, FIDE rating regulations have we=0.76 for 200 points of rating difference so that 7.5/9 would be better by 0.66. Thus you would gain 13.2 points of rating with K=20 or 6.6 with K=10.

Many titled players are heavily underrated in Chess960, as many of them only play titled arenas (where quite some people start with a rating of 1500) or prize events (where the best players have to berserk a lot, losing a lot of rating). In my opinion, titled Chess960 arenas are underrated by 150-200 points compared to the pool, although this is an oversimplification.

As for the opponents of berserking players, you have a point, but they will not decrease their Glicko-2 as much as berserking players, as non-berserking players will not play so many games. (Last but not the least, many berserking players at the top frequently berserk against each other as well as again everyone else.)

I like the time control 3+2 for Chess960, but out of the four options (me berserking or not, the opponent berserking or not) I do not especially enjoy any. One berserk creates a big imbalance, two berserks lead to a very quick game which is likely to contain many mistakes and cause a lot of stress, and a game without any berserk gives the other players a competitive edge, as it lasts long. Well, I love Chess960 and this is why I still play those arenas, in spite of my reservations to the tournament format.

@zwenna said in #12:

One could also argue that your fellow GMs are underrated, it felt a bit strange to see that the top players at yesterday's event had so low ratings compared to blitz and bullet.

The berserk feature distributes points from the top players to lower rated players and therefore caps the maximum ratings at a lower number than they would be without it, but I do not see how it would deflate the rating average. After all, the non-berserking opponents experience the opposite effect, their ratings get boosted temporarily and due to decreased deviation will only go down slowly after the event.

I think it's fine. Whatever your rating is, it tries to estimate your possible future results based on skill and behavior.

So if someone berserks a lot, then the rating should reflect that.

As for the slow climb after berserking a lot, maybe you tried to climb back way too soon. Let's say one berserks 100 games today and lost 100 points. At this point the rating system believes it has enough data to peg your rating to that number. Movement of rating would then be narrow. The system is confident that it has pegged you down. The system is agnostic whether it's berserk or not, which I think is fair, because who knows if you might go on a berserking spree again.

However, if you stop playing for a couple of weeks, the system slowly loses confidence that it has pegged you, so it will swing wider again.

It's best to think of rating movements as the system's confidence in a player's current rating. So another way to pose your question is, why is the system so confident that my rating is x?

A separate rating for berserk would make things "more accurate", but I think it's overkill.

@mkubecek said in #6:

Is this about updating after each game rather than batched over regular time periods again? Good luck convincing the ADHD online chess crowd to have their ratings updated once per month. :-)

I do not know