Rating deflation

When is Chess.com going to adjust their Elo ratings to account for deflation?  This is a known phenomenon that occurs as newbies improve and spend down the rating points of established players. It appears to me that the blitz ratings distribution is about 250-300 points lower on chess.com vs USCF or other sites. 

See https://www.chess.com/blog/SmarterChess/are-chess-com-ratings-the-most-accurate

NM Smarterchess judges chess.com ratings to be closest to USCF ratings. 

There was no talk of deflation.

Are you concerned that 231 point drop in your blitz rating since April was caused by deflation caused by the influx of new players?

PogChamps roadkill?

No. I’m concerned because the distribution curve of chess.com ratings is not bell shaped and does not have a mean of approximately 1500. These distribution parameters have historically been the case in the USCF and was the intention of Arpad Elo. 

Are you sure about that?  The USCF OTB players included a ton of low rated scholastic players. Also, the chess.com distribution is not bell-shaped, as it should be. It is skewed with a modal low rating and a longer rightward tail. 

The mean Rapid rating has deflated something ridiculous since the beginning of the pandemic, the average around 1020 at the start to 953 today. The addition of 10|0 mitigated it for a while but down it continues to go.

I haven't checked, but where does the initial rating come from?  I am a US Chess A-player (1800-1999) but I didn't link my account to my US Chess ID and started at 1593 after my first game.  If weak players start at that level then there is rating inflation with them having a lot of points to donate to others before their rating drops to their strength.  If strong players start at that level then there is rating deflation as they get points donated from others before their rating rises to their strength.  As players get stronger there is rating deflation as they need to get points donated from others until their ratings reach their now-stronger strength. As players age there is rating inflation as they have rating points to donate that are in excess of their weaker strength.  US Chess deals with this by tracking the ratings of a specific age and injecting bonus points (for very good results in tournaments) to keep that age group close to the same average rating (the players in the group changes from year to year as people age into or age out of the group, but the group is otherwise stable).

The lack of OTB events has pushed a number of stronger people on-line (deflation) and the lack of other activities has brought in a number of weaker players (inflation) but it may be that the smaller number of stronger players stay active longer and thus their deflation outweighs the inflation from the weaker players.

Since ratings change from game to game instead of tournament to tournament, something other that bonus points would be needed for rating point infusion. 

PS US Chess rating floors are also somewhat inflationary to counteract the natural deflation caused by players improving their strength, but they were created for an entirely different reason and are less significant than bonus points to counteract that deflation.

If there was not a US Chess minimum rating then it would more closely approach a bell curve (a different bell curve for scholastic and non-scholastic - not a bell curve for the entire population in one).  The 100-150 absolute floor minimum (100 + 4 per win + 2 per draw + 1 per 4+ round tournament until 150 is reached) is why there is a large mass in that minimum.  Without that floor you would see highly negative ratings that put a tail at each end of the bell curve.

I've also noticed the decline in average rapid rating. From what I recall, it was in the low 990s just before 10|0 was recategorized as rapid in September. It briefly climbed to ~1025 just after that, and then started to decline and hasn't stopped yet.

Things that I suspect might be having an effect: 

1) new players being able to choose their starting rating by answering the "what is your level of chess experience?" question during account creation and selecting a low rating like 800 or 1000,

(2) unusually many new players in the pool playing mostly each other with minimal external calibration from existing players due to the recent surges from the pandemic, Twitch/PogChamps, and The Queen's Gambit, and

(3) dissimilar 10|0 and 15|10 (and 30m) time controls sharing a single rapid category causing a lack of transparency in the meaning of rapid ratings/percentiles/etc.

I’ve sensed a Blitz deflation as well. I remember being in the 85% percentile when I peaked at 1510, and now I’m in the 83% percent at 1360, 150 below!

80% used to be above 1400. That may be caused by the pandemic, people who used to play otb only started to play online, combined with an increase in beginners. May also be due to cheating, I’ve been playing some really strange games, dude blunders a queen then defends like Karjakin. And, of course, the two reasons are not mutually exclusive.

I agree 100% with you. At least the blitz distribution needs to be rehauled. Chesscom's distributions have no resemblance to real world distributions from USCF or FIDE.

For example, a rating on chesscom of 1700 is approximately 95 percentile. In USCF 95 percentile is about 2000 rating. In FIDE 95 percentile is about 2300. 

@notmtwain, see the links below and feel free to compare with chesscom. 

http://www.uschess.org/archive/ratings/ratedist.php

https://en.chessbase.com/post/visual-presentation-of-world-chess-ratings

I wouldn't expect them to be analogous.  FIDE is recently starting to assign lower ratings. Prior to that only a US Chess 1800+ (and more likely 2000+) would be able to qualify to get a FIDE rating.  Thus the 95th percentile in FIDE was the 95th percentile of a group that was already above the 80th percentile thus translating more to the 99th percentile (or higher) of the US Chess rating group (for instance, a player in a US Chess event that only has a FIDE rating can be paired with an arbiter-assigned pairings rating of FIDE+100 which is a very direct indication that the US Chess and FIDE 95th percentiles measure very different populations with the FIDE-rated players consisting of stronger players on average).

Then you add that Chess.com has newer players that have not even gone to regular tournaments.  A US Chess 800 may well be in the 99th percentile of the general US population (includes those who don't even know the game) so it would be interesting to know what Chess.com's national breakdown of players is to get an idea of how many of the players do not play rated chess over-the-board.

You have a good point regarding FIDE ratings. I was only today made aware of their rating floor. However, ratings only measure active players so players who don't play are irrelevant in this topic as they aren't being measured in any of the organizations.

If we only consider USCF ratings chesscom ratings appear to be deflated at least in the blitz format. I saw a recent article by chesscom where they awarded 150 points to active bullet players across the board. If that's not admittance of problems with their rating distributions I don't know what is. 

I've thought more about this issue since I wrote my previous comment in this thread, and I've come to the conclusion that the deflation stems from people entering the rating pool with lower starting ratings than the current average rating at the time that they enter.

I written more about this near the end of this other forum thread about average ratings.

That makes sense. 

Disagree with everyone, focus on improving your games, try videos, lessons anything, but don't complain like a deprived child if your elo is under 2000

Exactly, how do you know melvin garney

Exactly, how do you know melvin garney? You must be the smartest person here.