Chess.com ratings have massively deflated since 2020

Some of the basics about ratings are easy to understand.
But some basics are not so easy to Thoroughly understand ...

For example - at the higher levels of chess ...
a difference of 50 ratings points between two players playing each other ...
is much more significant than at the lower levels.
(Provided established rating systems like FIDE or USCF or Elo are being used)
Example ...
a 2750 plays a 2700.
That 50 points is huge.
Could we call it 'difference dilation'?

Whereas - try instead a 1250 plays a 1200.
It might appear to be the same gap - if you're talking about addition and subtraction on a number line.
But it isn't.
Its a smaller difference in strength.
When I first heard about this ... I thought:
'Hey that's interesting. Why would it become more significant at the higher levels?'
I don't know all the details as to why.
But the first thing to realize is that rating numbers aren't like mile or kilometer markers on a freeway or motorway ...
I thought of a better analogy just now - but it only describes rather than explains ... incidental rather than key.
For example - in the hundred yard dash - as a runner improves he will find it harder and harder to shave tenths of a second off his personal record times.
Describes but doesn't quite explain.
But some of the people here in this forum might both know that differences in rating become more significant in the higher classes - and know Exactly why - including the relevant and technical math too.

Another thing at higher levels of chess ...
Draws happen much more. A a bigger percentage of drawn games.
GM draws for example. 
In the ultimate classic chess which is tournaments against each other in over the board chess at slow time controls.
Why?
Because GM's make a much lower percentage of easily exploitable mistakes.
And their mistakes are usually much smaller mistakes too.
Does that connect to 'difference dilation'?

I disagree on #69.

There is a bigger gap between 1200 and 1300 than 3000 and 3300.

There are few players having 3000+ rating and therefore their pool is more prone to bigger deviations. 1200's pool is abudant of samples (players) and therefore significantly more accurate. In essence it is basically about better accuracy on the larger pool of samples.

Well known fact says quite opposite to your statement, actually, the stronger the players the difference between their level of chess is more subtle.

If anything the ratings decreased when there were more players because there was a lot of competition. Now ratings are inflated because there's less players and it's easier comptetion.

'bigger gap' is interesting.
But actually - your post does not completely disagree with mine.
And - in my post I am raising points that I was told about and asking questions ...
as opposed to making claims.

2750 versus 2700 or close enough - is something that happens in classical tournament play.
So is 1250 versus 1200.
Would one find that the results differ?
That the 1200's while not doing as well as the 1250's - fare better than the 2700's against the 2750's? (not suggesting we should Care ... Lol)
One would probably find there's more draws between the higher ranked players.

But the issue would be ... how do the games go that are not draws?
Who wins more?
The claim I read on the net years ago suggested that the 2700 would have a much harder time winning even one game against the 2750 than the 1200 would have against the 1250 and the 2700 would have a lower frequency of winning any games.
Does it matter?
Well - we're discussing ratings right?

"...But it isn't.

Its a smaller difference in strength..."

Not completely, yes, but I take the later as your take....

See my post #73

Now you are talking!

Sometimes - its good to start with the obvious.
If you win - you never lose rating points (exceptions?)
If you lose - you neve gain rating points.
If both players are equally rated and there's a draw - nobody loses points.
Starting from those premises - how should a good rating system work?
To answer that - one could first consider what the rating system is for.
In tournaments with an entry fee - would all players want to be in the same section as master strength players - or would one want a chance to win and to win a prize?
And to enjoy rated chess ... including in non-tournament play - do 2000 rated players want to be playing a lot of 1000-rated players?
Players want a Fight! And to test themselves. And to improve.
So the answer is ... No.
I would say the above is fairly obvious too.
From there - how should ratings systems work and do they work that way?
Yes. But not always perfectly.

Did you ever consider someone on chesscom is just playing with the math? Trying to project fide ELO?

If wanted, 2000 rating of strength now could be easily projected to 65 000 rating in short time (or why not to 1800?). It is good to know it is only the playing with the math. It is possible to adjust it to any rating range family you wish to.

In the tactics problems they're definitely 'playing with the math' - especially in recent years.
Chess.com has an excellent marketing department it seems.
It knows well what the target markets want.

Has Covid and a much wider playing base of players affected the ratings?
Seems likely.
Its interesting and ratings are definitely an intellectual subject.
But also a 'practical' subject.
Paradoxically - I regard the Tactics puzzles as a very valuable training option.
Very efficient.
Gets right to the heart of the matter.
The study not just of 'refutations' ... but of mistakes.
I started getting more out of the puzzles when I started doing them Unrated and with no timer.
But again there's a paradox.
If you're taking a long time and you realize you're not breaking through ...
should one crunch and crunch forever to try to get the right solution?
Some strong players will and do say 'admit to yourself you don't get it and go ahead and make a move based on principles knowing you'll very possibly or probably get Incorrect'
Connects to ratings?
Certainly to rated play.
If you take too much time ... you know what happens.

Yes, the elo points are returned but only within a 90 day span, and only up to a certain amount of recent games of the cheating player.

However, there are also some cheaters who either don't cheat every game or use a weak engine or whatever, and if I play them multiple times and win and lose some games, I am refunded the lost points, but the gained points remain, thus inflating the ratings.

However, I think ratings are going down slightly on this site. Like others have pointed out, I also have anecdotal experience where I have looked at past games where my rating was similar to what it is now and noticed that I perform at a higher level now but the ratings I have now are only marginally better than in 2020-2021.

One potential cause for this could be that a bunch of players stronger than 400 join the site, win some gains and gain rating points, taking rating from others, but even though they have gained more than others have lost (due to high Glicko RD) they get bored of chess and no longer log in. Thus, it doesn't matter that they gained more because they are inactive now. With the chess boom, so many new members are joining and a lot are not staying. So looking at which players are going inactive is also would help with explaining possible rating deflation.

Facts

What’s Glicko RD

@83

"What’s Glicko RD"
++ RD is the rating deviation. An established player has a low RD. A new or young player has a high RD. The Glicko-2 rating system calculates the individual RD at any time. The 1970 elo system assumed the same constant RD for all players so as to simplify calculations, as computer time was expensive in 1970.

http://www.glicko.net/glicko/glicko2.pdf

You're both wrong.

By the defintion of elo, any two players 50 elo apart should have the same win/loss ratio to each other.

If this is not true, there is a flaw in how elo is calculated.

@85

You are correct: only the rating difference matters.
However, that implies that the ratings of both players are stable.
A lot of new or young players are underrated, as their rating has not yet kept up to their improvement.

If you look into it - you might find that its harder for a 2700 to win a game from a 2750 than it is for a 1200 to win a game from a 1250.
Somebody suggested this to me a long time ago.
That the same rating differences are more significant at higher levels.
Maybe its not so. Its not a 'claim'.
Or maybe it is so.
Consider that there's a higher percentage of draws at the higher levels.
Then ask yourself - does this mean its harder for the weaker player to win a game at the higher levels?
What answer do you come up with?

But the main subject seems to be that if a big new wave of strong players with low chess.com ratings have arrived during Covid and 'taken rating points' from other players and then left again with the pandemic now in the past - then 'rating points have been lost from the pool' as it were.
'deflation'.
Its an interesting idea and may be valid.