Feels like competition is harder at a given rating now.

I started playing chess last January for the first time, and quickly rose from mid-700s to mid-900s rapid across a TON of play in February and March, peaking slightly above 1000 in April.  I played a lot less for a while, and my rating dropped.  Now I'm bouncing around 850-920 or so.  OK, taking a break is likely to leave me out of practice.

However, as I've been playing a lot more and renewing my studying, I have been feeling subjectively like I've been improving, at least in small ways.  I've been doing tactics training and more often see beyond simple one-move tactical ideas.  I have more concrete plans for what I'm trying to do and far less often try things just because I'm out of ideas.

So, I pulled up some accuracy numbers using the new Insights feature in beta just to see if the data corresponded to my feeling:

Here's the period from April to August, when I was above 950 up to a peak rating of about 1014:

Here's the period from September to now, when I have been around 850-920:

I also use another site to do meta-analysis of my games on chess.com, and data there suggests I'm blundering more than I used to when I hit this rating range in the spring, so it's quite possible that I'm getting better on average but losing more games to one or a few catastrophic mistakes.

Still, it would make sense that with restrictions lifting around the world, and The Queen's Gambit effect wearing off, that ratings might shift downward a bit (since rating systems like chess.com's are dependent on overall skill levels of the player population.)

I guess I'm just curious whether other people have had the perception that competition is tougher now at a given rating than it was last year?

Personal note:  I'm fully aware that a common and very reasonable response is to say something like "who the hell cares and why are you worrying about your rating instead of trying to get better at the game?  Well, I'm doing that too, but I've always found statistically-based rating and matchmaking systems intellectually interesting, so no harm in taking a few minutes to look into the question.

I think you are close @CooloutAC but I think it is the opposite in a way. With a pandemic, MORE people have joined chess.com and decided to learn chess (chess.com reveals some statistics confirming this via a chess.com article during the pandemic), so this means there are more beginners. This means more players rated under 800 and this skews the average lower than it was. 

Once the pandemic ends, then we will see those players move onto OTB chess or let their chess.com accounts go inactive (out of chess.com rating pool) and then the ratings average may gradually return to 1100ish

oh okay, I thought you meant OTB ratings going up by itself, but you meant online ratings going up Okay, my bad, you are correct

It's well known lichess ratings are higher than on here and not comparable. 

So @KeSetoKaiba, that's exactly backwards.  Let's say you have a small population of skilled players.  For most players, matchmaking ensures a near 50/50 win rate, but those at the very bottom of the scale do worse because it's harder to find them a good match.  A large influx of new players with unknown ratings but lower average actual skill will mean players at the previous bottom of the scale get more good matches, which drives their ratings UP during pandemic boom/Queen's Gambit times.  Then, slightly higher-rated players win more, and their ratings get driven up, and eventually the upward pressure extends throughout the rating range.

Then, when the new players start to get bored, stop playing chess, move on, some people will drop out.  It's likely that those who are less engaged and drop out will tend to have a somewhat lower rating distribution.  So, now, at the lower end, it gets harder to find a good match and players lose more.  The lowest players start losing rating points and this creates downward pressure throughout the rating range.

TL;DR: A huge influx in inexperienced new players should make high ratings easier to get and so existing players' ratings will RISE.  As interest dies down, assuming those who remain are biased toward the more skilled, high ratings will be harder to maintain and will DROP.

Note that this is an expected feature of statistically-based rating systems like Elo and Glicko:  They measure individual skill relative to the overall skill distribution, not relative to any absolute standard of quality.

Also, note that if the large influx of new players had a similar or higher skill distribution to people already playing, this effect would be reversed.  I'm assuming they're less-experienced and probably less-skilled on average.

Edit:  All this is why my original post said that it felt my rating was going DOWN while quality of play was staying the same or increasing, and I was hypothesizing that this might be because casual players who came in with the pandemic/Queen's Gambit wave were dropping out.

Every since joining in 2017 my rating has gone up and then back down, so this is probably true. 

#3 its not weird, its exactly as it should be. Lichess compared to cc has heavily overrated players - influation is bigger. I can't get to 1900 on cc yet, but on lichess I never drop below 2050. Thats why I like cc out of two because its more realistic. My OTB rating would probably be 1500-1600 and even lower. Being a 2000 is my goal but I have to admit that I still have tons of work to reach it and just because I have that rating on lichess it doesnt mean I have that strength neither on cc and I wont even talk about FIDE, its a long way to get there.

#1 I haven't noticed anything and stats you showed as are slightly different. I neve experienced a big plateau so far and all my improvement was somewhat smooth and same. I dont think it would be the case if players got better. To reach 1000 you need to stop blundering free pieces and start taking every free piece your opponent gives you. This is enough to get you to 1000 in a month.

Nope, he was saying that new beginners would bring overall ratings down and that them leaving would bring ratings back up.  It's the opposite.

To be clear, I'm not raising this question based on how I feel about my rating, and I know that ratings fluctuate normally.  I'm raising it because my measured accuracy over two periods, each months, with hundreds of games each, has gone up, while my rating's dropped about 100 points.  (Also, it's well-known that rating numbers are not absolute, they're relative to the population.)

The accuracy number differences are relatively small, but they've all gone the same direction (for wins, draws, and losses) and averaging over hundreds of games for each period means that the effect of random variance from game-to-game will be reduced quite a bit.  That still doesn't establish that it's a significant change of course.

That's interesting!  I guess then the question would be whether the accuracies presented by the Insights tool (currently in beta, but it's where I got my numbers) are only the new accuracies, or are they a mix of new and old?  That's probably an implementation detail that only chess.com devs know for sure, but given that the Insights tool does require time to do a deep-dive analysis of historical games, I'd expect that they're all the new numbers.

Sorry, I realize this is extremely difficult to explain clearly because it's somewhat counter-intuitive.

An influx of new beginners would bring overall ratings UP for existing players even if their skill doesn't change, because the competition is now easier on average.  Then when the beginners drop out, leaving the more experienced players behind, ratings go back DOWN at a given level of skill.

(The mechanics about why this happens in a rating system like Elo or Glicko have to do with players at the very low end of the rating scale finding it easier or harder to find an even match, and their rating changes cascading upward through all higher ratings.)

Edit:  It's not a question whether this kind of drift happens with statistically-based rating systems.  It does!  The question is whether it's happening with chess.com in the aftermath of Queen's Gambit.

Insights uses the new move classifications and accuracy algorithms.

To the extent that people choose their opponents, that may have a counteractive effect.  However, the vast, vast majority of rated games on chess.com go through automated matchmaking, which strives, above all else, to give you as close to an equally-rated opponent as possible.  This is entirely where the upward or downward pressure on ratings comes from.

Note that even ratings among those who choose their opponents will end up affected by this.  As long as someone in a pool of friends plays automatically matchmade games sometimes, that will exert upward or downward pressure on the ratings of everyone they play.

Oh no, it really doesn't work like that at all.

First, FIDE uses Elo and chess.com uses Glicko, which are different systems (though each strives for the same goal, that difference in rating map predictably to likelihood of game outcome.)

Second, the numbers are arbitrary.  FIDE could rescale their ratings so that average were 1000 if they wanted to and it would affect nothing, since in Elo, only rating differences tell you anything, not the absolute number.  (This is why some experts speculate that the play quality at FIDE 2500 has increased over time.)

FIDE probably has rating drift but because they do not generally use automated matchmaking, it is unlikely to behave much like an online system like chess.com.

No, there will be more higher skilled people left, with lower ratings than they had.  Rating numbers do not directly correspond to skill (though rating differences do correspond to skill differences.)

I feel like what you're missing is that if you got rid of everyone below 2000, those who remain would eventually shift downward in rating.  The stable 2000-rated player from before would not be able to maintain their rating because they would *always* be matched against a stronger player, and would thus lose points more than they earn them back, until eventually they hit the artificial floor of 100 rating.

The reason ratings are usually stable is that every win has a corresponding loss, but if the worst players become more skilled on average than they were, they'll lose rating because they have to lose extra games to make up for the high win rates at the very high end.  (Presumably, the people who left were absorbing more than their share of those losses.)

Playing other people at higher ratings does NOT tend to raise your rating, by the way, unless you're already improving rapidly.  If you're the legitimate worst player in the skill pool, you will lose a lot more than you win and playing extremely higher rated players will make that worse.

It is true that kids who are already rising through ratings rapidly benefit from seeking out higher-rated opponents.  The only reason that's the case is that they have a higher-than-predicted chance to win against them in the first place, because rating changes lag skill changes.

I’ve been looking at the back and forth of the whole rating change. I’ve done a simple calculation to get some feeling for the effect of some changes in the players pool.

Assume, for the sake of arguments that in state 0 everyone has an ELO rating that is exactly equal to their playing ability with an average of 1200, let’s calculate using 1000 players. 

Then 300 new players enter the pool, 100 of them with a real rating of 200, and 100 of them with a real rating of 800, yet they start with a 1200 rating. 

The 1000 players with a 1200 playing ability represent 1.2M (1k x 1200) of the total playing ability of 1.3M (1.2M + 100x 200 + 100x800). This will translate to 92.3% of the playing ability. With a total rating points available of 1.44M, they get 92.3 of that number, which is 1.329M, or a rating of 1329 on average for each player. The 100 800’s are rated 886 and the 100 200s will be rated 222.I’ll ignore for now the effect of players added with their exact rating from day 1, because I don’t think that will matter,

Step 2. The 100 players with the 222 rating leave. The 1000 initial players now represent 93.75% (= 1.2M / (1.2M + 80k)) of the playing ability. The total points available on the site is 1.4176M (= 1000 x 1329 + 100 x 886). Their share is 1.329M which means this has no impact on their ratings. Though the average rating has gone up from 1200 to 1289, the ratings of the remaining players is not impacted. Right? 

Back to the start. Let’s say in the initial player group of 1000 players with an average of 1200, the actual distribution is 100 players at 1600, 200 players at 1400, 400 players at 1200, 200 at a 1000 and 100 at an 800 rating.

Then we get the same influx of new players 100x200 and 100x 800 and the free ELO they bring into the pool. The 800 rated players represent 18.46% of playing ability. Out of 1.44M points available the increase in rating would again be 886.

My idea of how this would play out:

If you were an 800 player before the influx of new players, your rating would increase. By how much depends on the free ELO added to the system. More importantly, while you go up, most of the players you face are still on their way down,  as their ELO rating is not adjusted to their playing ability. While the chess boom is happening, you will face opponents with a 800 or 900 rating on their profile, while their actual playing ability would be around 500-700. So, will will increase fast and you get better results with a lower accuracy.