Rating inflation on chess.com blitz

Hi ive tried sending an email but got no response, but there has been some pretty serious rating inflation on chess.com blitz in the past two years. Looking at the top players ratings, hikaru was struggling to stay over 2900 3 years ago and now just hit a new all time high of 3260. His graph shows a steady increase in rating over time. In the same time frame I dont think Hikarus fide blitz rating has gone up that much (hes 2934 fide blitz now and was 2906 fide blitz 4 years ago). So clearly Hikarus not really getting much better but the chess.com ratings have been getting inflated.

Also there are 60+ players over 2900 and 4 years ago Hikaru was the only one and had a large margin over second place. So its not just Hikaru.

It would be nice if the chess.com staff could address this issue and keep the ratings somewhat in line with FIDE ratings (ie the top player should be no higher than 3100 at any point)

Ratings in different pools don't need to be close. Players are not playing thousands of OTB rated games a year, which will have an impact on how high ratings can get. If they were playing 30+ OTB rated games a day against a wide range of ratings, FIDE ratings might be higher. 

Also, in OTB, most of the top players are playing in a pretty small pool, most of the time, and here they are playing a larger set of players, and if their performance is good enough, they will get higher ratings as well, in accordance to the ratings of the rest of the player pool.

Maybe FIDE ratings are deflated

ok but I dont think fide blitz ratings deflated 300 points in 4 years. And if this trend continues we could have hikaru hitting 3500 blitz next year.. that just seems a bit inflated to me

surely you can adjust the rating system so that you get less points for beating lower rated players and more points for being higher rated players (still capping out at 16 points total). That would do the trick for sure

The site uses the Glicko rating system and already acccounts for rating disparities between players and the expected outcomes based on that.

The point is, the chess.com player pool and the top-level playing pool for FIDE, which you are comparing, are different. Since the pools are different, the ratings will also be different. While some players are in both pools, the FIDE side of things is much smaller and the top players are more insulated from other members in the pool.

Ratings are not a measure of absolute strength. They are a gauge of player's past performances in a particular pool. Also, as mentioned, the sheer number of games will have an impact as well, especially in events like Arenas, where there is a wide range of players and ratings.

It doesn't take many games for any rating to quickly get close to what actually reflects their playing performance.

What he's trying to say is that it is caused by Hikaru not playing rated games in OTB against, say, 2000s etc and maybe gaining one or two points per game from that as it is online.

Yes but what do you think happens to that rating that was lost going from 1800 to 800?  It turns into inflation.  Those 1000 points are just added to the pool, increasing the average rating per player.  

That only happens when the RD of both players is low, but new accounts have a ridiculously high RD. So for example they can go from 1800 to 1650 in one game and their opponent might only win a few points.

If you look at blitz rating graphs of players who have been playing often the last few years, you wont see anything to suggest inflation.

That's not how math works, lol.

The rating system (the math) does a very good job of comparing players.

The range, and particular numbers will vary from place to place depending largely on where the administrators initially set the starting rating or average (in other words starting ratings today basically have no effect) and depending on the range of skill represented by the playing population.

That's a good point, I forgot about RD.  I do believe there is inflation though.  It's a shame that chess.com removed their bots, which gave a good indication of rating inflation over time since the bots always play at the same strength.  A way to test if there is inflation is to take a sample of games from a couple years ago at a given rating and compare the quality of those games (against an engine) with games in the same rating range played today.  

That's tricky though, because mistakes don't happen in a vacuum. 1200 vs 1200 in some boring french or slav exchange can easily be very high quality in terms of things like centipawn loss and how often the player's matched the engine's top 3. Meanwhile two GMs could play a sharp positions and have many blunders (if it's blitz).

---

Bots would also be tricky because not everyone played bots (I assume most people didn't). And those who played the most would figure out one thing that works, and win over and over to inflate their rating... although maybe this sort of thing was balanced by people trying to use the bots to sandbag, resigning to bots over and over. I'm not sure.

I agree with you on the bots.  Some people actually played the bots repeatedly, exploiting tendencies. Check out this thread here: https://www.chess.com/forum/view/general/computer-impossible-bullet-beat-down.  Ideally with the bot system, you wouldn't let players challenge the bots and you would disguise them in the pool so they end up playing random games against people.  

However, the trickiness you are talking about isn't really an issue.  With enough games in your sample (which you can decide based on game variance) your analysis can be rock solid, and that's just statistics.

Hah, it's sad, I'd almost forgotten what it's like to agree with something someone said here.

I agree with the points you made 

I think there is definetely something which feels like "inflation". Just looking at my opponents.

Can you please join my club sir?

Why? Moreover, this pretty much offtopic.

Okay

Can we talk in chat sir please