How does chess.com prevent rating deflation?

My game has improved significantly (no but really) since I started here several months ago, but my rating has declined from 1325 to 1275 (-50pts). In in-person games, I've dominated my old opponents.

Two of friends are on chess.com. During the same period, one has gone from 1475 to 1325 (-150pts). The other has gone from 1125 to 975 (-150pts).

(And those numbers are stable-ish averages, not the extremes, e.g. my highest rating was 1371.) 

Interestingly, against the chess.com bots my rating would easily be 150 points higher than my actual rating.

I know I'm just some guy on the internet with dumb anecdotes, but as chess.com membership surges and members get better -- especially in the lower tiers -- what is the mechanism to prevent rating deflation?

This isn’t the exact answer to your question, but I observed that based on your stats for Rapid, tilt might be to blame for your rating decrease.

Possibly, you start to get frustrated with your losses, maybe you have a bad day, and you play game after game without analyzing. As frustration builds, you want more and more to win back your elo, but you aren’t in the right state of mind to do so.

I noticed that your rating increases and decreases occasionally 100 points each week. After hitting your peak rating, you dropped about 300 points. Since then every time you hit a peak, you drop again. A possibility could be you simply need to stop playing when you start getting frustrated. This can also happen when you are tired, hungry, impatient, etc.

Alternatively, that might not be your problem at all (I do not know you personally). The thing is, 50 points of rating is nothing. A player rated 2000 will be of practically equal strength to a player rated 2050. If you assume both players are playing in their best condition of mind, I believe it would take at least 200 elo for a significant skill difference.
Of course, a 2400 IM is going to be weaker than a 2500 GM in most cases. But for us non-masters, that’s not always the case.

And the third, more likely possibility is that you and your friends ARE much stronger than in the past, you simply are not performing to the extent of your skill.

I’ve lost 70 rating points in the past week, yet I know I have improved in my openings and positional play. If you enjoy learning more about the game, and put the effort into it, the rating point boost won’t be too far away 😁

As you point out, your anecdotal evidence doesn't prove that there is any deflation.  As far as I know, chess.com hasn't published any data on it. 

However, they did do a 150 point upward adjustment in bullet ratings last fall.  https://www.chess.com/news/view/10-minute-chess-now-rapid-rated-bullet-ratings-increased They stated they did it because of a discrepancy between chess.com bullet and blitz ratings.

The bot ratings are not a good measure since they don't adjust. I have never heard they were set by playing rated players. Many people have made posts asking why they can beat some 2000 rated bot but lose to 1200 rated players.

And it really doesn't matter as long as your ratings stay relatively accurate in relation to the people you are playing here.

The ratings system here provide adjustments based on statistical measures of estimated accuracy.  The Glicko RD (rating deviation) estimates how accurate your rating is. 

 

> This isn’t the exact answer to your question, but I observed that based on your stats for Rapid, tilt might be to blame for your rating decrease.

It's not hard to go up and down 100 points in over only several dozen games, especially at the lower levels where opponents are so erratic. And I'm learning and trying new techniques. Variation is natural.

I won't say you're point is completely wrong, but this is over 900+ games spread fairly evenly over months.

And I play about the same # of games whether I'm losing or winning.

And regardless, any my flaws on my side wouldn't affect the other examples.

> As you point out, your anecdotal evidence doesn't prove that there is any deflation. 

I hardly think I *proved* there's deflation. I merely suspect there may be, and I'm curious how chess.com prevents it.

Which question you conveniently sidestepped to just attack my post.

> The bot ratings are not a good measure since they don't adjust

That's actually my point...the bots would be a consistent benchmark. Humans (especially < 1400) don't. They fluctuate, individually and as a population.

I'm not saying anything is definitely wrong. I'm just asking how chess.com ensures rating stability while its community changes.

It'd be great if someone actually had an answer to my question.

Other then the mentioned rating injection, which have happened a few times over the years, I don't believe there is any active methods to combat perceived deflation.

 

In blitz, where I play most often, my rating has remained relatively steady, within a 100 point range. That's pretty similar to how my OTB rating has been over the past few years.

>Other then the mentioned rating injection

Oh, yeah, for bullet. My examples weren't bullet, but good to know.

I share the same concern. I’ve talked about it a few weeks ago but except moderators no one was concerned (is there anyone there?).

I think there is a systematic rating deflation and they just don’t announce it. when new users join, ratings should inflate, but chess.com prevents new user inflation by some means that we don’t know.

You can look at your percentile instead of rating to understand whether you have improved, well, I don’t know if that’s accurate (given amount of ghost and bots), but at least maybe it is more accurate than rating.

I think you answered your own question. The player base is growing drastically. ELO is a reflection of the competition you are playing. For instance, I only got back into chess 5 months ago, but have already climbed into the 1500s. 6 months ago I, and many others were not on the website, which means the competition pool was easier and so your rating was higher.

For instance, if you are by far the greatest player amongst your group of ten friends, you are the GM of your friend group. But expand the player pool, and you might only be the 90th best of 100. Your ELO would be significantly lower, even if you were still just as strong.

Admittedly, I'm definitely not an expert in this area.  Nonetheless I believe that (other than establishing thresholds for titled players), the primary purpose of any rating system is to predict the most likely winner between two players who are using the same rating system. The absolute number of an individual's rating doesn't particularly matter. It's primary purpose is relative, not absolute.  So chess.com may not care very much if average ratings as a whole are decreasing.

As an example, I believe chess.com uses the Glicko rating system, whereas lichess.org uses a newer rating system called Glicko 2.  I think most would agree that Glicko 2 ratings are generally higher that Glicko ratings, and that the Glicko 2 rating system is a better predictor of who will win the game.  But I don't think you can use one player's Glicko rating and another player's Glicko 2 as a useful predictor of who will likely win the game

Using that logic, I don't think it really matters if your chess skills have increased while your rating has decreased.  That may feel emotionally discouraging because many of us use our rating as an indicator of improvement (or not).  And I suppose chess.com kinda feeds into this mentality by showing us our rating change over time.  

Perhaps a better measure of improvement is if you begin beating players with higher ratings than yours?  Also I think your stats page shows where your rating sits as a percentage of all members of the site.  If that percent is increasing, that's a good indicator.  Unfortunately, chess.com doesn't graph that % over time.

Well, I hope this is of some help.  

ELO is an algorithm. It is designed to calculate a relative rating for zero sum games between two opponents. There exists no such thing as a “rating deflation mechanism”. Anyone’s rating can bottom out at 100. 
On the other end- there is no set limit to a maximum rating. Current trends are inflating ratings through the roof as the numbers increase and players playing longer.

As a bit of an aside. Chess.com doesn't use the ELO rating system (which is used - in part- to establish FIDE Titles).  I don't believe anyone's chess.com rating can properly be used for any purpose other than comparing their results to other chess.com players playing games in the same class of time format (bullet, blitz, rapid, ...)  And as I mentioned earlier, your chess.com rating number only has relative significance among that population.  It's absolute value is irrelevant. 

So I believe the bottom line to the OP's question is that chess.com is unlikely to do anything about general rating deflation (assuming such a thing is actually occurring).  They have at times made significant offsets to change the ratings of large numbers of players in certain groups (such as significantly increasing the ratings of all blitz players depending on ratings), but I believe these large changes were made to better align players blitz ratings to their rapid ratings.  In fact a couple months ago, chess.com implemented a large increases in the ratings of daily chess players.  I think the change was rescinded after about a day due to member feedback about the unintended downsides.