How correct is the chess.com rating

I find it probable that the chess.com rating is too wrong mainly because they place opponents of almost the same rating as yours. Whereas (Wins+0.5Ddraws)/(Wins+Draws+Losses) would surely be correct (without Beyesian inference if the number of games becomes large enough) if they placed opponents from all ranges of ratings, i.e. randomly.

The site uses the Glicko rating system which is a well tested and sound system. Correctness is only valid within the pools of chess.com players within each rating pool. Usually when people say there is something wrong, it when trying to compare to other playing pools, such as other sites, FIDE or a national rating organization. While there may some correlations, and estimated formulas to convert, that may or may not be valid for any specific player.

 Are you able to explain your sureness that it's correct? I find it impossible to see whether it's correct or not.

It seems that nobody can explain why it is correct or wrong enough compared to (Wins+0.5Ddraws)/(Wins+Draws+Losses), the latter with placing opponents randomly from all ranges of ratings. E.g, suppose that they used the rating (Wins+0.5Ddraws)/(Wins+Draws+Losses) but placing opponents only of almost the same rating. Would it be correct compared to placing opponents randomly form all ranges of ratings? I doubt so.

quest

It’s pretty close I’d say

Btw I think that the chess.com rating is better than lichess rating. My chess.com rating is 1650 whereas my lichess rating is 1800+. I don't know why.

And chess lovers. Please connect with me by sending me a friend request. I just want to get connected with more and more chess lovers across the world.

I'm not sure what you're saying... are you saying let every player play (almost) every other player many times, and then rank them according to win+0.5draw/total games?

That's... enormously impractical right?

Ratings are very good at predicting results. The basic idea is a 1300 player is expected to score a certain amount (depending on the opponent's rating). If more is scored the 1300's rating is adjusted up. If less is scored it's adjusted down. The adjustment is proportional to the overperformance or underperformance. This reliably ranks people after very few games. I believe the mathematical details of Glicko are free to view online.

When you quickly rank players you can pair them such that games are more enjoyable. Noob vs master isn't enjoyable for either player.

Neither is better. They both use good (and I assume very similar if not identical) math, so both are correct.

BTW, dude is a Harvard statistics professor, and Glicko (which gives similar results to Elo) is used not just by chess.com, but by many different (non-chess) organizations. This isn't some random system they pulled out of the garbage

It's not "impractical", i.e. with random selection of opponents it's like everyone has faced everyone, i.e. suppose 1000 (or even just 100) games of your opponents selected randomly. Some of the 1000 opponents will be very strong, some very weak, some medium, etc. Therefore the (w+0.5d)/(w+d+l) of these1000 games is close enough to the (w+0.5d)/(w+d+l) that would have been if all the 1000000 games would have happened, 1000001 players having faced each other once, like a soccer championship of 1000001 teams. And it is surely correct system in all respects. Whereas the Glicko Elo-like system for me is impossible to conclude whether it's correct or not. Especially that they place opponents of only almost the same rating as yours. Have you read the Elo-Hell? Is it that you get stuck in the same rating because of that, i.e. that you face opponents of almost the same rating as yours? By the way a perhaps irrelevant question: I started with 1200 and I lost 200 points in very few games. Have you got an idea how this happened? You lose or win more points when you have high rating facing an opponent of high rating?

In the beginning you can win or lose many points (over 100) in a single game. This is the system trying to get you to your correct rating quickly. It helps keep the system stable overall because it's essentially creating or destroying rating points... after a user is established, the exchange is (roughly) equal, meaning the amount gained by one player is (roughly) the amount lost by another.

Well, sometimes a player can improve a lot in just 100 games. So with your method that player would have to start over (those 100 results couldn't be used to accurately rate him).

Elo hell doesn't exist, at least not here. Yes I've heard of it.

How can I be sure it's accurate? Well the dude is about 1-million times more competent at math than me (he's a Harvard statistics professor). But on my own I admit I did some statistics on WLD data on chess.com users, and from what I can tell the system is amazingly accurate... like... really surprisingly good at predicting a person's future results based on past results.

However, you are not able to convince me that it is as correct as the system I refered.

Ok, but, I don't care, because that's not my job.

I already said the math is online. You can look for yourself if you're interested.

I don't grasp the math, but perhaps it is not necessary to grasp it in order to conclude on its correctness (it is just me that I cannot conlude it). There is a rough idea of what it does, e.g. in the system I described there is also an unknown variation which gives more or less points according to the strength of the opponent, but the unknown variation is unecessary when the sample becomes e.g. 100 because then it is like 1000001 players have faced each other once. There is also an unknown modification in the system I described that uses Bayesian inference to correct the rating to more true, when the number of games is very small, e.g, just 1 game. 

Well, I don't think I understand it all either. I haven't looked at it in a long time... but like I said, lots of people and places use it, and if you can't even understand the math then your objection doesn't mean a lot to be honest. There are plenty of math students who graduate every year, some of them play chess, I never hear anyone complain.

I don't see it as a complaint, but I think I proved that the (w+0.5d)/(w+d+l) is surely correct, whereas the Elo and the Glicko and the Trueskill systems remain to be proved on their correctness.  I also said that perhaps a very clever dude can without understanding completely the math, conclude and prove on those 3 systems correctness, first to himself and then to the relevantly stupid dudes like me with simple words.