How correct is the chess.com rating

Oya Da

Given the amount of evidence that was given to you over there (and over here as well now), I think this is the only conclusion. I have to admit that the math can get quite complex, but it checks out. There's nothing hidden and those are pure facts, which can be checked. 

The thing about math is that things can actually get proven there. Without a doubt, pure proof. The Pythagoras theorem was proven. You can't say that in your opinion it's wrong, it just doesn't work that way. The literature just proves it. It's not opinions, it's facts.

One soccer team has 15 wins, 10 draws, 5 losses and another has 14 wins, 10 draws, 4 losses. They both have w-l=10, whereas the first team has (w-l)/(w+d+l) =(15-5)/(15+10+5)=1/3 and the second has (14-4)/(14+10+4)=5/14. The rating  (w-l)/(w+d+l)  shows that the second team is stronger than the first, as 5/14>1/3. The (w-l)/(w+d+l) is equivalent to (w+0.5d)/(w+d+l), this is easy to prove, so one can choose which of the two he prefers.  The (w-l)/(w+d+l)<=>(w+0.5d)/(w+d+l) is blind obvious that it is a correct rating, whereas Elo Glicko and Truskill all the experts say it is also correct and I say God surely knows whether it’s correct or not (I do not say any more that my poor understanding says it is most probably wrong because it appears to count w-l, because at physics.com considered this as rudeness).

So basically you're saying that if you can't understand something, it must be wrong? Well, I guess I'll fly away now (since I don't understand everything about gravity). 

I told you that it appears to count (initial rating)+8*wins-8*losses, i.e. wins-losses. A physics.com professor says that he proved I am wrong but I did not see that, and because of that locked the thread. And in the best case is under doubt whereas the (w+0.5d)/(w+d+l) is surely correct. He got also furious even if I just I doubt it "whereas the math shows it is surely correct"! And here, they say it is proven like the pythagorean theorem thus I should trust the opinion of professors without doubt.

The funny thing is that they actually did prove it on the physics forum. So no, it's not under doubt at all. Again, if you don't understand the proof that doesn't mean it's incorrect. But don't feel bad. There's actually whole groups of people doing things like this. Flat earthers for instance are a perfect example. They don't understand the proof for a round earth, so they say it's flat.

In fact, it's just human. In many ways it's good to be sceptical and don't accept something because someone says it. There's a big difference between scepticism and ignorance though. Saying you don't accept something because you don't have enough information yet and need to study it more is perfectly okay, saying something is incorrect because of the same reason just says you don't want to learn and have your mind made up.

So this is where I stop my contribution to this topic. You're looking for something that isn't there. We can't help you with that.

hi

That all who write disagree with me, is fine. The bad thing is if someone takes my discovery that Elo etc appear to count wins-losses, and claim it as his, i.e. not mentioning that I made the discovery, i.e. my name or at least my nickname.  Also, the surely correct system (w-l)/(w+d+l)<=>(w+0.5d)/(w+d+l) I haven't seen it somewhere if I remember well, so quite probably this is mine too. By the way, w-l appears more correct than (w-l)/(w+d+l) when the number of games is very small (!), for example consider 2 players, one has 1 win in 1 game and another has 80 wins, 15 draws, 5 losses in 100 games. The first has (w-l)/(w+d+l) =(1-0)/1=1 and the second has (80-5)/100=0.75, whereas obviously it is not the first one who is stronger, but the second, and this is shown (!) by the w-l as the first has only 1-0=1 and the second 80-5=75. Thus (w-l)/(w+d+l) and (w+0.5d)/(w+d+l) are generally more correct that w-l but need an unknown modification when the number of games is very small. And I remind you that the rating I discovered needs the opponents to be selected randomly (when the opponents are too many) for reasoning I explained with the example of the premier league and league 2. AND Elo etc for the same reasoning needs that.

P.S., how do you compare two players at chess.com and at chesstempo.com? At chess.com I am stuck at ~1050 and at chesstempo.com I am stuck at ~1500. Which of the 2 is my true rating? I bet they are in-comparable, e.g. perhaps they don't use Glicko at chesstempo. At chess.com I started with 1200 if I remember well, and at chesstempo.com I started with 1500, lol.

I lost 270 points at chesstempo in 6 -7 games because it was the first games after a long time that I hadnt played. Now this is definitely also WRONG. E.g, I actually improved since the long time I hadnt played, and this was shown including the next (of the ~6-7) games, but my wins counted crap.

I feel like its somewhat accurate to uscf but no too accurate to fide

I play at chesstempo.com (my account is restricted at chess.com), and  I just beated one rated 1740 whereas I was rated 1582, and I earned 6.1 points and he lost 43.2 points because of that win. Do you know why I didnt earn as many points as he lost? Edit: It seems that it is not  because this was of his first games after a long period that he did not play any games, nor that he was a new player, as this has happened again in all the next games I played.

If they the rating was just wins-losses (as Glicko essensially is) it would be more correct than the WRONG gaining or losing huge points when you are a new player or have a long time to play. 10 consecutive wins or 10 consecutive losses only, are determined a lot by the fluctuation of luck and should not determine the rating. Luck, i.e. in other games you blunder a lot and in other games a little, and it is unpredictable in which games you will blunder a lot and in which games a little, thus in which games you will blunder a lot and in which games a little is determined by luck. Thus you might be a weak player and have 10  wins in 10 games and you might be a strong player and you might have 10 losses in 10 games. Thus the sample of only 10 games is incapable of identifying player's strength.

LOL bro.

I gess the problem is your definition of correct.

What does correct mean to you ?

To me if a system predicts with accuracy who is the stronger player based on its rating, then the system is "correct" enough.

LOL bro. You were thinking about this answer for two years?

It is extremely inaccurate my rapid rating is 600 but it should be around 1300

The rating system is a way of pairing you with an opponent of similar ability. Nothing more.

Given that I have won 49% of my games over a 5 year period, it appears to work.

Example:

Suppose I get sent to prison. Naturally, I can play only against other prisoners. So if the average playing strength of the prison population is very low, then I can run up a rating worthy of a Grandmaster because my (Wins + 0.5 x Draws)/(Wins+Draws+Losses) is close to 1.0

yes

of course