inflated ratings

 

 I would like to make a modest suggestion about the rating system at chess.com, first I would like to talk about the starting rating 1200 which i consider a little unrealistic being the fact that most chess fans play at a rating between 1500 to 1600 I would like to suggest that starting rating starts somewhere within that range, I would also like to suggest to do something about inflated ratings, I don't think it is fair that some players get ratings of above 1700 just after playing less than 20 games, I don't think players should get ratings of above 1700 with less than 20 games no matter if their opponents had  high ratings, I think the points to be gain by a player before game 20th should be somehow limited or like in some other chess sites players are forced to play players within their ratings range before they could play higher rated players for example, it would be nice to limite players to only allow them to  play lower rated players than themselve and players rated 100 points higher, I think that inflated ratings don't really let the serious chess fans  measure their playing strength against others, and another thing about inflated ratings is that would take the website right to where other websites do, and it is that they become a ratings contest chess sites rather than a website where players come to enjoy and polish on their chess skills, I think that the reason we like chess.com is because it gives a place where we can stay away from the ratings contest live blitz chess sites where people compete for rating points at the expense of the quality of the game, i think adjusting the rating system to eliminate inflated ratings will really set this chess site appart from the others more superficial ones

just an idea!  what do you think? 

In my opinion, 1200 is the average rating of the new players here.  I have played dozens of new players (1200), and half seem to be slightly stronger than that, and half seem to be weaker than 1200.  The more games played, the more the ratings become more accurate.  I have the opposite feeling.  I think the ratings are under-inflated.  All the 2000 rated players are really masters with 2200 to 2400 ratings.  I have also mapped the real ratings (USCF, Elo, ICCF, BCF, CFC) of the stronger players, and almost all of them have ratings here much less than their real OTB or correspondence ratings.  I also don't mind the big jumps.  If someone has a 2000 rating after 10 games, then they had to beat some strong players to get there.  They will drop when they lose to someone else.  Everyone seems to be a class below their real ratings in my opinion.  If anything, I would like to see new players with real OTB ratings to use that rating or that rating minus 100 points.  But I think 1200 is the average of the new player here.  And its easy to tell.  Look at the quality of games of the new player at 1200 (or 1300 to 1100).  Openings are bad, pieces are dropped quickly, and most fall into well known chess traps. 

I think the starting 1200 rating is fine but I can understand your point about inflated ratings. Yes, maybe as you suggested limit players for their first 20 games to play people rated within 200 points of them either way, so if I was rated 1522 on my 20th game I would still have to play someone from 1322-1722. Then after you have an established rating after the 20 games then you are free to play any rating.

 

Seems logical to me. But what do I know...hehe

 

 yes Bill no doubt that many of the 2000 and above rated players at this site are CM or Masters, I just don't see anything wrong with making players keep their rated games within a 100 or maybe a 200 points range either way, I think that by keeping rated games within a 100 or 200 points range you will eliminate the 2000 ratings based on average much lower rated opponents and will also eliminate inflated ratings, I think it will make things more realistic, unrated don't matter much, unrated games could be let open to players choice.

just a thought!.. what do you think?

If there was a way to restrict the ratings to the true OTB or correspondence ratings, that would be ideal.  I also like the idea of a floor.  When I was 2200 USCF, then dropped to 2000 and could go no lower, it was fun to play chess knowing that I could not drop below 2000.  I played lots of games out that may have drawn, and I risked it and lost to up and coming players.  I wasn't playing for a rating.  Then USCF changed the rules again and took away the floor.  At the other end, I don't think anyone should have a rating higher than 200 points above the strongest player they have played.  In other words, someone should not get a 2000 rating if they never beat anyone higher than 1800.  They should be stuck at 1999.  I see some players who have high ratings but have not played anyone stronger than several hundred points below them.  So if you want to go over 2200, you have to beat someone 2000 or better, but not a bunch of 1900 players or below.

 

 yes, that what you explained above is kind of where my ideas for this topic came from

Why not start everybody above the national average at 1700 and let them work their way down instead? If they lose to a higher rated player the higher rated player could gain minimal points and the lower player could be adjusted accordingly.

great discussion and some really good points have been made. here is where we stand:

you must pick some midpoint/average rating to start with. and, as Bill pointed out, Chess.com attracts slightly more new players than other sites might. 1200 is also the midpoint between 0 and 2400, which is probably the range on our site (with some outliers).

as for the ratings moving around, you'll have to take up issue with Mark Glickman. :)  he is the professor who created the Glicko system (see here). you should also read about it here: http://www.chess.com/article/view/chess-ratings---how-they-work

the ratings pushing up far higher above the best person they have beat comes from mathmatics that I don't entirely understand to perfection, but it has to do with the RD (ratings deviation). once you have played a lot of games, that number comes down and your rating moves in a more predictable fashion. we are only seeing wild swings in the first few games - much how a provisional rating in the USCF works. we may be able to implement the idea of a floor, but i'm not sure. i also am not a huge fan of people only playing within a certain ratings range - why not leave that open to the players themselves?

 

thanks so much for bringing this up DeepNF3! we're trying hard to make this the best possible chess community :) 

 

 

 

 my pleasure!, I like this site a lot

I have to disagree with Bill Wall about the ratings being under-inflated. Maybe above the 2000 level, that's true, but they're definitely over-inflated at lower levels. My rating here (1700ish) is a little more than 300 points higher than my USCF rating. But that's consistant with other internet sites. At FICS (freechess.org), my rating in slow games (usually 30-75 minutes per game) is slightly higher than my rating here.

 

And most of the opponents I've played here rated at or below my rating are severely over-inflated, as well. I started with a 1200 rating here and challenged people with 1300-1500 ratings here, since that's near my USCF rating. I beat everyone I played in that rating range, most of them quite easily. I played one guy with a rating in the 1300's here who moved his queen on something like 6 of his first 9 moves. That type of play would be lucky to earn a 4 digit rating in USCF tourneys. 

 

I do agree with the idea of a ratings max based on your opponents. If you win or lose against someone with a rating too far away from yours, it shouldn't affect your rating. Why should a master gain points for beating a 1200 player?

 

--Fromper 

Fromper,

I stand corrected.  However, lets look at your case.  You are around 1400 USCF (is that established rating or not).  You are around 1700 here, but based only on 11 games.  That means provisionally, you are beating 1300 rated players to get a 1700 rating (you get about 400 rating points when you beat someone by taking his rating and adding 400 when you start out provisionally).  Now play 10 more games with someone rated over 1800, and if you lose to them, it averages out to 1400.  Let's see what your rating is after 20-30 more games with people rated 1800-2000.  I bet it averages out to 1400 to 1500.   Also, you played one person 3 times.  You need to play a variety of players.  Also, when was the last USCF rated tournament you played in?  Since you did not give your true name, I cannot look it up with USCF.  Maybe you really are a 1700 rated player now.  Finally, maybe it is skewed that it becomes more accurate with the higher ratings.  It's just that my experience is that half the 1200 rated players I have played are 1200 or weaker, and half are 1200 and stronger (and moving up).  But established ratings of 30 or more games tend to become more accurate.  

Most of the ratings are inflated here. This is just a matter of math. Everyone comes in with 1200. A lot of them play , time-out, and never come back. Now all the players who won a game are 1400 players boom! just like that. Same pattern for the second game: half the players below 1200 but who won on time-outs are over rated.

As the site ages this will become less of an issue. But the newest players will always have skewed ratings. It can work the other way as well if a group of experts or masters all log in and play each other we get half or them rated low because they lost (actually at this level expect more draws ^_^). It just happens that we get more beginners than masters probably because there are more beginners than masters.

Until the site ages a bit more its going to have a majority of players that are new and thus have inflated ratings.

But what IS an inflated rating anyway? Are we comparing chess.com ratings to non-chess.com ratings? This comparison works best if they have the same formulas and the same starting rating and the same population. They do not! The chess population for any Internet site is generally lower rated than a real world chess organization. That's because a lot of people who spend the travel time and gas on visiting a chess club don't mind taking a few clicks to visit an online site.

Is any of this making sense? Good. ^_^

Because it is also a matter of geography. In Texas someone who drops pieces and falls for opening traps is a class C 1400-1600 player. I was a candidate master in Texas (about 20 years ago) but in New York (especially near the Marshall Chess Club) I was much closer to 1850 than 2050. This is based on small sample though so its way too exaggerated to be thought of as exemplary of the ranges. But it is exemplary of the rating slant since similar ratings in New York beat those in Texas.

Could this have been a fluke? Sure, why not. Could the inflation or deflation of ratings on chess.com be a fluke too? Sure, why not. ^_^

Right now I estimate my ELO to very close to my chess.com rating ~ 1700+. If you take the time to look over my games you might rate me as low as 1500 or as high as 1800. (How the mighty have fallen!)

Basically: No two ratings are equivalent for all players except as mathematical coincidence.

Personally I would have preferred a rating system more useful like 50 points to start with 100 being exceptional. There has NEVER been a good reason for using the numbers used (I'm looking right in your eyeballs Mr. Glicko).

But then I also prefer to use the number of games played to measure fun levels rather then the quality of the games so what do I know? I'm having more fun than most of you WHEEEE! ^_^ ^_^ ^_^

I looked at ratings of 21 people rated over 1800.  Only 4 had a higher rating here than their real Elo or USCF rating.  People like Dembo, Volvk, Pernnoir, Mont-Reynaud, Schiller, Gill, Milanovic, and Khmelinitsky (all title players) have real ratings hundreds of time higher than here.  Here is a sample list of the rest (Elo or USCF rating / chess.com rating): Wall (2200/2169), Rogersky (2210/2107), BruceHayden (2200/2079), rowrulz (2265/2057), chessiq (1964/2048), erad (1891/1994), montey (2297/1890), littleman (1646/1888), Creg (1937/1838), startstek (2168/1844), matthelfst (1784/1843), and Jay (1850/1836).

Maybe you could just place a ? symbol after the rating of members who've played less than a certain number of games so those who care too much about ratings have a quick way of determining if it's likely to be accurate. 

pawnshover wrote: In Texas someone who drops pieces and falls for opening traps is a class C 1400-1600 player.

---

*boggle* Sounds like USCF ratings in Texas are highly inflated compared to here in Florida. For instance, I'd be afraid to try Blackburne's Shilling Gambit against anyone over 1200 here, for fear of getting my butt kicked.

 

My USCF rating is well established, including 22 games since returning to tournament play 3 months ago. I've dropped a little, from staying slightly above 1400 for an entire year in 2003 to around 1340 today, partly because I was a bit rusty when I first returned to the game after a couple of years away, and partly because I've been trying to completely change my style of play. I used to be a very quiet, defensive player, but I had no clue what to do in the middle game. So I've taken up a more attacking opening repertoire, including mostly gambits, in order to force myself to learn to attack. In the short term, this is causing me to lose more games as I adapt and learn, but I'm sure it'll make me a better player in the long run.

 

As for my games here, I haven't just won against players in the 1300-1400 range. I've really slaughtered them. My USCF rating is in the 1300's, yet nobody below 1500 on this site has really given me a challenge. I don't intend to play my next 20-30 games against players above 1800, as you suggest, because I would probably lose almost all of them. I'll play some people rated that high, and some in the 1600-1800 range. I suspect my rating here will probably stay around 1700 for a while, give or take a few points here or there. And as I said, that seems consistant with other internet sites that are known for inflated ratings, as well.

 

--Fromper 

Hi im littleman/ian my Australian chess federation(A.F.C.) rating is 1646 and this site its currently 1888 much higher then i thought it should bemind u it has been awhile since i played a tournament so maybe i got a little better since then because of this and other sites, i have only once played in a OTB tournament where my rating was over 1850 and i had a couple of 1700 ones but even considering that i still think the ratings are some what inflated my friends but who cares to be honest just play and enjoy the game thats what i think anyway...

So it doesn't help the ratings inflation when a player rated almost 1900 resigns against me because he hung a pawn. I still thought he had a better position overall. Why do people resign so early?

 

--Fromper 

My friend thats what seperates the good players from the great ones there fighting spirit... They must have thought i lost this so its just a matter of time before i will lose the game since they know that they would be very likely to winin your position so they decided not to waist time and energy on it any further, but still i think we should fight untill there is no chance at all and i mean none!!. other wise we learn little to nothing on how to defend a weaker position and that lessens our overall skill in my personal opinion

I usually don't resign to early but when I do its because of the psychology of it. Some positions are not fun to play. Down a couple of pawns really hurts me a lot more than down a piece.

This is because good chess players use their pieces while I push pawns. It's too bad pawns don't have the option to stay pawns when they reach the other side.

"fail; fail again; fail better" - Samuel Beckett

 :)