Inflated Rating

 But to say they're "inflated" is to say they should be the USCF rating, which isn't the case. And choosing USCF as the norm is rather dubious, even if you could compare ratings from different pools. I would expect ratings here to be more accurate than USCF ratings because more games are played here.

 Ofcourse otb ratings are far more reliable in telling the chess strength/knowledge of a player than online ratings. Lets face it, you have chess nobodies online beating top professional GMs, need I say more ?

Well, Elo type ratings are all about relative strength, so it's always going to be arbitrary to choose the ratings of any particular pool as the norm. That being said, the argument for choosing USCF is that it gives, as close as possible, a common frame of reference - at least for American players. Since almost everyone has played USCF rated players at one time or another and has a concept of how those ratings relate to playing strength, then one can use those ratings as a rough benchmark in comparison of other ratings pools. 

 Actually, USCF uses glicko, not elo. And I fail to see that it is a more common frame of reference. I expect more Americans have played online chess than have knowingly played a USCF player of a given rating. And even if it is a more common frame of reference, I fail to see how that qualifies it as being a norm by which to judge other rating pools.

This from Wikipedia entry on the USCF:

http://en.wikipedia.org/wiki/United_States_Chess_Federation

" ...Recently, the USCF has transitioned to a rating system that was proposed by Mark Glickman. Glickman, currently the chairman of the USCF Ratings Committee, is a college statistics professor who plays chess and has written numerous papers related to rating systems. The current rating system as implemented by the USCF is still an Elo rating system, but with a sliding K factor..."

At any rate, a given ratings differences between players in Glicko ratings corresponds to the same ratings difference in Elo ratings when using ratings to relate scoring probability. So it's not as though the systems are incompatible in that sense.

As far as choosing USCF as a norm, I admit it's arbitrary. But being arbitrary hasn't stopped people from choosing, for instance, feet and inches as units of length measure since what is arbitrary may still be useful.  

 That's irrelevant. The ratings are a relative measure of your chance of beating other players here, not absolute measures of your skill. Therefore if everyone has the same advantage it won't affect the rating, and you can't compare ratings from one rating pool to another.

But for every game won because this advantage was utilized there is a loser who failed to utilize it as effectively if at all so while on an individual basis some users might be higher rated than their opponents than the would be under another set of rules some users will also be ranked than they would be relative to their opponents under another set of rules.  On the average it should roughly balance (taking out other real sources of inflation) and all we've really done is highlight a very good example of why we cannot compare ratings between pools.

Here is my estimate of correspondance between chess.com ratings and FIDE ratings (first # is chess.com rating, second is OTB rating) :

• 2700+ => 2300+
• 2500-2700 => 2200-2300
• 2300-2500 => 2000-2200
• 2000-2300 => 1800-2000
• 1800-2000 => 1600-1800
• 1600-1800 => 1400-1600
• U1600 => U1400

Mine used to be. I got from 1200 to 1560ish in about three games. i've been falling steadily ever since even though i'm getting better in LC =P

Is that for correspondence chess, long chess, or blitz/quick?  And what led you to this estimate?

This is true only if you think that style of play has a large effect on the outcome. That is, you would need situations where two players of different styles have reasonably even scores against each other, but significantly different scores against a third person.

If style of play has an effect on the outcome is it any more detrimental to the accuracyof a large pool as opposed to a small one though?

I'd say it's likely more of an issue when there are islands of activity -- cliques if you will, outside of which players rarely venture in search of opponents.  I'd say this is more likely to occur in a larger pool, and certainly in a geographically dispersed one.  One of the advantages of the online community is that the geography doesn't play a role, however I'd venture a guess to say that the groups here at chess.com might.

Someone, I forget who, did gather a dozen or so USCF member's ratings for comparison. If I can find that post I'll do an edit.

edit:

chess.com 'Online Chess' vs USCF 'Standard'

2250 1300

2050 1750

2050 1400

1900 1400

1800 1500

1750 1700

1400 1500

by likesforests

billwall also has done a poll but I haven't seen it.

 Actually,I was referring to the ratings of players who had never played one another.

For example,...   Let's say two players are about to play one another for the first time. One is rated 2100 'XXCF' and the other 2000 'XXCF'.

The greater the number of players in the 'XXCF' rating pool, the less useful the 'XXCF' rating is for the purpose of predicting the probable outcome of that particular game.

I'm not sure how much style of play would affect the rating of different size pools. However, without something like that large pools are no less accurate than small pools. It was just all I could think of that would be a problem for large pools: an interaction effect between the players independent of the effect of the two players themselves.

I have done some simulations with regional groups that have limited interaction between groups, but have not noticed a significant decrease in accuracy. I need to do more detailed simulations in that area, though. I just haven't had the time to do the programming lately.

No, I understood you. That's what I was saying is incorrect. As long as those two players have played a reasonable number of games, those ratings will accurately predict the outcome of the game regardless of the size of the pool.

 I am well aware that many chess players believe this last statement to be true.

I am not one of them.

Moreover,I wonder how many of them would continue to believe it,if they understood that the very fact that they believe it, is a serious impediment to the improvement of their chess skills.

Online turn-based chess is not the same as regular over-the-board chess. Therefore, it is only natural that your online rating would be at least somewhat different than your OTB one. I mean, on Turn-Based chess you get days to make your moves, and you even get to see your moves before you make them! If you don't do at least somewhat better when compared to regular timed games then you're doing something wrong.

I understand what you're saying, but the average number of games that players in the pool have played is important too (the more the better so long as it's not the unlikely case where it's all small, independent groups of people continually playing only amongst themselves). In practice, even in a very large pool, the ratings of two players who have never played one another should predict the likely outcome of games between them fairly accurately as there will not be that many degrees of separation between them.

i agree that to start with your initial high correspondence rating can be quite deceiving until the eventual big smack-down, quickly bringing us all back to earth fast