Are you saying the ratings within the pool of players at chess.com are inflating as time goes by or are you using the word "inflated" to mean high compared to a different pool of players.
If the latter then you have to somehow work out the specific number to subtract from the chess.com rating.
Are you saying the ratings within the pool of players at chess.com are inflating as time goes by or are you using the word "inflated" to mean high compared to a different pool of players.
If the latter then you have to somehow work out the specific number to subtract from the chess.com rating.
I meant the latter. I don't think it necesarily matters whether the ratings match real OTB ratings since the conditions are different and the ratings will be comparable to other chess.com ratings. Based on the evidence I have seen between real OTB ratings and chess.com ratings though I think USCF ratings are approximately chess.com rating - 200 or -400, where the difference represents whether the player is better at OTB or correspondence.
This is actually a slightly different thing to the ratings that players start with though, since before they play any games we have no evidence what-so-ever as to what their rating should be.
A 
Hi
Firstly I explain the basics behind the Rating Deviation RD and then show how it is implemented here and why new players here are under-rated by about 100 to 150 points. By altering this there would be a knock-on effect to the entire site and all the ratings would become more accurate.
The Rating Deviation RD determines the accuracy of a rating and automatically increases over a period of time where a player is not playing, hence the accuracy of their rating becomes less because they have not played for a while so their strength may have gone up considerably or down considerably.
How the RD determines the accuracy of a rating:
Example
RD = 50
Grade = 1700
Glicko suggests 95% certainty that the players rating is between:
Grade - (2 x RD) and Grade + (2 x RD)
1700 - (2 x 50) and 1700 + (2 x 50)
1600 and 1800; and so because this player has a high RD their actual playing strength is within 1600 and 1800 with a 95% certainty, but we cannot guage it more accurately until they play more games.
How is RD calculated?
The formula for graded players is:
> min( ,350) ensures the RD doesn't exceed 350.
> t is the number of rating periods since last competition.
> c is a constant that governs the increase in uncertainty over time.
> RD2old is the old RD of the player.
So new RD = the square root of (RDold squared + (c squared x t)), where if it exceeds 350 then it is rounded down to 350.
350 is the default RD for new/unrated players.
RD on Chess.com
New players start with a rating of 1200 here. So using the Glicko system above we assume (as new players) their RD is 350.
RD = 350
Grade = 1200
Glicko suggests 95% certainty that the players rating is between:
Grade - (2 x RD) and Grade + (2 x RD)
1200 - (2 x 350) and 1200 + (2 x 350)
500 and 1900; which is true for the vast majority of players but it's not as often as 95% of the time. If it were then 95% of chess.com players would fall between these ratings and they don't. I would suggest it more accurate to either start players as 1300, 1350 or start them as unrated completely since the ratings on this site are inflated. 1300 or 1350 would match more accurately with the actual spread of ratings, even taking into consideration the probable daily fluctuation and using it as a margin of error.
Any thoughts?
A