17

I'm wondering if there's any way to get an approximation of my FIDE rating by using one of my online ratings, say ICC or chess.com rating? If not, is there, at least, a correlation between how well you rank on those sites and your eventual FIDE rank?

  • 1
    Just answering the title: the correlation is positive. – Evargalo May 17 '18 at 11:47
  • It really varies depending on the online site. But online rating is usually quite a bit higher. – Inertial Ignorance Sep 28 at 23:12
8

I asked this on ICC once and was directed to the following link:

http://www.chessclub.com/help/ratings

IMHO, I don't think there is a failsafe way to approximate a FIDE rating based on online ratings.

7

Online and FIDE ratings can't be directly compared or converted with any real accuracy. They have a different pool of players and different constraints. I'd be surprised if the actual mathematical rating method was exactly the same.

The 'pool of players' issue is one of practical import. It turns out, players' geographical isolation highlights an under-appreciated fact about FIDE's rating system. The FIDE rating system doesn't tell you how good you are; it tells you how good you are compared to other players. Thus the ratings of isolated groups of players are effectively unrelated.

4

NM Matt Jensen has done similar research in 2014, in 2015 and later in 2016 using data from survey with over 370 surveyees. The survey shows the correlation between Chess.com, Lichess.org, USCF and FIDE ratings in Bullet, Blitz and Rapid/Classical categories.

The results can be found on Google Sheets. Here's the line chart showing the correlation between ratings:

enter image description here

  • How should we interpret that graph? It does not seem to show correlation at all... – Federico Poloni May 18 '18 at 10:44
  • 1
    @Federico, if your Chess.com Blitz rating (red curve) were let's say 1200, you should locate it in the graph and see that it correlates to Lichess.org Blitz rating (purple curve) of approx. 1540 and to FIDE regular rating (lime green curve) of approx. 1280. The vertical axis represents the ratings and the horizontal axis represents the player. – gdrt May 18 '18 at 11:26
  • @FedericoPoloni: my Lichess classical rating hovers around 1600. Therefore, my projected FIDE rating would be about 1250, a difference of 350 points. To see this, follow the Lichess classical curve to its extreme left, which happens to rate 1600; and then drop straight downward 350 points to find the FIDE curve. – thb May 18 '18 at 15:37
  • I'm not sure we use the word correlation in the same way; I was expecting a coefficient between 0 and 1. – Federico Poloni May 18 '18 at 16:46
4

I took the raw data from the source linked to by gdrt and looked at some pairwise correlations, but not all. There are too many pairs to consider and not enough data for most of them, so I focused on chess.com blitz ratings vs classical FIDE/USCF ratings, which were the most abundant. I also looked at chess.com vs lichess blitz. You can find them at this Google sheet.

In general there is clear correlation between the ratings, with r2 0.7-0.8, but the standard deviation of the residuals is about 200. That means that to have a 95% confidence on a prediction obtained from the regression model, the error bar is roughly +/- 400! (Two standard deviations.) The correlation between chess.com and lichess blitz is slightly better.

You can find all the equations in the link above, but since the original question was specifically about predicting FIDE rating from chess.com, I'll also post that one here:

fide = 0.7360*chess.com + 458

Again, this is classic FIDE vs chess.com blitz, and the error bar is about 400 points so don't expect a very accurate estimate!

FIDE vs chess.com correlation

  • Brilliant analysis! Could you do the equations for rapid/classical as well? :) – gdrt May 17 '18 at 14:23
  • 2
    This is embarrassing: I completely misinterpreted the spreadsheet! It turns out that the "comparisons" tab which I used to derive the equations above is already a model, not the raw data, which is in another tab. That explains the suspiciously good fit! I'll update my answer later, but just to give an example, the correlation between chess.com blitz and FIDE regular has an r^2 of 0.75. – itub May 17 '18 at 19:32
  • I also overlooked that you used the model labels for your analysis. Hence the perfect residual sum of squares of 0. – gdrt May 17 '18 at 22:48
  • 3
    @gdrt, I completely rewrote my answer based on the raw data instead of the comparisons tab. I don't think there is enough data to say much about online rapid/classical or about offline rapid/blitz (well, maybe there is enough chess.com rapid but I got tired... :-) ) – itub May 18 '18 at 14:07
1

This is roughly the conversion for different sites. CC stands for Chess.com, and Li stands for LiChess.

1000 FIDE->1200 USCF->1100 CCBlitz->1250 CCRapid->1450 LiBlitz->1500 LiClassic

1200 FIDE->1350 USCF->1300 CCBlitz->1450 CCRapid->1600 LiBlitz->1700 LiClassic

1400 FIDE->1500 USCF->1450 CCBlitz->1600 CCRapid->1725 LiBlitz->1850 LiClassic

1600 FIDE->1700 USCF->1600 CCBlitz->1725 CCRapid->1850 LiBlitz->2000 LiClassic

1800 FIDE->1900 USCF->1850 CCBlitz->1825 CCRapid->2050 LiBlitz->2150 LiClassic

2000 FIDE->2075 USCF->2050 CCBlitz->1900 CCRapid->2200 LiBlitz->2250 LiClassic

  • 1
    Do you have a source for this? – Herb Wolfe Jul 5 '18 at 17:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.