@Sarg0n Very cool! Thanks for sharing!
Adding a quadratic term will almost always improve model fit, but the difference is pretty trivial. Look at this graph, where I plot your quadratic formula on top of a linear approximation:
https://imgur.com/a/43k8j
Over the range of reasonable DWZ ratings, the predictions of the two models are almost always extremely similar. Interesting nonetheless to play with different models.
The second graph you posted is also *extremely* interesting. In particular, the spread in the scatter plot of DWZ vs. Fide is similar to the spread in the scatter plot of Lichess vs. Fide. The correlation seems slightly higher in the former (ignoring the made up data points), but probably not by that much.
This reinforces my intuition that Lichess ratings are actually a pretty good proxy for OTB skills.
Probably a trivial reason: a lot of German tournaments are evaluated nationally, but weaker players lack Elo tournaments.
Sometimes the ratio is one Elo tournament within a couple of years in contrast to a couple of DWZ tournaments within one year.
Btw, my current DWZ=2139. Actually lot of players with the same DWZ have a higher Elo or put it the other way: lot of Elo mates have a lower DWZ, which is reflected in the formula (Elo -> 2200)
I like it dudeski. I remember a discussion we had some time ago about rating comparisons, and someone mentioned that the Glicko2 formula Lichess uses starts everyone at 1500 where the Elo formula in use now starts people at 1200. I thought this was a salient point that people were missing. So, I've always thought maybe your Lichess rating was about 300 points too high. From what I've heard (and calculated with your formula) this may be a real, real rough rule of thumb that holds water - and you don't need a calculator.
@Sarg0n makes total sense.
@LDog11 Glad you found this useful. If you don't want to use the formula and you need an even rougher rule of thumb, it should be something like:
Fide ELO = Lichess Classical - 170
or
Fide ELO = Lichess Blitz - 80
btw, what is the most updated formula?
is it still
FIDE Rating = 187 + Lichess Classical Rating X 0.38 + Lichess Blitz Rating X 0.48
?
@mathtuition88
I consider that none of the critiques made in this thread pose a significant challenge to the original formula (given all the caveats already expressed in the original post). So yes, that's still the formula I would recommend using.
Plug in the numbers
F = 187 + CL x .38 + LB x .48
The result should be very close to
FIDE = LC -170 and LB -80
So the formula should predict FIDE 1700 for LB of 1780 and LC of 1870 (OP's post #204)
Guess what? It predicts 1751 ... 51 points higher
Plug in the numbers for a LB rating of 1080 and LC rating of 1170
It should predict a FIDE rating close to 1000
Guess what? It predicts 1149 ! a whopping 149 points higher.
The lower the online rating, the estimated FIDE rating accelerates to higher numbers above what is expected.
Plug in the numbers for a LB rating of 2480 and LC of 2570
It should predict a FIDE rating of 2400.
Guess what. It predicts 2353. ...47 points lower.
The higher the online rating, the estimated FIDE rating decreases at an accelerated rate than what is expected.
Higher rated online players, using the formula are estimated to have a lower FIDE rating than their online ratings, while lower rated online players using the formula are estimated to have higher FIDE ratings than their online ratings. With an online blitz rating of 1080, a FIDE OTB rating of 1149 simply is not the case.(except maybe in voluntarily given profiles which the OP has based his "evidence").
Make sense to anyone? The less skilled online players will fare better in OTB play (who have never played OTB before) and the more skilled and experienced will fare worse than their respective online ratings, according to the formula.
@mdinnerspace
There is an empirical way to assess this ridiculous line of criticism that you keep trotting out. I have addressed it explicitly in post #196 and #197. See here:
https://lichess.org/forum/general-chess-discussion/how-to-estimate-your-fide-rating-conversion-formula-inside?page=20#196
There are many players rated around 600.
Lets use the formula to predict their FIDE rating by their online Blitz rating of 680 and Classical rating of 770.
+80 and +170 ... follow so far? We should expect an estimate somewhere close to 600. Correct? The OP made the claim his formula is accurate "within a few points".
Guess what the formula predicts?
A FIDE rating of 805 !!
205 points higher. (115 points higher than their online blitz rating).
So a 670 online player, having never played OTB, should expect a FIDE performance substantially higher than their online performance?
Nonsense
@mdinnerspace
A) You didn't address posts #196 and #197 at all. In those posts I explain why your concern is unfounded.
B) You say "There are many players rated around 600". That is wrong. In my sample of 196,713 players who have played at least 50 classical games on Lichess, there are exactly ZERO players with a classical rating under 700.
The formula was derived from the observed data. Obviously, it cannot be expected to make accurate predictions in ranges of ratings that are *not observed on Lichess*.
In post #197, I calculated predictions for ALL LICHESS PLAYERS. I show clearly that the predictions are reasonable.
You can make up crazy examples all you want. I have demonstrated that your examples don't show anything useful. They are edge cases that apply to a tiny minority of Lichess players (or to none at all).
This topic has been archived and can no longer be replied to.