We can't deny the reality as you do. The numbers don't lie - it's the basis, the facts. We didn't create the basis - the formula just interprets them.
"For that which must not, cannot be."
@mdinnerspace No, it's perfectly fine mathematically for lower rated players to have OTB ratings higher than blitz and vice versa for great players. Let's look at the equation y=2x-4. At the point (3, 2) the x's are bigger than the y' s. At the point (5, 6) the y' s are now bigger. Is the equation not a viable formula? No, not at all, it's just that the y's get larger faster than the x's. As for drawing lines through clouds, that is pretty much the heart of statistics.
Edit: I would be curious to see the standard deviation to see how much of an approximation this is. r-squared would be good, too.
Let's review the equation..
Based on the OP's median rating he found at Lichess, a reasonable starting point of players ratings..
.38 x 1078 blitz rating = 410
.48 x 1169 classical rating = 561
187 + 410 + 561 = 1158 estimated FIDE rating
a full 80 points above their blitz rating.
.38 x 2078 blitz rating = 790
.48 x 2178 classical rating = 1045
187 + 790 + 1045 = 2022
A full 56 points below their blitz rating
(note: the same +78 and +169 was used)
This implies that higher rated online players are estimated to have lower OTB FIDE ratings than their online blitz rating, while lower rated online players have a dramatically estimated higher OTB FIDE rating. (Of course it applies to a limited set, simply stated, the formula can not be applied as a general principle, unless it can be shown with verifiable evidence to be accurate) which has not been presented. There is no evidence that lower rated online players will have a higher OTB FIDE rating. In fact, the limited evidence suggests completely the opposite to be true.
Obviously the inherent problem with the formula is the constant "187". It is being added equally to the higher/lower ratings skewing the results. The "187" must be interpreted as a variable, with a lower number that corresponds to the lower ratings.
Simple mathematics. In any given formula with 2 variables (.38 x n + .48 x n) + y = A ..
y can not be a constant number (must also be a variable) to establish a consistent equation.
Entering different numbers representing n and then adding a constant number renders the equation useless.
WHAT EVIDENCE?
Sorry...
Anyway, let me get this straight. Your problem with the equation is that a low rated player will have a higher OTB rating than his lichess blitz rating, and a high rated player will have a lichess blitz rating higher than his OTB rating. This is NOT a problem! It just means that as someone's ORB rating increases, their lichess blitz rating increases at a faster rate. One lichess Glicko point is not the same as one FIDE ELO point. The equation given by @dudeski_robinson shows the relationship between the two.
@mdinnerspace
Edit: and as far as evidence, did you look at the graph given in the original post? That is not made up at all; it is a scatter plot of the actual data pairs of lichess rating and FIDE rating.
Also, as a simplified, real life example, let's look at temperature. At the freezing point of water, Farienheit is "higher" than Celsius. (At a 2000+ish rating, lichess blitz is "higher" than OTB.) At -50 degrees Celsius, the Fahrenheit temperature is "lower" (-58 degrees) than the Celsius value. (At a 1000+ish rating, the lichess blitz rating is "lower" than the ORB rating.)
Is our Celsius-Farienheit conversion equation wrong, then? No! Is the rating conversion equation wrong? Maybe, but not for the reason you have given.
@mdinnerspace
Entering different numbers representing n and then adding a constant number renders the equation useless.
No. That's called a constant. Mathematicians use it all the time. Do you remember the "c" in the quadratic formula? That's always a constant, a number not multiplied by a variable. Not the best definition on my part there, but still.
@mdinnerspace
#107, this "constant number" sounds somewhat mystic used by mdinnerpace as invented arbitrarily by the thread-starter. Can we call it intercept of the regression which sounds much more scientific? A well-determined intercept?
A natural "event" doesn't always go through "zero", often there are constants, offsets, intercepts - call them whatever you want, but you need them for a proper fit.
(Sorry, this is tough for a non-native speaker. Hope you got it.)
And back to my handy dandy example: the equation that converts Celsius to Farienheit is F=(9÷5)×C+32. Is the "+32" wrong? No!!!!!
This topic has been archived and can no longer be replied to.