Any particular reason why lichess rating is sooomuch higher/differnt from FIDE??

#19 Thank you for that plot. great and fine to put as argument. yes I struggle putting mathematics into words... But i keep trying. I hope my word images are fun to consider though.

Yes I know scattered, but not uniform. And as I understand these are individuals that have both ratings. Blitz being the most populated (from looking at the plots).. I find that blitz tells even a linear story. I stand corrected. The other ones don't seem to be populated as much. (or it is just that blitz being tighter, makes it darker). One graph like that and it is enough for me to get some idea of the relation, i would dismiss the others as not enough data. and the scatter might be explained by the difference in the underlying statistical models for individual ratings... ELO might be assuming a static distribution with no volatility (don't know, but may have read that).

So to answer the op. The reason is the initial condition (and the increased flexibility of glicko-2 over ELO), and different population make up. For from previous blitz graph, you could hand wave as relationship.

I think choosing an average of 1500, makes more room below. As I have not yet read how the bell shaped Gaussian probability density of individual rating being defined over the whole real line, does not yield possible negative values. So pushing to 1500, might be a way to have only positive ratings. It appear to me that 1500 is arbitrary. I should read about ELO underlying model if it can be view under that same dynamic perspective (individual bell shape evolution, upon successive collisions as pairings).
But now with your plots. I don't feel it to be priority. you have the best answer. They are different, but not unrelated. It might also mean that bullet is more "random" and does not hold as much meaning in any system anyway.

and classical, sadly, shows how blitzy lichess population might be anemic (in numbers i mean). I can take as much room with words as your plot image... but you win.

When you give a bajillion points to new accounts, what do you expect?

I have posted on this before. You should get a base 10 points for a win, 10 points for a loss. Get more wins, get more points, 10 wins get you 100 points or a ratio similar (11 games, 10 wins, 1 loss). Then as you go up 100 points you earn less points, say 8, then 6, and then 3 when playing people at your level.

When you reach 2000, it then enters a "Big League" rating system more in tune with FIDE. If you are that good, you have nothing to worry about.

This would help those 1500-2000 because they get trampled on from the 2000+ players who start accounts at 1500? and it would avoid the retards who got lucky their first few games, so they have a 1800 rating when in fact they are more like a 1400 player.

I do have my own questions about the different population make ups between FIDE data and lichess data.
populations of ratings.
They are about game context, and individual rating trajectories, (in transitioned, plateaued, etc....) , and time scale of collisions.

Isn't FIDE strictly data from tournaments? grouped salvos of collision frenzy (with some meta-game of chess within).
How are the collisions spread over time... and paring distances statistics.

While lichess might have a more dense and better spread over time statistical characteristics of pairing. So while the scatter plot is about same individuals, the pools their ratings are based on may have completely different characteristics. I would not trust FIDE over Lichess. They may even be about different games (depends on lichess weight of the tournament data over non-such structured data, games with more individual weight in rating stats, games sequence more statistically independent from each other (tournament are clumps of games, with their own strategies, linking them).

If one could models the differences in dynamic characteristics to include above factors (and the pairings are documented in time, so are the rating trajectories i would guess from both pools), the scatter might get even more reduced under such corrected coordinates to plot.

there is no need to make room below the rating is logarithm of "power rating" ans negative ones work just as well as positive.

And rating are lamost normally distributed. On fide NOT as they still have reltively high cut point i.e low end tail is missing.

Strength of player is assumed to be normally distrinbuted random variable with mean on strenestima standard deviation of RD. Pretty simple maths at the end of the day. Much better systems are available but Glicko-2 is good enough and actually in Elo system is good enough with high amount of games played online.

As for almost linera mapping between blitz and FIDE: it is to expected as well it can be assumed to best correlelated simply because it most played here. in particular strong player do not play rapid or classic.

Also an issue is there good be pools within pools. people playing 3min blitz might not play 5 min blitz at all making these pools to some degree separeted and could have different mapping

@petri999 ..you made a point that rating only exist to pair players of a similar strength... so is it possible that events like the candidates tournament will be at some point be run based on the lichess rating of the players... and if so what happens when you want to achieve a GM title... when will lichess start issuing their own GM titles

No rating will not be based on any online rating. Candidates tournament is played of long time limits over the real board with arbiters supervising the event hence the event will be run on rating earned on long time limit games over real board. Also lichess rating are based on single games not single tournaments. Struggling in 9 round 9 day event is quite different stress level excercise.

Obviousli Lichess could create its own titles but why? Actually there was LM title and there are people with it just that it is no longer awarded. Very hard to thing title by an online site could gain such prestige that it would make any sense. FIDE arena has its own title system . and limits are so low that most of people could get if they really want to.

I do also feel the need to point out that there is not one unified Lichess rating so the comparisons between Lichess and FIDE don't necessarily match up for all time controls.

For example the top 10 bullet players here are all well above 3000 (the highest is over 3200) while only 1 blitz player is over 3000 and the top rapid player isn't even 2900. Oh and the top classical player? Try mid 2600s.

This can have a lot to do with how many players play (rapid and classical is much less popular for example and may attract less casual players) and other factors.

The graphs that @petri999 posted actually shows this relatively well.

So I guess statements like "a 2400 lichess rating is 2000 FIDE" is just oversimplified and only correct for one time control.

Basically, if you're talking about bullet or blitz an estimate of inflation of 300-450 points is justified depending on the rating, but if you have a Rapid or classical rating it is definitely NOT inflated by close to 400 points. Probably much less.

Is it really necessary to talk about OTB bullet? At least they aren't carpenters trying to hammer in nails that quickly.

https://youtu.be/St1Udrk6VG0

#26 @petri999 I am glad to hear that those long game still exist somewhere in the game of title chess. (kidding amicably).

And I am sure that the rigour of competition is strong there. But these are not individual games but series of games.

For the population of games pairings is actually the number of series of games pairings (if I am not wrong on the many games for pyramid climbing). The games within series are not comparable to off-tournament lichess games just in that regard.

I am not saying that lichess games are less worthy, i might actually be saying that non tournament games provide for better statistics, unless the tournament games within series linkage were (or is) acknowledged, documented and compensated for.

So, while the winners of the tournament may be clear cut, and their rank ordering as well for that tournament, the statistical power of one tournament with weakly random pairing (actually is it not very tiered?) and the number of players, would make for a questionable (just asking) base for a precise or robust rating (over time).

The frequency or number of tournaments held by the same person over a year might make for a better estimate for that person, perhaps, but I wonder about the pairing random coverage. However, I have no clue about the numbers i just assumed to be on the small side, I just never saw any data at the scale of that which lichess does display. Other servers may not be as transparent, i don't know. And I don't think FIDE maintains a database of public access and internal consistent (?, ignoramus there).

But i would be curious about well documented databases where such question could be tested. going back in history too.

However I understand that databases have often been made proprietary by the need for human inspection or curation. or something like that. so i might have to wait and keep hoping... (not that i would do the work, but at least I could consider it, or dream about it).

Sometimes, I don't bother to put a question mark to my questions. most of the above is questions or hypothesis.. even when not presented clearly as such. I am missing data, to make those questions high degree of belief or conviction. but in absence of such, i think they are plausible questions or hypotheses.

@hendrixmaine some of your asumptioins are very incorrect. Classiical offset between lichess and FIDE at 2000 is very big. More so than blitz there just is not any reliability on classical ratings. Lichesss 2000 maps between fide 1400 to fide 2400. Strong players just dont play classical online. I think thos 2400 are strong player who played few games and then notice that opposition is soft.

Highest correlation in rating was in blitz. If you the reddit post is says that lichess blitz to FIDE had correlation rating was 0.6 which means that lichess rating "explains" about 36% of FIDE rating. And classical just about non existent. Classical 2000 can be good player or someone who drops his/her queen for absolutely no reason