how accurate are LiChess ratings?

This seems to be one of those immortal questions, its been answered in full by lichess staff members several times now you might have some luck finding a suitable answer by running a forum search.

One of the main differences to note is that lichess uses "glicko" in place of the more familiar elo system.

I also notice in some cases it differs depending on the players rank. For example some titled players have the same rating on lichess as chess.com. it seems to even out some after so many thousand games (30,000/40,000 or so). It takes some looking into to fully understand it again I would suggest some forum searches.

One can find approximations everywhere which predict the numbers with a certain probability for most users.

docs.google.com/spreadsheets/d/1GcVG_slcTUJfi0sQMeaxes_jwYFBrfWaR1ejSSOH-SM/htmlview

@Sarg0n :

sorry but I dont think this is right. For example: do you have FIDE-Elo of around 2000 as CM titled player? I think you should be realistically at around 2200 FIDE. Your lichess-rating is 2200.

IM Chessexplained (very strong blitz-player) has 2500 lichess and 2450 FIDE. Many other cases the same. I dont think that a FIDE 1750 player is able to keep 1990 lichess-Rating. Maybe some peak or beginners luck but being better than around 93% of all lichess players with just 1750 FIDE is...not very realistic I have to say.

1750 FIDE is around median. Maybe these datas are very old? Just my thoughts.

I have different accounts. Peak blitz is 2302, so that's not a bad comparison, I think. I am rather slow.

We all have fluctuations of some hundred points a day, so there's no exact formula for everything.

PS: table data of 2016, there are more recent ones available with the current population of lichess.

@Sarg0n :

I have just recognized that these datas are obiously based on a handful of examples. So they are not representative anyway.
The one I have once found is more similiar to my impression that FIDE and lichess ratings are not that far away.

Of course there are always crazy blitz-addicts (I know some younger players with average skills in normal time controls but extremely fast and strong in blitz/bullet with totally different ratings - 1800 long and 2200 bullet an so on...) but thats not the regular thing. And as I have just said: Often FIDE-Ratings are old or useless when players rarely play tournaments.

Greetings!

Those are by no means correlated in any way. In the current state of affairs GMs are 2700+, IMs are 2550+, FMs are 2450+ etc

This however can't be predictor of any kind because OTB classical and online blitz are two different sports for all intents and purposes.

I'm almost certain that any average GM could reach 2700 blitz here, but the opposite doesn't hold, 2700 blitz here isn't a proof of GM level.

What you should do is play 9 rated OTB games and see where do you stand on the FIDE rating list.

Well @PetarBosnjak they are correlated. Obviously as say sta average GM is 2700 i.e. you have some sort of mapping from Elo to Blitz. Rating just inst overly strong for many reasons BUT rating has some realtong FID/USCF rating. But accuracy of such estimation is about +-200 pts so I woudl say pretty meaningless. In link above there some data but raw data is pic
here
imgur.com/a/nWy4x
From there it obvious that there is correlation i.e. there noisy line that is clearly linear. But also evident is that amount noise is pretty huge. This is before classical/rapid split

You cannot disprove such average connections by single outliers. Of course there’s a correlation which predicts certain intervals well.

All the tables on the internet I found are yet pretty good for the majority of cases.

Ye s there is correlation and it is actually better on stronger side. But if you look about average Lichess player about 1500 then Then the scatterogram berely has darker part.. And this also expected. At least in Finland vast majority of tournament are not FIDE rated but only nationally rated. Hence the FIDE rating of anyone below 2000 is not nearly as accurate as the national rating. So there we have noise both on mapping and noise on target value ans result cannot be very accurate

At 2000 in very no scienfistic way if pick the almost black are up and down I get bracket of 1750 to 2100 in FIDE. So not quit 400 points I said but about 350. So you do get an estimate but still kinda noisy. Probably of same accuracy if attend couple of OTB tournaments only.

Also further prediction error for someone pondering how he would do in the OTB is the additional skill of actually thinking of your moves after playing years of blitz only. At least I find it very hard. And my ELO/SELO/Lichess Blitz are pretty close each other

If you really want to concoct a function which translates Lichess blitz into FIDE ELO you should take more parameters into than simple value of the corresponding ratings.

One of the things I'd definitely pay attention to is whether player wins/loses many games in worse/better positions. How much does he rely on time, is his play consistent in terms on cpl or there are spikes all over the graph indicating that he's tricky, but not nearly as precise as rating would suggest.

IMHO sheer rating doesn't indicate anything unless it's complemented by other parameters which may actually tell something about player's skills necessary for OTB classical.