How to estimate your FIDE rating (conversion formula inside)

Now that lichess has a realistic rapid and classical designation, the OP's formula might have a smidgen more validity, as it would include games 25+ minutes played as "classical"; which is slightly more comparable to longer OTB games.
However, "classical" previously was any game 8 minutes or longer. 10 minute games at lichess are 2nd most popular and accounted for 70%+ of all previous "classical" games. 10 minutes and less is a blitz game by comparison to actual classical. This is the primary drawback of his hypothesis lichess ratings could accurately predict an OTB Fide rating.
There simply exists no comparison between online blitz and 90 minute OTB ratings. (except of course for the statistical averages and medians that apply to a very small percentage of players who to fall within a narrow range.)

The original chart, alone, is enough for us to take lots of conclusions. I added a trend-line in red :

imgur.com/a/4nda9

PS: Each grid line is spaced 250 points.
PS: The blue line crosses the chart symmetrically (were X=Y).

My personal conclusion is that, ignoring the false FIDE rating reports floating up away from the trend-line, Lichess ratings are about 100 to 150 points above FIDE ratings, on average. Though for beginners below 1500 lichess blitz this difference becomes larger. Probably because they aren't confident enough to play concentrated OTB a long game.

No surprise that the red line is curved. It's because we are comparing fast Lichess time controls against slow Fide time controls, while faster time controls always result in a larger rating span. That's why Magnus is almost reaching 3000 in blitz Fide.

So, for people like me, who play better at longer time controls, the FIDE rating may be even higher than Lichess rating. It depends on the individual.

I doubt that the data is correct. There are simply no 3000 fide ratings in reality. There are few GM's playing on lichess atm, so the Fide ratings numbers should be even lower.

Also the point clouds seem to vary too much. How did they collect the data?

I suspected a higher gap between lichess and fide rating.

@EvilChess That seems very sensible to me. But the curvature is *really* slight. Unless you're at the extremes, it won't matter. So for the typical forum user, the linear fit is probably "good enough".

@TkzonKy Obviously, there are fake data points in there. Read the original post, where I described in detail how I collected the data and how I attempted to minimize the influence of fake reports. Of course, it's still not perfect...

@TkzonKy, there is no validation of the reported Fide rating, but that's the best data available. I didn't use any math, just plain old paint brush, visually ignoring the high ratings ( including all 3000 false reports ).

Now let's check for your estimated FIDE rating, using your Classical 1750 rating, which is the best category to compare against FIDE. In the drawing below, my guess would be about 1550 +/- 200 with a confidence level around 80%.

Yes it varies a lot, because:
1) Online ratings vary a lot on their own. Charts show it.
2) The OTB performance with long time control varies depending on the individual. It's a very different category.

For instance, in past when I played OTB tournaments, my club rating was slightly higher than my internet ICC rating. Because I do better at longer time controls and I need no distractions.

@dudeski_robinson, thanks for your work. I hope you may soon have nice data to collect from new Classical and Rapid categories and plot something interesting.

I understand that adding the blitz rating could provide greater precision. And I see that blitz data displays a smaller gap. However, it doesn't seem correct that blitz ratings ended up with a greater pounding factor in your formula (0.48). Because the longer time control (Classical) should have a greater relevance when estimating FIDE ratings, since FIDE ratings are measured at very long games. Indeed, LichessClassical-to-Fide relationship alone is stronger than LichessBlitz-to-Fide relationship alone.

In case online cheaters were interfering too much in Lichess Classical ratings, then we should see less of this effect in new Rapid category.

@EvilChess

An alternative explanation, is that people with Fide ratings are more likely to play blitz online.

In any case, the two ratings are super correlated anyway, so I wouldn't draw much inference from the fact that one has a greater weight than the other (the difference in coefficient sizes is pretty small anyway, relative to standard error).

To be honest, I'm probably done with this. I got the info I wanted personally, and am unlikely to push it further.

I am happy to share the python code to scrape the data if anyone wants to tweak it and run it again.

This "assumption" that classical time controls are more prone to "cheating" (because of the longer time involved), bothers me.
Computer assisted play is employed at every time control, even MORE so at fast time controls. More cheaters are caught at bullet (simply because of it's popularity) than rapid or classical. People are simply uneducated regarding the subject, make false statements, not possessing all the facts.
IF someone is to make the choice to cheat, the time control played is irrelevant.

> " For instance, in past when I played OTB tournaments, my club rating was slightly higher than my internet ICC rating. Because I do better at longer time controls and I need no distractions." EvilChess

Guess what? the same applies to a very high percentage of chess players; just as the opposite applies; a great many do better at fast time controls. Statistics are an average; are not relevant to the majority. The hypothesis that an accurate prediction (within a few points...OP) can be made is false.

The prediction made matches the statistical average of all players. That's it.

Thanks for the explanations EvilChess and dudeski_robinson.