There is absolutely no correlation between an internet blitz rating and a OTB Fide rating that can be made by any "formula". Nice try by the OP but far to many assumptions are made in the premise, making the hypothesis completely invalid. A persons internet blitz rating vs. standard (90 min. OTB) could differ by 400 points or be very similar. Everybody has different strengths/weaknesses.
Well, establishing an empiric relation which suits for a lot of people is not bad at all.
It basically says something like: reducing the over-rated lichess rating to 85-90% and adding 150-200 points. This works for most, there are outliers, sure.
More than outliers. A very high percentage of players OTB rating can be 200/400 points higher than their blitz rating. Or vice versa, many players excel at fast time controls. Their blitz rating is 200/400 points above their standard OTB rating. Internet chess and OTB are two entirely different animals until one has spent a lot of time at each and becomes familiar with the settings. The OP is suggesting to new internet blitz players, his "formula" gives an accurate portrayal of what to expect OTB. He makes too many assumptions for it to be a valid hypothesis.
SebVC: Producing a formula with bounds is actually more complicated than it sounds.
DrHack: I would argue that this is not a flaw at all. What the formula assumes, is that blitz and classical ratings are two (imperfect) proxies for FIDE rating. There is absolutely no assumption that these two variables are the same; the model absolutely does not require them to be correlated. Even so, I can tell you that Blitz and Classical ratings are correlated at 0.88 (in the sample of players who played at least 50 games of each — 95,000 people). This is a super strong correlation.
mdinnerspace: Nobody’s arguing that the formula makes perfect predictions. So I take your point there. However, when you say that online and OTB ratings are “two completely different animals”, you are overstating the case. On average, a 2000 FIDE player will tend to play better online than a 1600 FIDE player. Not *every* 2000 OTB player will beat the 1600 OTB player online, but ON AVERAGE this will be the case. My model does not assume that there is a one-to-one mapping of online vs. OTB. This is a descriptive linear model, and the only potentially bad assumption is that people report their true FIDE ratings in their Lichess profile. As I’ve stated above, this is not always the case — there are many ratings that are obviously made up. I have taken many different approaches to deal with these “outliers” or “bad” data points.
My FIDE rating estimator formula:
(What people would guess your rating to be + what you think your rating should be) / 3
Why not just calculate the squared deviation to your model, it would give a basic estimation of how volatile is the ratings are
In defence of dudeski's formula, it is the best and most comprehensive formula that I have seen regarding Lichess and FIDE ratings. (Others I have seen are: https://www.reddit.com/r/chess/comments/5gac0h/chess_rating_comparison_2016_results_chesscom/)
Clearly, there should be some correlation between FIDE and Lichess, though not perfect as many have pointed out.
dudeski's formula attempts to illustrate that relationship, and I really appreciate it. Good job.
Your formula gives me an estimated 1874 FIDE rating. This might be a close approximation as I know my USCF rating of 1592 is inaccurate due to relative inactivity in OTB tournaments in recent years. Thanks for taking your mathematical understanding and statistics and creating this!
I Fail to see how an internet blitz rating can correlate to a OTB standard FIDE rating. The reason for this is simple: A persons blitz and standard chess skills can vary greatly. If by chance, the players skills are similar, than the formula has some validity.
For many players a rating difference, whether higher or lower, exists between their blitz (5 minute) and standard (90 minute OTB) skills. Hence a valid comparison can not be made.
Especially taking into account players who 1st start playing chess do so by internet play at the faster time controls. When they 1st experience OTB 90 minute games, everybody reacts in their own particular fashion. Some play well, others do not. It takes dozens of games, at a minimum to begin establishing a OTB rating, which may or may not be any wheres close to their on-line blitz rating.
For as many players that the formula comes within 100 points, there will be as many players who after establishing a OTB rating will find the rating is either higher or lower by 200 points as compared to on-line blitz play. View the difference of ratings of slow vs. fast play for the same player on-line. Very often 200 point spreads are seen.
Want to "estimate" your FIDE rating? FIDE will do it for you after playing x amount of games. Anything else is purely speculation, such as on-line ratings are inflated by x, y and z. The OP downloaded 300,000 Lichess accounts? Of which 1 % voluntarily listed their FIDE rating. The resulting formula is based on this control group to give an estimated FIDE rating ???
On-line ratings are not a reliable indicator of OTB ratings within+/- 200 points and greater for a great many of players. Where is this information available to readily download all the users accounts at Lichess?
Well, you don‘t see it and it doesn‘t have to be necessarily so but for most people IT IS quite sound. So no bad estimation for most users. Not more, not less.
This topic has been archived and can no longer be replied to.