Data analysis: Lichess vs. FIDE ratings • page 1/2 • General Chess Discussion • lichess.org

Problem: There are lots of threads with speculation about the relationship between Lichess ratings and FIDE ratings, but almost no systematic analysis of the question.

Objective: Measure how “inflated” internet ratings are relative to over the board ratings.

Method: I looked at the profiles of 11,235 Lichess players and identified the subset of players who self-reported a FIDE rating in their personal profiles. Then, I compared their self-reported FIDE ratings to their current Lichess ratings.

# Full sample

11,235 individuals
Median Blitz rating: 1588
Median Classical rating: 1639
Median Puzzle rating: 1578

# Subset of players with self-reported FIDE ratings

230 individuals
Median Blitz rating: 1856
Median Classical rating: 1902
Median Puzzle rating: 1860

# Graph

I plotted the relationship between FIDE and Lichess ratings across the three categories. Check out the graph here:

http://imgur.com/a/vUMG8

# Median difference between FIDE and Lichess ratings

Blitz: 127 points
Classical: 196
Puzzle: 81

Based on this small highly imperfect sample, it looks like Lichess Classical ratings overshoot FIDE ratings by about 200 points.

# More

If you’re interested in the relationship between Tactics Puzzles and In-Game ratings, check out my previous data analysis here: https://lichess.org/forum/general-chess-discussion/tactics-vs-game-ratings-a-statistical-analysis-5000-lichess-players

Problem: There are lots of threads with speculation about the relationship between Lichess ratings and FIDE ratings, but almost no systematic analysis of the question. Objective: Measure how “inflated” internet ratings are relative to over the board ratings. Method: I looked at the profiles of 11,235 Lichess players and identified the subset of players who self-reported a FIDE rating in their personal profiles. Then, I compared their self-reported FIDE ratings to their current Lichess ratings. # Full sample 11,235 individuals Median Blitz rating: 1588 Median Classical rating: 1639 Median Puzzle rating: 1578 # Subset of players with self-reported FIDE ratings 230 individuals Median Blitz rating: 1856 Median Classical rating: 1902 Median Puzzle rating: 1860 # Graph I plotted the relationship between FIDE and Lichess ratings across the three categories. Check out the graph here: http://imgur.com/a/vUMG8 # Median difference between FIDE and Lichess ratings Blitz: 127 points Classical: 196 Puzzle: 81 Based on this small highly imperfect sample, it looks like Lichess Classical ratings overshoot FIDE ratings by about 200 points. # More If you’re interested in the relationship between Tactics Puzzles and In-Game ratings, check out my previous data analysis here: https://lichess.org/forum/general-chess-discussion/tactics-vs-game-ratings-a-statistical-analysis-5000-lichess-players

ParadoksB

#2

Quite interesting, thanks. But people who have reported FIDE ratings around 3000 should probably be removed from the analysis :)

dudeski_robinson

#3

Thanks. And sure. But that's why I used the median and not the mean, so that outliers don't skew the analysis.

legend

#4

good stuff. bumping this for myself since im in process of similar analysis though im taking it easy and going slowly.

mathtuition88

#5

+1 for your good effort. Really appreciate it.

thanks!

Just curious, is the 200 rating inflation consistent throughout the rating band?

Or the higher one's rating, the less the inflation? I recall I see one CM (presumably elo >=2200) with Lichess rating 2200.

E.g. For high rated players, Lichess 2400=Elo 2200?
Mid rated players, Lichess 1800=Elo 1600?
Lower rater players, Lichess 1400=Elo 1200?

Just curious, is the 200 rating inflation consistent throughout the rating band? Or the higher one's rating, the less the inflation? I recall I see one CM (presumably elo >=2200) with Lichess rating 2200. E.g. For high rated players, Lichess 2400=Elo 2200? Mid rated players, Lichess 1800=Elo 1600? Lower rater players, Lichess 1400=Elo 1200?

BigGreenShrek

#8

I second that question! My impression is that the ratings are less deviant for higher rated players, but I'd love to see the actual stats.

dudeski_robinson edited

#9

Very good question!

Unfortunately, my sense is that the sample size is way too limited to really answer a nuanced question like this properly.

That said, your expectation is basically that the relationship "flattens out" at the top, where lichess ratings would be more indicative of the truth.

But if you look at the scatter plot above (imgur link), the relationship seems pretty linear (i.e., constant across bands). But again, very limited data...

Very good question! Unfortunately, my sense is that the sample size is way too limited to really answer a nuanced question like this properly. That said, your expectation is basically that the relationship "flattens out" at the top, where lichess ratings would be more indicative of the truth. But if you look at the scatter plot above (imgur link), the relationship seems pretty linear (i.e., constant across bands). But again, very limited data...

petri999

#10

I think one thing you can say that correlation is pretty weak. Its more like shotgun scatter than real mapping