How to estimate your FIDE rating (conversion formula inside) • page 9/36 • General Chess Discussion • lichess.org

Since there are many (fake) outliers, summarizing these data using the average would be misleading. We use the median instead:

A typical (median) user's FIDE rating tends to be 78 points lower than her Lichess Blitz rating
A typical (median) user's FIDE rating tends to be 169 points lower than her Lichess Classical rating

And proceeds to create his formula:

FIDE Rating = 187 + Lichess Classical Rating X 0.38 + Lichess Blitz Rating X 0.48

We can then conclude that a persons classical rating of 1169 and a blitz rating of 1078 should equal FIDE 1000

Lets plug in the numbers.
1169 x .38 = 444
1078 x .48 = 517
444 + 517 + 187 = 1148

hmm.. is this the new math you instruct 1000 = 1148 ???

Please explain when using your formula, how a 2078 blitz rating and a 2169 classical rating results in a rating of 2008 (should equal 2000) while a blitz rating of 1078 and a classical of 1169 results in a rating of 1148 (should equal 1000) ?

@dudeski_robinson writes in 1st post: Since there are many (fake) outliers, summarizing these data using the average would be misleading. We use the median instead: * A typical (median) user's FIDE rating tends to be 78 points lower than her Lichess Blitz rating * A typical (median) user's FIDE rating tends to be 169 points lower than her Lichess Classical rating And proceeds to create his formula: FIDE Rating = 187 + Lichess Classical Rating X 0.38 + Lichess Blitz Rating X 0.48 We can then conclude that a persons classical rating of 1169 and a blitz rating of 1078 should equal FIDE 1000 Lets plug in the numbers. 1169 x .38 = 444 1078 x .48 = 517 444 + 517 + 187 = 1148 hmm.. is this the new math you instruct 1000 = 1148 ??? Please explain when using your formula, how a 2078 blitz rating and a 2169 classical rating results in a rating of 2008 (should equal 2000) while a blitz rating of 1078 and a classical of 1169 results in a rating of 1148 (should equal 1000) ?

dudeski_robinson

#82

There are two steps in the analysis. First, I offer simple descriptive statistics (medians). Second, I develop a prediction formula using ordinary least squares regression. This is all explained very clearly in the original.

dudeski_robinson edited

#83

Did you even make an effort to read and understand my detailed post comparing the four formulas? I spent a lot of time explaining exactly why the median comparison doesn't give the same results as my formula.

The fact that the two approaches dont produce the same results is TOTALLY NORMAL AND 100% EXPECTED.

I also explained in great detail why it's better to use the prediction formula based on least squares regression than the simple differences relative to the medians like you propose. In short: my formula is much more precise than the strategy you suggest.

There's no "new math" here. There are just different prediction formulas. One of those formulas is better than the others, because it minimizes prediction error in my sample of nearly 3000 player profile. In my original post, I use that formula.

Did you even make an effort to read and understand my detailed post comparing the four formulas? I spent a lot of time explaining exactly why the median comparison doesn't give the same results as my formula. The fact that the two approaches dont produce the same results is TOTALLY NORMAL AND 100% EXPECTED. I also explained in great detail why it's better to use the prediction formula based on least squares regression than the simple differences relative to the medians like you propose. In short: my formula is much more precise than the strategy you suggest. There's no "new math" here. There are just different prediction formulas. One of those formulas is better than the others, because it minimizes prediction error in my sample of nearly 3000 player profile. In my original post, I use that formula.

CM Sarg0n

edited

#84

Even a linear regression doesn‘t require that all points lie on the curve. A curve is determined that way that the sum of deviations for each point is minimized. Picking out single points and stressing the „distance“ s no good - you have consider all of them.

This is scientific, well done!

Even a linear regression doesn‘t require that all points lie on the curve. A curve is determined that way that the sum of deviations for each point is minimized. Picking out single points and stressing the „distance“ s no good - you have consider all of them. This is scientific, well done!

mdinnerspace

#85

The OP is speaking in "medians". Generalizations. Assumptions such as classical ratings are higher than blitz ratings to arrive at a formula. For a great many players their blitz rating is quite a bit higher. Their skill set is better at faster games relative to slower games. For a great deal of other players, their skills at fast time controls are quite a bit less relative to longer time controls, especially those required for 90 minute OTB.

It simply is foolish to suggest that any formula can remotely "predict" a players FIDE OTB rating based on an internet blitz rating. Especially from Lichess ratings as blitz and classical time controls are very close. 8 + 0 is classical with 5 + 3 is blitz. Identical.

The OP is speaking in "medians". Generalizations. Assumptions such as classical ratings are higher than blitz ratings to arrive at a formula. For a great many players their blitz rating is quite a bit higher. Their skill set is better at faster games relative to slower games. For a great deal of other players, their skills at fast time controls are quite a bit less relative to longer time controls, especially those required for 90 minute OTB. It simply is foolish to suggest that any formula can remotely "predict" a players FIDE OTB rating based on an internet blitz rating. Especially from Lichess ratings as blitz and classical time controls are very close. 8 + 0 is classical with 5 + 3 is blitz. Identical.

CM Sarg0n

#86

Well, it is developed scientifically. It works pretty well for the majority with a high probability. Case closed.

We should recalculate Maurice Ashley‘s Elo! Anyone knows his Blitz rating?

Well, it is developed scientifically. It works pretty well for the majority with a high probability. Case closed. We should recalculate Maurice Ashley‘s Elo! Anyone knows his Blitz rating?

mdinnerspace edited

#87

@dudeski_robinson

The issue with your formula is not the prediction it makes, but that it is not consistent.

A steady progression is seen when doing the calculations from 2000 down to 1000. The resulting prediction continues to produce a higher and higher number as lower ratings are calculated. The formula "over estimates" the FIDE rating for the 1000 by 148 points while is only 8 points different for the 2000.

There is something missing in the formula. The expected result differs greatly from low ratings and high ratings.

@dudeski_robinson The issue with your formula is not the prediction it makes, but that it is not consistent. A steady progression is seen when doing the calculations from 2000 down to 1000. The resulting prediction continues to produce a higher and higher number as lower ratings are calculated. The formula "over estimates" the FIDE rating for the 1000 by 148 points while is only 8 points different for the 2000. There is something missing in the formula. The expected result differs greatly from low ratings and high ratings.

dudeski_robinson edited

#88

Nobody is saying that the same exact skills are involved in playing a 10+0 game online and an 4hr game OTB. All we are saying, is that the two sets of skills are related.

The graphs I posted in the original post showed clearly that -- except for some outliers, which I exclude from the analysis -- there's a VERY strong relationship between people's Lichess ratings, and what they report as their FIDE rating.

https://imgur.com/a/nWy4x

All I've done is calculate the formula that best approximates that relationship.

And by the way, medians don't have anything to do with the formula at the top of my original post. They don't enter the calculations AT ALL. I gave the medians only to give people a sense of what the data look like.

My theory: You don't engage with my actual arguments in post #80, because you are unable to understand them.

Nobody is saying that the same exact skills are involved in playing a 10+0 game online and an 4hr game OTB. All we are saying, is that the two sets of skills are related. The graphs I posted in the original post showed clearly that -- except for some outliers, which I exclude from the analysis -- there's a VERY strong relationship between people's Lichess ratings, and what they report as their FIDE rating. https://imgur.com/a/nWy4x All I've done is calculate the formula that best approximates that relationship. ------- And by the way, medians don't have anything to do with the formula at the top of my original post. They don't enter the calculations AT ALL. I gave the medians only to give people a sense of what the data look like. My theory: You don't engage with my actual arguments in post #80, because you are unable to understand them.

mdinnerspace edited

#89

Your "premise" that the "two sets of skills are related"...

Is false. For some their FIDE rating and online rating is similar, but that can only be determined after ratings are established. For a great many others , possibly even a majority, these 2 ratings will vary by 200 points, either higher or lower.

You then proceed to "find the math" to back up your premise. This can easily be done, in fact is very common in the scientific field. Mathematics is a great tool, is it not ? It can be used to prove anything! But your math, the formula itself, is missing a factor, as the prediction progressively gives a higher expected number as lower ratings are calculated.

Your "premise" that the "two sets of skills are related"... Is false. For some their FIDE rating and online rating is similar, but that can only be determined after ratings are established. For a great many others , possibly even a majority, these 2 ratings will vary by 200 points, either higher or lower. You then proceed to "find the math" to back up your premise. This can easily be done, in fact is very common in the scientific field. Mathematics is a great tool, is it not ? It can be used to prove anything! But your math, the formula itself, is missing a factor, as the prediction progressively gives a higher expected number as lower ratings are calculated.

dudeski_robinson edited

#90

The premise is NOT false. Look at the graph!

https://imgur.com/a/nWy4x

Again, I'm not saying that the two sets of skills are PERFECTLY aligned. Some will be better online, and some will be better OTB. NOBODY is arguing that the relationship is perfect.

I'm trying very hard not to let the conversation fall into ad hominems here, but your refusal to understand simple concepts and arguments is truly spectacular.

The premise is NOT false. Look at the graph! https://imgur.com/a/nWy4x Again, I'm not saying that the two sets of skills are PERFECTLY aligned. Some will be better online, and some will be better OTB. NOBODY is arguing that the relationship is perfect. I'm trying very hard not to let the conversation fall into ad hominems here, but your refusal to understand simple concepts and arguments is truly spectacular.