English Chess Forum

Robert Jurjevic wrote: "Elo suggested scaling ratings so that a difference of 200 rating points in chess would mean that the stronger player has an expected score (which basically is an expected average score) of approximately 0.75."

Approximately is the key word. The Elo scale is approximately 75%, the ECF scale is exact. The difference is enough to drive the practical conversion factor below 8. Look at the table. A rating difference of 198 to 206 is an expected score of 76% with 75% corresponding to a rating difference of 193, or reversed it's a difference of 189 to 197 that has the expected score of 75%.

P.S.

'FIDE2 - FIDE1 = 8 * (ECF2 - ECF1)' holds for 'FIDE = 8 * ECF + 650' but not for 'FIDE = 7.5 * ECF + 700'. For 'FIDE = 7.5 * ECF + 700', 'FIDE2 - FIDE1 = 7.5 * (ECF2 - ECF1), which again, if by definition it must hold that 'FIDE2 - FIDE1 = 8 * (ECF2 - ECF1), may indicate that 'ECF2 - ECF1' is larger than it should have been, i.e., that the ECF grades are stretched, as '8 * (ECF2 - ECF1)_correct ~ 7.5 * (ECF2 - ECF1)_streched', and you even have an estimate of the current stretching (by fitting the ECF grades to FIDE ratings).

Roger de Coverly wrote:

Robert Jurjevic wrote: "Elo suggested scaling ratings so that a difference of 200 rating points in chess would mean that the stronger player has an expected score (which basically is an expected average score) of approximately 0.75."

Approximately is the key word. The Elo scale is approximately 75%, the ECF scale is exact. The difference is enough to drive the practical conversion factor below 8. Look at the table. A rating difference of 198 to 206 is an expected score of 76% with 75% corresponding to a rating difference of 193, or reversed it's a difference of 189 to 197 that has the expected score of 75%.

I think that the expected score is, by definition, exactly 75%, and that is why it is called expected. The player, of course, may not, and usually, does not, perform as expected. Ratings (grades) are based on finite number of games, and that is why they are not 100% accurate. Even if the grade estimate could have been calculated based on infinite number of games, the definition would keep the attribute 'expected' for the 75%, as no matter how accurate is one's grade, one cannot predict future 100% accurately, as people may not, and usually do not, perform as expected, even the estimate of their grades may have been done 100% accurately. Therefore, the expected score should, by definition, be exactly 75%, no more, no less, and shoud be independent of the errors in grade caclulation (owing to the fact that one cannot take into account infinite number of games).

Matt Fletcher wrote: The best fit line across all these data points is FIDE = 7.4 x ECF + 690 - alternatively (with rounder numbers) 7.5 x ECF + 675.

This has cropped up in another thread. Would an update be possible? I ask that with no idea of how much work is involved.

NickFaulks wrote:

Matt Fletcher wrote: The best fit line across all these data points is FIDE = 7.4 x ECF + 690 - alternatively (with rounder numbers) 7.5 x ECF + 675.

This has cropped up in another thread. Would an update be possible? I ask that with no idea of how much work is involved.

I've downloaded the FIDE February 2017 list and compared to ECF January 2017. I can't remember exactly what I did last time so just matched up all people with ENG, SCO, WLS nation codes and a grading code of A-E for whom I could find a matching FIDE rating - there were 2,191 players in total where this was the case.

I'm now getting a best fit line of FIDE = 7.6 x ECF + 625 - working with 7.5 that gives me 7.5 x ECF + 640 which is somewhat lower than I got before, and quite a bit less than the '+700' rule of thumb.

Happy to do a bit more analysis now I've got the dataset together if anyone's interested.

Matt Fletcher wrote: Happy to do a bit more analysis now I've got the dataset together if anyone's interested.

The "new" URS ratings still have limited credibility, but how does the ECF to URS conversion look? Apart from some seemingly spurious data every player with an active FIDE rating has a URS rating and vice-versa.

http://universalrating.com/

How can I have a different URS rating vs FIDE rating when I don't play any blitz or rapidplay chess?

Apologies if I've missed something earlier in the thread.

Mike Truran wrote:How can I have a different URS rating vs FIDE rating when I don't play any blitz or rapidplay chess?
Apologies if I've missed something earlier in the thread.

Maybe if you played some Blitz and/or Rapid it would pull your URS back in line.

Mike Truran wrote:How can I have a different URS rating vs FIDE rating when I don't play any blitz or rapidplay chess?

URS follows different and unknown rules from FIDE. There's a default Rapid and Blitz adjustment which applies even when there are no Rapid or Blitz games recorded.

Apart from unexplained anomalies, it appears the calculators of URS ratings have access to the FIDE data.

Mike Truran wrote:How can I have a different URS rating vs FIDE rating when I don't play any blitz or rapidplay chess?

As Roger has already mentioned, the two systems work on different methodologies.

The following quotation is taken from the URS press release about the February list:

"Commentators with a key eye will note that GM Vladimir Kramnik has lost 3 rating points on the February list despite not playing any rated games during the month of February. This is one of the key differences between the URS and the ELO rating systems. URS continuously re-rates all players in the database regardless of individual activity during the month and these movements are hence normal for the system. A more detailed explanation of some of the notable changes between the January and February rating lists will be released in due course."

David Sedgwick (URS press release) wrote:URS continuously re-rates all players in the database regardless of individual activity during the month and these movements are hence normal for the system.

Would I be right in thinking that if you took the December 2016 data, added zero new games, told the system that it was now 31st January 2017, that the ratings would change? That's something Elo based systems don't do, but the ECF system would (on a six monthly look).

Is it a desirable feature of rating systems that purely the passage of time should change the rankings produced?

Roger de Coverly wrote:

David Sedgwick (URS press release) wrote:URS continuously re-rates all players in the database regardless of individual activity during the month and these movements are hence normal for the system.

Would I be right in thinking that if you took the December 2016 data, added zero new games, told the system that it was now 31st January 2017, that the ratings would change?

That's a question that can't be answered with certainty without knowing the algorithm being used to calculate the ratings, but the following statement on their website suggests the answer is yes:

http://universalrating.com/faqs.php wrote:URS™ is a weighted performance rating, calculated across several years of previous game results for all players. Older games are given less importance than more recent games, by applying an exponential decay rate.

I'd assume that some, maybe all, games would be affected by the decay factor, and the oldest month's worth of data would drop out of the calculation.

Matt Fletcher wrote:

NickFaulks wrote:

Matt Fletcher wrote: The best fit line across all these data points is FIDE = 7.4 x ECF + 690 - alternatively (with rounder numbers) 7.5 x ECF + 675.

This has cropped up in another thread. Would an update be possible? I ask that with no idea of how much work is involved.

I've downloaded the FIDE February 2017 list and compared to ECF January 2017. I can't remember exactly what I did last time so just matched up all people with ENG, SCO, WLS nation codes and a grading code of A-E for whom I could find a matching FIDE rating - there were 2,191 players in total where this was the case.

I'm now getting a best fit line of FIDE = 7.6 x ECF + 625 - working with 7.5 that gives me 7.5 x ECF + 640 which is somewhat lower than I got before, and quite a bit less than the '+700' rule of thumb.

Happy to do a bit more analysis now I've got the dataset together if anyone's interested.

I accept that it might not make much difference, but wouldn't it be better to compare the FIDE January 2017 list with the ECF January 2017 list?

Alex Holowczak wrote:wouldn't it be better to compare the FIDE January 2017 list with the ECF January 2017 list

Hastings would be in the February FIDE list. If it is in the ECF January list then FIDE Feb is the correct one otherwise FIDE Jan is the more appropriate when trying to compare like with like.

Roger de Coverly wrote: Is it a desirable feature of rating systems that purely the passage of time should change the rankings produced?

As always, it depends upon what your ratings are intended to do. The FIDE system is explicitly an attempt to optimise predictive power. On that basis, some such adjustment does make sense, but only on a very long time scale. Noticeable changes from one month to the next make no sense.

One problem with the URS is that we have not been told ( unless I've missed it ) what their underlying philosophy is.