ECF grades compared with FIDE ratings

General discussions about ratings.
Post Reply
Sean Hewitt
Posts: 2193
Joined: Sat Mar 10, 2012 8:18 pm
Contact:

Re: ECF grades compared with FIDE ratings

Post by Sean Hewitt » Sun Feb 02, 2014 9:34 pm

Jonathan Bryant wrote:...if it was just a case of finding the right formula then we'd all be in the same order on both lists wouldn't we?
Nope, not necessarily. Take any two subsets of games played by a player and it is possible to get vastly different results. It's just probabilities.

Jonathan Bryant
Posts: 3452
Joined: Sun May 11, 2008 3:54 pm

Re: ECF grades compared with FIDE ratings

Post by Jonathan Bryant » Sun Feb 02, 2014 9:48 pm

Sean Hewitt wrote:
Jonathan Bryant wrote:...if it was just a case of finding the right formula then we'd all be in the same order on both lists wouldn't we?
Nope, not necessarily. Take any two subsets of games played by a player and it is possible to get vastly different results. It's just probabilities.
OK, but my problem with the rating system is to do with what happens when the same group of games is rated. Unlike Chris, it's not that I play all these games that are ECF rated and all these other games that are FIDE rated. There are some I play that are only ECF rated and some that are only FIDE rated, but mostly the games I play are both FIDE rated and Elo rated - but the results that come out are very different.

I'm not saying everybody's experience is exactly the same as mine, but I'd have to be an outlier in a number of different ways (bad luck in number of unrateds played, bad luck in number of under-rated opponents played) for there not to be something wider going on. That and the fact that my experience in Spain was just totally different is what leads me to feel it's not just a problem with the conversion.

User avatar
Christopher Kreuzer
Posts: 8781
Joined: Fri Aug 06, 2010 2:34 am
Location: London

Re: ECF grades compared with FIDE ratings

Post by Christopher Kreuzer » Sun Feb 02, 2014 11:07 pm

Jonathan Bryant wrote:@Chris: I'm not sure about numbers required. I suspect it's the wrong question anyway. If the system is inherently deflationary then it doesn't matter how many games you play.
Out of interest, why do you think the system will correct itself? The factors that are causing the deflation seem pretty permanent to me. And my personal situation is that as I play more and do better my FIDE is rising a bit, but my ECF is going up even more so relatively speaking the situation is indeed getting worse.
I think it can correct itself if you play games outside of a national pool - it is meant after all to be an international rating system. If you get groups in each country only playing within their own rating pool and (crucially) gaining their first ratings against under-rated people from within that national rating pool, then you will get discrepancies. Talk to people from other chess federations and see if they think the ECF conversion formula is correct. I don't know enough people from other chess federations to be able to comment, but from what the couple of conversations I've had, the impression I get is that the conversion formula is not that helpful in predicting results. What it needs is someone to crunch the numbers for a wider range of games and results, not just anecdotal evidence from individuals.

Jonathan Bryant
Posts: 3452
Joined: Sun May 11, 2008 3:54 pm

Re: ECF grades compared with FIDE ratings

Post by Jonathan Bryant » Mon Feb 03, 2014 7:52 am

Christopher Kreuzer wrote:
I think it can correct itself if you play games outside of a national pool ... not just anecdotal evidence from individuals.

Oh for sure - playing games outside England (perhaps even outside the London area, I don't know) - will definitely help. Not sure it's going to happen on a systematic basis though.

And you're right of course, in my survey n=1. Although everybody I meet is also under-rated on the conversion, as I've said.

Talking of conversion formulas,
somebody mathsy/statsy could remind me, what's that thing you do when you have two variables and you want to draw a line that's a best fit between the data points. Is it regression analysis?

Whatever, I'd be interested to see what the best (or least bad) conversion formula actually is at the moment.

Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: ECF grades compared with FIDE ratings

Post by Roger de Coverly » Mon Feb 03, 2014 9:11 am

Jonathan Bryant wrote: Whatever, I'd be interested to see what the best (or least bad) conversion formula actually is at the moment.
See earlier in the thread
http://www.ecforum.org.uk/viewtopic.php ... 15#p130770
http://www.englishchess.org.uk/wp-conte ... 201401.pdf

MartinCarpenter
Posts: 3041
Joined: Tue May 24, 2011 10:58 am

Re: ECF grades compared with FIDE ratings

Post by MartinCarpenter » Mon Feb 03, 2014 2:47 pm

It does seem to be two different problems maybe in some ways the same root cause :) Elo is a conservatively updated grading formula. Hence the need for juniors to play lots of games a year to keep up with their rapidly improving grades. Something of the same requirement will apply to someone who's initial grade is a good bit out. Obviously, unlike a continously improving junior, that's a one off error so it'll eventually fix itself over time but it won't be a speedy procedure unless you're playing quite a few games a year.

So, at lower levels with many players on only a few games/year, you get major systemic deflation from juniors not playing enough, being badly underrated and then this spreading etc and then some general randomness to go with it.....

I'm sure you could design a grading system to cope somewhat better with this, but it'd then behave fairly awfully when applied to the top players. Their grades are of course hugely more important than those of UK club players :)

User avatar
IM Jack Rudd
Posts: 4815
Joined: Tue Apr 17, 2007 1:13 am
Location: Bideford
Contact:

Re: ECF grades compared with FIDE ratings

Post by IM Jack Rudd » Mon Feb 03, 2014 2:55 pm

MartinCarpenter wrote:So, at lower levels with many players on only a few games/year, you get major systemic deflation from juniors not playing enough, being badly underrated and then this spreading etc and then some general randomness to go with it.....
That's not quite the case. If juniors were playing enough, the system would still be deflationary. The root cause here is that the improving juniors have to take their rating points from other people, and there's no systemic way of damping this down. (The ECF system, by contrast, does have such a method: play lots of games against adults and the odd game you play against a junior will have a much lower effect on your grade.)

Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: ECF grades compared with FIDE ratings

Post by Roger de Coverly » Mon Feb 03, 2014 3:16 pm

IM Jack Rudd wrote: The root cause here is that the improving juniors have to take their rating points from other people, and there's no systemic way of damping this down.
Until the ECF came up with the potentially dubious hack of treating Juniors as new players every year, it would add bonus points if you played a Junior. In its classic form from about the mid-seventies to the early nineties, you would score (junior grade -40) for a loss, (junior grade + 10) for a draw and (junior grade + 60) for a win. When first introduced, it had only been +5. Before this was introduced in the late sixties, the pool of top players appeared to be reducing in size. The was despite or perhaps because of the popularity of chess among teenagers and young adults.

It was later believed the 10 point addition was an overcorrection causing mean and average adult grades to increase. As a consequence it was cut back so as to be age related with older ages getting less increment. For convenience the printed published grades included the add-on.

I'm not sure when or whether the changes will go through, but FIDE have a proposal to increase the K factor for all Juniors under consideration. This will accelerate rating changes upwards for improving Juniors at the cost of magnifying the effect of the odd bad performance.

Matt Fletcher
Posts: 271
Joined: Wed Mar 09, 2011 9:42 pm

Re: ECF grades compared with FIDE ratings

Post by Matt Fletcher » Mon Feb 03, 2014 8:30 pm

I had a quick look at the pdf which looked interesting - was a bit confused by the couple of 2800+ players (presumably there are some overseas players in the calculation), and had a go at seeing if I could knock some data together.

Based on the February 2014 FIDE list and the January 2014 ECF list, it looks like there are ~1500 people listed as from England or Wales with both an ECF and a FIDE grade.

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.

If you convert all ECF grades to FIDE on this simplified basis and compare to actual ratings, you get an average difference of just 2.5 points which seems close enough. Of the 1500-odd players, around 1100 (73%) have conversions within +/- 100 of their actual FIDE ratings, 580 of these having conversions below and 520 above their ratings which again seems reasonable. It also works pretty well across the grade bands and seems better for the highest bands (which is probably desirable) - I've attached a picture which shows the proportion of various bands whose conversions fall close to and far from their actual grades.

Compared to what I always remember as the conversion of 8 x ECF + 600, this crosses over at 150 ECF = 1800 FIDE and will rate higher-graded players lower and lower-graded players higher.

For a 180-graded player (just to take a random example :) ) it would give a rating of 2025 instead of the 'standard' 2040.
Attachments
ecf fide.png
ecf fide.png (32.88 KiB) Viewed 4031 times

Jonathan Bryant
Posts: 3452
Joined: Sun May 11, 2008 3:54 pm

Re: ECF grades compared with FIDE ratings

Post by Jonathan Bryant » Mon Feb 03, 2014 9:58 pm

IM Jack Rudd wrote:
MartinCarpenter wrote:So, at lower levels with many players on only a few games/year, you get major systemic deflation from juniors not playing enough, being badly underrated and then this spreading etc and then some general randomness to go with it.....
That's not quite the case. If juniors were playing enough, the system would still be deflationary.

Aside from what Jack says, my experience is not that the under-rated juniors aren't playing much. They seem to be much more active than me. So their ell-rating is going up - it's just that their ECF is going up even faster.

Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: ECF grades compared with FIDE ratings

Post by Roger de Coverly » Tue Feb 25, 2014 6:44 pm

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.
They've gone for 7.5 * ECF + 700 = FIDE

http://www.englishchess.org.uk/fide-elo-conversion/

The 4NCL uses a conversion formula, since you can now elect at an individual player level for ECF grades to be used to determine board orders. Otherwise the change in formula affects the use of foreign ratings in the domestic system when players report their Foreign results. You might use it to get an initial ECF estimate in leagues or non-FIDE tournaments for players new to the UK.

There's still the problem of the International ratings lagging the ECF grades and ECF grades jumping all over the place for active players because of the half-yearly updates and the double count problem this can create for the marginally less active.

User avatar
David Shepherd
Posts: 912
Joined: Fri Nov 23, 2007 3:46 pm

Re: ECF grades compared with FIDE ratings

Post by David Shepherd » Wed Feb 26, 2014 9:30 am

Matt, thanks for posting the above chart - I think it is a very good way of presenting the data. I don't know how easy it is with the data you have, but I would be interested to see the same chart but with only juniors included.

Matt Fletcher
Posts: 271
Joined: Wed Mar 09, 2011 9:42 pm

Re: ECF grades compared with FIDE ratings

Post by Matt Fletcher » Wed Feb 26, 2014 4:43 pm

David Shepherd wrote:Matt, thanks for posting the above chart - I think it is a very good way of presenting the data. I don't know how easy it is with the data you have, but I would be interested to see the same chart but with only juniors included.
David, it's fairly easy as FIDE ratings have DOB attached for the majority of people. Attached is the same chart for just those with DOB >=1996, based on the conversion FIDE = ECF x 7.5 + 700. There are around 200 juniors covered
ecf fide juniors.png
ecf fide juniors.png (24.88 KiB) Viewed 3609 times
Basically if my calcsulations are are right, the ECF grades for juniors give converted ratings that are about 150 points on average higher than their actual ratings. (equivalent to 20 points ECF difference)

Brian Valentine
Posts: 574
Joined: Fri Apr 03, 2009 1:30 pm

Re: ECF grades compared with FIDE ratings

Post by Brian Valentine » Wed Feb 26, 2014 4:57 pm

Matt's interesting figures bring us back full circle to my original 2009 post.

In the consultation process there was one interesting contribution from Michael Whale. He pointed out that if one does linear regression FIDE on ECF rather than what Matt has done and I (in the consultation) did then the fit (after inversion) is different. Technically the slopes are proportional to the respective variance of the independent variable.

The FIDE to ECF slope on the same data as in my graph was 0.127, nearly 8 in "sterling currency". I don't think this changes the situation with the ENG juniors but it does show we need to treat the statistics with care. (My pairs included non ENG players in both lists hence some 280+ from the London Classic).

Brian
Manager ECF Grading

User avatar
David Shepherd
Posts: 912
Joined: Fri Nov 23, 2007 3:46 pm

Re: ECF grades compared with FIDE ratings

Post by David Shepherd » Wed Feb 26, 2014 5:57 pm

Matt, many thanks - so the best fit for the adults and juniors is different.

In the range ECF 160 to 179 in a significant number of cases using the conversion leads to more than 250 points difference and the difference of 150 still remains for the players graded over 200 over.

This data shows that the juniors are significantly under-rated compared with adults (assuming the ECF rating is a correct reflection of their playing strength which it may not be). I suspect this will come as no surprise to most adults who play regularly in FIDE rated tournaments.

If the juniors were removed the fit for the remaining players would be different, so the difference between the two is likely to be more than 150 FIDE points. The average adult graded 150 above their junior opponent should expect to lose a significant amount of rating points :(

Post Reply