Page 15 of 19

Re: ECF to scrap its grading system

Posted: Thu Jan 31, 2019 1:02 am
by NickFaulks
Richard Bates wrote:
Wed Jan 30, 2019 10:06 pm
And I think they will have to come up with a better solution to the “junior problem” than FIDE have managed.
The latest massive analysis of results showed that, aside from initial ratings still needing work, the predictive power of FIDE ratings for juniors is pretty good. At the risk of banging the point over the head, there is nothing that any system can do with results that are not reported.

Re: ECF to scrap its grading system

Posted: Thu Jan 31, 2019 6:07 am
by Richard Bates
NickFaulks wrote:
Thu Jan 31, 2019 1:02 am
Richard Bates wrote:
Wed Jan 30, 2019 10:06 pm
And I think they will have to come up with a better solution to the “junior problem” than FIDE have managed.
The latest massive analysis of results showed that, aside from initial ratings still needing work, the predictive power of FIDE ratings for juniors is pretty good. At the risk of banging the point over the head, there is nothing that any system can do with results that are not reported.
Is there a link to this analysis? I should add that this does not necessarily surprise me - I am not one who advocates the view (post K factor changes) that “all juniors are underrated”. My impression is that what has happened is that the volatility resulting from the large k factors has meant that “on average” a junior is much more likely to be closer to their true strength. However the corollary is that any individual junior is just as likely to be significantly overrated as underrated. So if you play a large number of juniors (as you say factoring out problems caused by initial ratings in combination with few games played) then on average the predictive power might be quite good, even if there could be huge variation on an individual game by game basis. But that’s just a hypothesis - i’m not sure whether analysis bears that out.

Anyway, I re-emphasise that this is in some ways irrelevant to the point I was trying to make - which is that it is important that players retain confidence in the system, and combining monthly ratings with a complete overhaul of the system that people understand and are used to, contains an obvious risk that they won’t. And that may not be easily assuaged by rational argument. That’s all.

Re: ECF to scrap its grading system

Posted: Thu Jan 31, 2019 7:53 am
by Adam Raoof
NickFaulks wrote:
Thu Jan 31, 2019 1:02 am
Richard Bates wrote:
Wed Jan 30, 2019 10:06 pm
And I think they will have to come up with a better solution to the “junior problem” than FIDE have managed.
The latest massive analysis of results showed that, aside from initial ratings still needing work, the predictive power of FIDE ratings for juniors is pretty good. At the risk of banging the point over the head, there is nothing that any system can do with results that are not reported.
It's not just that events are not all fide rated, but events that ARE rated do not include the results for all games that are played. That must make the rating system less accurate, though I don't know the actual impact. Anecdotally this is the hardest thing to explain to new players who want to understand their fide profile and the absence of a complete list of their games.

Re: ECF to scrap its grading system

Posted: Thu Jan 31, 2019 8:51 am
by Roger de Coverly
Adam Raoof wrote:
Thu Jan 31, 2019 7:53 am

It's not just that events are not all fide rated, but events that ARE rated do not include the results for all games that are played.
This is one of the issues that a domestic Elo system may have to address and resolve, namely a less stringent rule for the inclusion of results against new players. It's not totally against the rules for operating such systems to include them in some manner, the FIDE rules for all play all tournaments featuring unrated players being a case in point.

If you do want to include all results, less frequent lists such as the current ECF approach support this, as it enables a half season or season's worth of results to be built up before attempting an estimate of strength.

Re: ECF to scrap its grading system

Posted: Thu Jan 31, 2019 9:34 am
by Brian Valentine
Picking up a few points in recent posts.

I share Richard Bates concerns about acceptance of a radical change to Elo. However the ECF current approach to new entrants/juniors tends to breakdown with the smaller samples generated by monthly lists. The approach can be very volatile and explaining the inner working is far harder than Elo. There is also the technical issue more likely to arise of "islands" of results with no (or few) rated player benchmarks.

On Adam's point, nearly 8% of ECF results would not be graded if pure FIDE was introduced and that rather defeats the point of going monthly.

I can reassure Roger that using five years of recent results that there is a fix to the problem he outlines. The method gives better average predictive power based on a bigger dataset, where the added results are those least reliable. It also seems to address the FIDE shortcoming in an ECF context of deflation of grades/ratings particularly amongst juniors.

Re: ECF to scrap its grading system

Posted: Thu Jan 31, 2019 11:09 am
by Stewart Reuben
In earlier days the FIDE system did not give a player a published rating until he had played 9 games or more. The results were collated, but didn't affect the opponent's until after 9 games. Now it is just 5. That must make initial ratings more volatile.

Nick will remember the discussion about iteration of new players' results in Swisses. This would mean that all your games counted, provided your opponent achieved a rateable performance. He was wholly against that. But it has the effect that, playing an unrated opponent, means your game doesn't count for you. Thus you can throw the game. i had such an opportunity in an International Senior Team Chess Championship one year, but resisted the urge. I am sure not rating the game for the rated player is inflationary in some federations.

Re: ECF to scrap its grading system

Posted: Thu Jan 31, 2019 2:26 pm
by Alex Holowczak
Richard Bates wrote:
Wed Jan 30, 2019 10:06 pm
I would say it is simplistic to state that problems with FIDE/ELO rating are simply explained by not all games counting towards them. It is the excuse usually given by its defenders but that doesn’t mean it’s true, or at least isn’t the only issue.
On the contrary; this is not the "excuse usually given by defenders", it's the reason given by those who have a problem with the system. I know this because I've had the conversation with those people in England, at various events. You can hardly blame the "defenders" for then quoting that argument themselves!

Re: ECF to scrap its grading system

Posted: Thu Jan 31, 2019 3:06 pm
by IM Jack Rudd
The issue with FIDE ratings is not that most English games don't count towards them, but that it matters that they don't.

Re: ECF to scrap its grading system

Posted: Mon Feb 04, 2019 9:04 am
by Roger de Coverly
There's quite a bit more in the recent Board minutes.

https://www.englishchess.org.uk/wp-cont ... ration.pdf
Use of ECF/ Clarke or Elo ratings – There was discussion on the preferred rating system given the need to change the grading system to support monthly grades and the resulting opportunity to change the rating system. The Board had an open mind at this stage as to whether the Clarke methodology should be retained (subject to its ability to deal properly with monthly grading) or whether a move to an Elo methodology should be made. The board's preference is to move from the Clarke system’s 3 digit grades to 4 digit ratings, but it was agreed that there should be a consultation process to confirm this would be broadly acceptable to members and supporters, recognising that there may be strong views on both sides.
Also
Treatment of non-rated player games – It was agreed that all results should be graded including non-rated player games, notwithstanding this is different from the FIDE approach
So a direct copy of the international system is out of the window.

Re: ECF to scrap its grading system

Posted: Mon Feb 04, 2019 11:01 pm
by E Michael White
NickFaulks wrote:
Thu Jan 31, 2019 1:02 am
The latest massive analysis of results showed that, aside from initial ratings still needing work, the predictive power of FIDE ratings for juniors is pretty good. At the risk of banging the point over the head, there is nothing that any system can do with results that are not reported.
I might be saying the same thing as Richard Bates but don't assume that if the theoretical predicted results leave the ratings in the same state that that shows that the ratings or model probabilities are correct. For a set of ratings and results there is more than one set of model probabilities that leave the ratings unchanged.

Re: ECF to scrap its grading system

Posted: Tue Feb 05, 2019 12:54 am
by Roger de Coverly
E Michael White wrote:
Mon Feb 04, 2019 11:01 pm
For a set of ratings and results there is more than one set of model probabilities that leave the ratings unchanged.
General reasoning suggests that if "on average" players score 50% and also "on average" they play opposition of the same strength as themselves, it's only at the extremities that the absolute values of ratings matter.

When the international ratings had a qualification cutoff of 2000, that was inflationary because only results of players rated above 2000 would count. The same premise is implied with a cutoff of 1000, but at that level, improving by hundreds of points is commonplace, so the effect is diluted.

Re: ECF to scrap its grading system

Posted: Tue Feb 05, 2019 1:10 am
by Stewart Reuben
Initially the FIDE cutoff was 2200 and the ECF one 200.

Re: ECF to scrap its grading system

Posted: Tue Feb 05, 2019 4:01 pm
by Brian Valentine
E Michael White wrote:
Mon Feb 04, 2019 11:01 pm
I might be saying the same thing as Richard Bates but don't assume that if the theoretical predicted results leave the ratings in the same state that that shows that the ratings or model probabilities are correct. For a set of ratings and results there is more than one set of model probabilities that leave the ratings unchanged.
I'm trying to understand this remark, maybe I'm missing a nuance. If n players play in a period and their ratings are unchanged, in normal circumstances, there are n equations and n ratings to recalculate. Doesn't that mean that there is a unique solution?

Re: ECF to scrap its grading system

Posted: Tue Feb 05, 2019 6:30 pm
by IM Jack Rudd
Brian Valentine wrote:
Tue Feb 05, 2019 4:01 pm
E Michael White wrote:
Mon Feb 04, 2019 11:01 pm
I might be saying the same thing as Richard Bates but don't assume that if the theoretical predicted results leave the ratings in the same state that that shows that the ratings or model probabilities are correct. For a set of ratings and results there is more than one set of model probabilities that leave the ratings unchanged.
I'm trying to understand this remark, maybe I'm missing a nuance. If n players play in a period and their ratings are unchanged, in normal circumstances, there are n equations and n ratings to recalculate. Doesn't that mean that there is a unique solution?
There's a unique solution for every specified set of win probabilities, yes. EMW's point was that two different sets of win probabilities can result in the same unique solution.

Re: ECF to scrap its grading system

Posted: Fri Feb 08, 2019 6:22 pm
by Roger de Coverly
The latest ECF newsletter announces a proposed implementation of January 2020. It means that those leagues, most of them, who ignore January grades, can at least complete their season.