When is the new grading list coming out?

[Adam Raoof]

Please can we scrap the outdated and unreliable ECF grading system completely and move towards FIDE ratings for all players, regardless of age?

Adam,
I hope you really meant to suggest that we should move to an Elo based system, something that many others would support. As Roger has clearly pointed out, however, there are a large number of obstacles to make moving towards universal FIDE rating almost insuperable. Even if they were possible, why would you want to mortgage the future of chess in England to a such an incorrigibly corrupt organisation?
Regards,
Paul.

Yes, I meant Elo, sorry.

[Roger de Coverly]

Yet nothing dramatic has happened to their playing ability. It is just that one won most of his matches this season and the others didn't.

It's been variously pointed out that grading systems use (known) results as an attempt to measure/estimate (unknown) playing strength. It's also been pointed out that there's a trade-off between use of most recent results and stability of the estimates.

The ECF grading team seem to got fixated almost to the point of obsession with the notion that the measure of (unknown) playing strength is correct only if it predicts results in the forthcoming season. They seriously need to take on board the notions that (a) there is (and needs to be) such a thing as ratings lag and (b) that for inexperienced players in general and younger juniors in particular that playing strength is itself volatile rather than stable.

The ECF grading system is getting on for fifty years old. None of these ad hoc panic measures like not grading juniors have ever seemed necessary in the past.

Could we not return to two simple premises?

(a) the grade is a measure of actual performance over the previous season which is "stabilised" with previous season performances for players falling below a "number of games" qualification standard.
(b) the grade is used as the input to next year's calculations unless there are good reasons not to. These reasons might include the anti-deflation measure of a junior increment. They might also include ignoring the previous season grade if the current season's results were very different from the previous. ( this is for 40 point reasons as much as anything else)

[Paul McKeown]

Roger,

I agree.

Paul.

[Alex Holowczak]

The ECF grading team seem to got fixated almost to the point of obsession with the notion that the measure of (unknown) playing strength is correct only if it predicts results in the forthcoming season. They seriously need to take on board the notions that (a) there is (and needs to be) such a thing as ratings lag and (b) that for inexperienced players in general and younger juniors in particular that playing strength is itself volatile rather than stable.

Indeed, they said that instead of someone graded 190 having a 60-40 chance against someone graded 180, it was 57-43 instead. As a result, the system was changed. What exactly was wrong with it not being 60-40?

[Roger de Coverly]

an estimate will be calculated based on repeated iterations, with the iterations stopping when it is determined that "the figures [have] (more or less) stop changing". I know I'm not the first to say it but, by itself this surely won't work if a player scores highly or lowly.

I think you are correct and I think this solves the GC Taylor mystery in the London League. He had a really good returning season and scored 9.5/10 in division 3 against a field ranging from 107 to 157 in 2008 published and 133 to 171 on new grades. However one of his opponents was ungraded ...

As a consequence, the grading program went off in an uncontrolled loop and applied the 40 point rule for as long as it could. This put him in the top 50 as previously noted.

The same effect may well have afflicted some of the juniors with unrealistically high grades.

[Paul McKeown]

Indeed, they said that instead of someone graded 190 having a 60-40 chance against someone graded 180, it was 57-43 instead. As a result, the system was changed. What exactly was wrong with it not being 60-40?

Alex, the thing is, though, the 60-40 is actually a limit to the expectation for the stronger party. It depends on the uncertainty with which the two grades are estimated. If they were known with 100% certainty, then the expectation of the stronger player would be 60%. As they are never known with 100% certainty, then the expectation of the stronger player should be less than 60%.

[Roger de Coverly]

ndeed, they said that instead of someone graded 190 having a 60-40 chance against someone graded 180, it was 57-43 instead. As a result, the system was changed. What exactly was wrong with it not being 60-40?

Various statements were made with little (in my opinion) convincing backup as to exactly how this was measured.

All I think they did was used the (possibly bugged) recursion program to pretend everyone was a new player. They then compared the outputs from this program to the actual grades and started jumping to conclusions.

It's my personal opinion that it's the recursion program itself that's the major issue. I say that because it's the only new feature over the last 10 years and it's only been over the last 10 years that there's been or seems to be a problem with grades in the sub 120 area. My pet theory is that since the recursion program replaced the "graders estimate" for new players, it inadvertently removed an anti-deflation device from the foot of the system. I don't think a (human) grader would ever estimate a new player as negative or indeed 1 or 2. My fix would have been to give all new players a minimum grade of say 50 (or something age related). They've got detail results back to about 2002, so they could have parallel run this to find out the practical effects. My guess is that many of the now "overrated" players would only play for a couple of years and so the effect on the 120 - grandmaster range would have been limited. It would however have boosted grades of improvers in the sub-120 range

[Paul McKeown]

Alex,

You might like to read what Mark Glickman had to say about the perceived deflation of rating systems: A Comprehensive Guide to Chess Ratings . Bear in mind he received a Ph. D. from Harvard for his work on grading systems.

Essentially the deflation premise should be treated with a healthy dose of scepticism.

This whole fiasco started when it was pointed out in the early 2000's that FIDE ratings and BCF grades seemed out of kilter according to the formula FIDE = BCF * 8 + 600. Of course this was true, but the real problem lay with the FIDE ratings being inflated. What has happened since has had the same heavy inevitability as a Greek tragedy.

You can't criticise people's brainchildren though: they take it personally.

Best Regards,
Paul.

[Roger de Coverly]

Come the new grading system one is graded 156 and the other two are both graded 73 . Yet nothing dramatic has happened to their playing ability. It is just that one won most of his matches this season and the others didn't.

I now think this (and other junior anomalies) may be the same issue as the GC Taylor London League anamoly. For heavily scoring players playing at least one unrated opponent, the new player estimation routine can rampage out of control until it runs out of 40 point differences to consume. The same may well apply to players scoring near zero percentages.

I had always supposed the estimation program had a sort of safety switch to prevent this "infinity" effect. Perhaps I was wrong and this out of control effect is also the cause of zero and negative grades and deflation at the foot of the grading system.

[Leonard Barden]

Roger may have his finger on it. In the case of Kilpatrick and his hyper-inflated 234 grade, he scored 9/9 against an approximate 170 field on Board 1 for Dulwich in London League Div 2. One of the nine opponents was ungraded.....

Incidentallty Kilpatrick's new ECF grade comes from only 26 games. Consult his Fide card and you will see he has 35 Fide rated games in the relevant period (Andorra Open, U16 Olympiad, world U16, 4NCL) which makes a minimum of 44 actually played. It looks as if the ECF haven''t graded the two events in which he was representing them.....

[Roger de Coverly]

Does anyone know what Nigel Short was graded at at 13 and at what age he became a grandmaster?

According to the November 1979 BCM he was 215 for 1979 (age 13 in ECF rules) up from 206 in 1978.

According to wiki, he became a grandmaster at the age of 19 -- but it was more difficult then !

[E Michael White]

I now think this (and other junior anomalies) may be the same issue as the GC Taylor London League anomaly. For heavily scoring players playing at least one unrated opponent, the new player estimation routine can rampage out of control until it runs out of 40 point differences to consume. The same may well apply to players scoring near zero percentages.

High scoring players will cause issues and ideally should be switched to a Poisson or similar rare event distribution. I dont think this is the only problem we have now or previously. In the past graders were convinced the only issue for deflation was the grading of juniors; this is clearly not true for example underactive slowly deteriorating seniors produce similar but lower volume amounts of internal deflation. I think its a part of chess player psyche to believe that they can identify a single cause, fix that and the rest is a walk in the park. Chess players spend a lifetime training themselves to pick the best move and system and probably overanalyse that to the exclusion of alternative thoughts.

I dont know the exact new basis for new starters but I did manage to obtain replies on this forum from Richard Hadrell and later Howard Grist described the old iterative process used. Their two reply sets conflicted slightly but the basis may have altered in the meantime. What mathematicians describe as iterative processes might be better described as a reiterative process and similarly when an MP says he wants to reiterate something they usually mean repeat it. The iterative process used previously is really a red herring as the same figures can be obtained with a decent array processor solving lots of simultaneous equations. But what can happen as I and others have said on this form in one guise or another is for local pyramiding to occur, a bit similar to banking overdrafts.

Ian Thompson gives an example of local pyramiding but it is not necessary to enlist Michael Adams to make it work. Basically if a new club of juniors starts up with unrated juniors and the coach splits them into 4 groups either by age U14, U13 etc or on his assessment of strength into premier, major, inter etc and they play events to be sent for grading, which may be swiss or APA at that stage they cannot be graded. The problems come if the lowest player from group A plays the top player from the next group down in a graded adult event such as a second team county fixture and wins. All players in Group A will eventually be rated higher than group B. Take this down in steps and the lowest group D player plays as a reserve in just one adult 4th division league match against a graded player and whether he wins or losers, group A will have very high grades. If that adults mobile phone goes off he could create several 275 junior 14 year olds ! Obviously this is an extreme example but to one degree or another this happens in junior clubs and rating limited events.

[E Michael White]

Presumably only in the first iteration. His other games in the season will probably give him a grading performance of 200 or so, in which case for the second iteration, the game will count as 110 because of the 40-point rule.

Jack I don’t think the iterative process can use a 40 point rule as it would not then normally converge.

[Roger de Coverly]

High scoring players will cause issues and ideally should be switched to a Poisson or similar rare event distribution.

It's not just high scoring players. There may also be problems with low scoring players. I may have found a relatively local player who scored 1.5 from 6 against a field averaging 108(new). I would have thought this was worth about 75 -- instead 24. Needless to say, a couple of the opponents were unrated.

underactive slowly deteriorating seniors produce similar but lower volume amounts of internal deflation.

If improving players cause deflation, then surely deteriorating players cause inflation. They will have reduced the skill set of their grading band thereby handing out free points to players who have only improved relative to said senior.

[Paul McKeown]

If improving players cause deflation, then surely deteriorating players cause inflation. They will have reduced the skill set of their grading band thereby handing out free points to players who have only improved relative to said senior.

Absolutely - a standard idea with respect to ratings/gradings that has been understood for decades.