Chess Grading Destruction

[Stewart Reuben]

Brian Towers >I'm curious as to whether the ECF has sound reasons other NIH and tradition for not moving towards its own version of Elo as many other countries have done. If they do exist then perhaps they should be publicized so that other federations and perhaps even FIDE itself could seriously consider moving towards an ECF grading system.

THE BIGGIE IS THAT THE ECF SYSTEM CAN BE EXPLAINED VERY SIMPLY AND QUICKLY
Each game you play, you get your opponnt's grade + 50 if you win; your opponent's grade if you draw; your opponent grade - 50 if you lose. You take the average of all the games played.
(I do realise there are special points, e.g. the 40 point rule and what happens if you play +-30 games in the period.)
The FIDE System is much harder to explain and understand.

Arpad Elo and Sir Richard Clarke worked together on the USCF Rating System and BCF Grading Systems. Sir Richard was obviously deeply conscious that people might attach to a small difference in grades. Thus he started out with categories which basically covers 8 grading points. If you have a grade 200 as opposed to 201, we know it isn't very different. But an unsophisticated person might well think that 2200 and 2210 are statistically signicant.
3 figures may also have been preferred because printing such numbers was cheaper than 4.

The ECF system is based on the assumption that players meet opponents quite close in strength. Basically in the expected range 40-60%. Then the relationship between difference in grading or rating is linear. According to Elo the system outside that range is non-linear. ECF ignores this.
percentage rating difference
95 470
90 366
85 296
80 240
75 193
70 149
65 110
60 72
55 36
50 0
In FIDE each game is rated separately and the plus/minus determined over a month. The ECF takes the average over 6 months. It is taking an average of figures that have a non-linear relationship.

In my opinion, if the ECF moved to monthly grading lists, there would be an increase in competitive play, in due course, of approximately 10%. That assertion is based on what has happened in the US, FIDE and the London Rating System when it existed.

[MartinCarpenter]

Look this incentive thing is basically nonsense too If you genuinely improve then your ECF grade will go up and stay up. Quite fast too if active.

That's an inventive, because it means something. Randomly pushing peoples grades up/down by 10/20 pts just because of which of their games happened to start/finish a season simply makes the grading system look/feel like nonsense.

It also makes the grades rather worse at their more important tasks like assigning people to congress levels, board orders etc.

Yorkshire actually does both Live grading lists during the year but then a summary list produced every ~July. The summary list is what everyone pays any attention to. The live grades are just amusing/sometimes plain silly.

[Michael Farthing]

Then perhaps Yorkshire has the answer - an annual grade that means something and a day by day grade that's nonsense but exciting. Should satisfy everyone.

[Maurice Lawson]

Martin, I do not agree at all that the incentive thing is nonsense. On the contrary it is central in a competitive activity. Neither do I have much confidence in either the ECF or the FIDE systems to actually get to what you call a genuine grade. This is proved by the wide disparity between grades and ratings going significantly both ways when trying to make comparisons. But at least the FIDE system is designed to track sensibly. ECF is not. Stewart makes excellent points about the archaic nature of ECF - but at least previously it was a more up to date tracker, and a better competitive tool. It is now even more archaic and should be buried.

[MartinCarpenter]

Having a stable list and a strong tracking one does seem a sane compromise. They shouldn't mix too much though.

If the old (new!) 30 game rolling window was a good way of tracking I'd be much more sympathetic! It really wasn't though. The problems all come round the edges, because you get games/tournaments suddenly going from counting at full weight to not counting at all from one list to another and huge jumps in grades from one list to the next.

The Yorkshire system tracks much better than that - it starts the weight of each played game at 1 then exponentially degrades it by 1/12 each month (think he eventually rounds them down to zero!). So its all a continuous process and you don't get the horrible randomness the game window introduced.

The Yorkshire system has a big issue mind. Its great if you play 20+ games a year. Once you only play 10-15 Yorkshire graded games a year your grade ends up a bit volatile - but less than 10 a year and the grade is really fairly random. It can't go back to previous years to up the sample size ECF does of course.

[Richard Bates]

I struggle to see the "logic" in arguing that a grading system which is essentially(with the minor tweak that it is now published six monthly, and other things which pre-date the six-monthly lists eg. junior calculations) the same system that has operated for the vast majority of time that the ECF has published a grading list, can be considered "corrupted and unfit for purpose". With nothing in the way of hard evidence provided. Especially any attempt to try to hold it up an unfavourable in comparison with FIDE ratings!

Unless i'm misunderstanding even some of the criticisms don't seem to make much sense - the grading list has always been based on periods of time - the number of games has only ever been considered a 'minimum' to generate statistical validity. It is true that the previous six monthly method made it more likely that games would be prioritised over time period, but it wasn't how the system was designed.

The argument over rolling yearly averages vs six monthly averages is worthwhile on some level, but hardly a fundamental corruption of the system. The most important change IMO was the getting rid of the previous "countback" system which generated significant noise and produced very unwelcome perverse incentives not to play.

My main bugbear at the moment would not be the system itself, but the fact that having introduced six monthly lists, ECF (and other competitions) have a habit of preferring to use the July list exclusively for qualification purposes. Which seems to me to undermine some of the value of two lists.

[Michael Flatt]

When making comparisons between ECF and FIDE grading systems there are other points to note other than simply the mathematical method and the use of either a 3 or 4 figure digit number to indicate ones rating. Time controls, rates of play, number of games played in a day and Arbiter qualifications are different and generally more restrictive in FIDE rated events.

Many English events simply would not satisfy the FIDE rating criteria. The current ECF grading system has undergone some tinkering in recent years but it still remains an effective tool for distinguishing whether, on average, one player is stronger than another and ranking them in accordance with the requirements of the Swiss pairing system.

The maximum grade of each graded section in a competition are varied according to the whims of the organiser. There is no significance to assigning a 160 or 150 grade limit other than to divide the players into conveniently populated sections. The preference for weekend tournaments in England seems to support narrow banded sections rather than simply an Open tournament.

Do typical club players play sufficient games each season to justify going over to monthly grading? Six monthly list seem to work well, perhaps the next step is quarterly?

[IM Jack Rudd]

When making comparisons between ECF and FIDE grading systems there are other points to note other than simply the mathematical method and the use of either a 3 or 4 figure digit number to indicate ones rating. Time controls, rates of play, number of games played in a day and Arbiter qualifications are different and generally more restrictive in FIDE rated events.

This is true but irrelevant. Brian was talking about a national Elo system, which can have whatever requirements the federation wants. Many other countries have them.

[Michael Flatt]

This is true but irrelevant. Brian was talking about a national Elo system, which can have whatever requirements the federation wants. Many other countries have them.

To be honest, it is not at all clear as to what point the original poster is trying to make.

What specifically does the ECF grading system fail to do and what is the evidence?

[MartinCarpenter]

The coherent version of the argument seems to be that he wants a grading system strongly tracking recent form - preferring sporting interest (or some such) vs 'accurate'/stable grades.

The recent reversions unarguably make ECF in general a fairly stable/conservative system. The counting back to previous years to get to 30, taking all games every year etc. It can often take 2/3 lists for a jump in form to fully work its way through, although more often that not it vanishes before it does

Not that the previous way of trying to do that really worked as it introduced too much noise, but there is a legitimate debate as to whether you'd prefer something closely tracking ups/downs in form through the year and frequent lists, or something smoothing them out.

Or both, yearly one for competitions, board orders etc, frequent one for 'fun'.

[Roger de Coverly]

Tbut there is a legitimate debate as to whether you'd prefer something closely tracking ups/downs in form through the year and frequent lists, or something smoothing them out.

I think you should do both. A relatively stable value used where you need a ranking for eligibility, board orders, seeded pairings etc. coupled with a measure, however defined, of recent results.

The ECF system by its vary nature includes a number of disparate events with a mix of time controls and playing conditions. From that point of view, treating a year of chess activity as the basic building block is very logical as it's only year on year that the mix of events has a reasonable chance of being comparable.

[Maurice Lawson]

Many of you have widely diverted yourselves from the point I am making, on to other peripheral issues. In fact what I'm saying is that the ECF system over the last few years, whilst of course not perfect, has actually been a much more reliable indicator of recent playing strength than has been FIDE. ECF has indeed been a good system for the last few years! It has also been better than it will now be after the January changes. It was a good playing strength indicator and a good motivator as well. It would have been even better if it were to be published more often, and if it were weighted towards recent games instead of being simply averaged. In short however, it was a decent compromise whilst being imperfect.
The FIDE system is particularly poor for club strength players - it reacts far too slowly because of the too low K factor (despite the recent change) and misses too many games actually played. The new ECF system will now react even more slowly than FIDE (which many of you posting comments will think is wonderful!). My view is that there is no benefit at all in the changes made (since artificial smoothing over a long period is in my opinion a disbenefit) and there has been much lost in order to achieve that disbenefit. That is why I describe what is left as anodyne. The system is spoiled and chess players will not benefit from the change.
Justifying this backwards move on the grounds that it was done for decades is a very lazy argument. What we now have most definitely will not stand up to the modern world. We need to be progressive, not backward.

[Michael Flatt]

Martin,

Thank you for explaining. So, to be able to calculate grades at more frequent intervals wouldn't it require leagues, for instance, to upload results to the grading database at more frequent intervals? Yet more work for the County grader!

Sometimes league games are not completed in one evening and are carried over to a time convenient for both players to meet and conclude the game, or await an adjudication. Would the change lead to more pressure on players to finish games in a single session?

Are there sufficient players who regularly compete in weekend and week-long tournaments to justify changing the grading system to publish monthly grades?

Would there be any benefit to the average club player, who might only play 12 league games and a few club championship games in one season?

[Maurice Lawson]

This is what I actually said....
ECF has indeed been a good system for the last few years! It has also been better than it will now be after the January changes. It was a good playing strength indicator and a good motivator as well. It would have been even better if it were to be published more often, and if it were weighted towards recent games instead of being simply averaged.
I said it was good! I did not say it had to be published monthly - only that it would be (even) better if it were. Please don't try to divert attention from the main issue.

However, it is inevitable that in time, with further integration and development, we will indeed end up with greater frequency of publication. It would be rather luddite to think any differently. So for example at some stage quarterly publication might be an interim stage before going to monthly - but it will inevitably happen at some stage, and that will happen at the point where the software allows it to happen without the leagues of people having to intervene.

At some point the ECF needs to stop focussing excessively on grading stability and start using the grading system to promote competitive chess. Requiring the system to track back over a full year, regardless of the number of games played, and regardless of the recency of games played, cannot possibly endure as a system and it would be better to recognise that now and start planning for something contemporary and competent which takes chess forwards instead of backwards.

[Michael Flatt]

The confusion may have arisen from the title of this thread, "Chess Grading Destruction".

I think that there would be merit in giving greater weight to more recent games.