English Chess Forum

I dont think the subsequent comments relating to this question are correct.

Previous to 2009 an ungraded junior would have been graded 260 taking your previous grade as 200. I pointed this out and several other anomalies during the 2009 rehash and I think the basis was changed so that the junior would be 250.

Since 2009 juniors, who have played more than 20 games, as in your example are treated as ungraded so his new grade would be 250 most likely or possibly 260.

The basis of grading juniors is not clear in the documentation produced after the 2009 rehash, and if you burrow into the calculations for juniors along the way you can have a discussion with Roger de C about lack of stability over time and Brian Valentine about whether start values are needed in the markov process and a few other people who think that the ELO system solves all the problems. Additionally arbiters seem to think that both systems predict results and justify the accelerated paring system in tournaments. Neither system is based on any concrete mathematics and thus are approximations to a goal which can never be achieved.

Even the name is approximately referred to as the Clarke BCF/ECF grading system, whereas in fact it was invented by a Scotsman Kenneth Harkness primarily for RR tournaments. Clarke instigated its approximate use for unbalanced leagues where players had different activity rates and, thus made the issues of deflation, stretch and local inertia more prominent. Cynics suggest that Harkness gave the ECF the grading system as a joke and later Scotland switched to an ELO based system but the ECF havent realised the joke yet. Perhaps this last point could be confirmed by Geoff Chandler who is an approximation to a Scotsman or Alex McFarlane who spends so much time over the boarder he could be considered as extrapolated English.

@Matt Mackenzie I assume because it's easier to programme the spreadsheet this way?

(Since the ECF website has no justification for or apparent awareness of this side-effect.)

_________________
http://streathambrixtonchess.blogspot.com/

The ECF grading team are a bright bunch of people. They know what they're doing. They probably just have other things to do than write a 50-page manual explaining the statistical ins and outs of it. The 2009-level explanation could probably do with an expansion to reflect how it currently works though.

_________________
February 16-17: 2013 BUCA Team Championship
April 20: 2013 British Blitz Championship
July 27-August 10: 100th British Chess Championships

Funnily, this feature (losing pts from winning games) is what will (according to my calculations) bar me from applying for "Club Master" status (when my next grade comes out) should I have both the desire for such a title and be a member of the ECF, neither of which will happen.

_________________
AKA Scott Stone

"Give a man fire and he's warm for a day, set fire to him and he's warm for the rest of his life."

That's Mr Stone to you, f**kface.

Back in around 1970 when this rule first came in, they didn't use computers and didn't itemise individual results. So a grading calculation was to add up the total of grades adjusted by the 40 point rule and then add the excess of wins over losses. That was all the central collators of grades (in those days the SCCU etc.) ever saw.

It's become an intrinsic part of the grading system, so you would need to know what and why you were doing it if you tampered with it. General reasoning suggests that it would accelerate the grading uplift of players doing well, whilst knocking down the grading of players doing badly. There are always persistent suspicions of players who always seem to do well in rating restricted events with high prize money, but who never seem to gain the points their results or even strength would indicate. You wouldn't want to make it easier for them .

For sure. Keeping things simple back then was very reasonable. Actually, it still is now. The current system clearly isn't perfect but at least those - like me and Sebastian Stone, for example - who like to calculate our own rating performance can do so the ECF way. I just had a look at how elo ratings are calculated. Hardly straightforward.

I do think an amendment to the 40 point rule is worth considering although even there the argument for keeping things as simple as possible could trump the benefit of the 'improvement'.
(my thinking, btw, is simply that there is sense behind the idea that one shouldn't lose rating points by winning games nor gain points by losing them, but to restrict the amount of points you can collect by winning a game doesn't have any logic at all as far as I can see; if changing the rule impacts on the grading system then so be it, would be my argument).


Actually, my interest in all this tends to be theoretical rather than practical. I find discussions around these sorts of statistical questions interesting in and of themselves. In practice it doesn't bother me too much that a system has flaws because a grade is only a general overview anyway.

When I break my results down by league/colour/strength of opponent etc the results vary greatly and since creating grades for each circumstance would be a nonsense, one rough approximation is good enough. So it seems to me, anyway.

_________________
http://www.streathambrixtonchess.blogspot.com

"Of all the stupid questions so far, This one is particularly annoying, and irrelevant and when I meet you, I propose to write it down on cardboard and nail it to your face."
- Jerry Sadowitz

In a reasonably pure form as for example used in the International rating by FIDE, it's relatively simple. You have a rating which is updated every two months. When you play a rated player you look up both their rating and yours and take the difference. You then look this up in a table to find your expected score. The difference between what you actually scored (1/.5/0 ) is differenced to this. This is multiplied by a k factor to get the change in rating that one game will contribute to the next published list. The k factor is 25 for new players, 15 for established ones and 10 if your rating has ever been above 2400.

You can do the expectation table approximately in your head if you note that a 200 point differential expects you to score 76%/24% and a 350 point differential 90%. You can use a linear approximation between 0 and 200 and 200 and 350 without being far wrong.

So a win against a player 200 points higher scores you 3.6 (at k=15), whereas a win against one 350 points higher is about double that at 6.75.

I had a bet with myself that somebody would respond with a post like this!


I'm not saying it's impossible, and I doubt I'd object if such a scheme was introduced by the ECF, but, really, following your calculation method really isn't as simple as

"If you win you get their grade plus 50; if you lose you get their grade minus 50; if you draw you get their grade"


is it?

_________________
http://www.streathambrixtonchess.blogspot.com

"Of all the stupid questions so far, This one is particularly annoying, and irrelevant and when I meet you, I propose to write it down on cardboard and nail it to your face."
- Jerry Sadowitz

Tournament organisers are increasingly adopting the annoying habit of basing rating prizes on the best "actual to expected" performance. Where they publish a computer generated cross-table, earlier rounds are easy, but you need a rough and ready way of calculating expectations in your head for the last round.

By comparison to obscurities like Glicko and Glicko-2, the FIDE system is indeed simple if a little more complex than the ECF.

I guess expected performance is the same number as their grade. So the grading prize goes to Tournament Performance Grade - Actual Grade. Shouldn't be too difficult, and I guess it's another way of skinning the Grading Prize cat.



But we're not comparing it to Glicko and Glicko-2, we're comparing it to the ECF system. The advantage of the ECF system is that you can calculate it in spreadsheets, or a separate document. At the very least, the FIDE algorithm requires carrying around look-up tables (or Stewart Reuben's book, before he plugs it ) if you want to calculate it on the fly. It is unquestionably more complicated than the ECF system.

_________________
February 16-17: 2013 BUCA Team Championship
April 20: 2013 British Blitz Championship
July 27-August 10: 100th British Chess Championships

The traditional way of doing grading prizes was to say that it was the best score by a player under x. This has the advantage that it's easy to see from the wall chart who your rivals are, or even whether you have any chance at all. The Elo based one has the difficultly that it's actually quite difficult to figure out who's in the running and for that matter who won.



It's not especially difficult to do look up tables in a spreadsheet, but yes it's more complex than the ECF method.

Agreed; I would never do a grading prize like that, but could understand why some would - particularly if you have a FIDE-rated Open offering grading prizes on the same lines.



It's just another thing to do though, isn't it?

_________________
February 16-17: 2013 BUCA Team Championship
April 20: 2013 British Blitz Championship
July 27-August 10: 100th British Chess Championships

Claiming that this supposed "flaw" in the ECF system demonstrates the superiority of the FIDE system is unjustified IMO. Any grading system which bases a new rating on performance based on the old one will suffer this "flaw" that a win against any player can cause a reduction in performance if that performance is sufficiently extreme. See also the standard claimed flaw in the ECF system that two players who play each other exclusively and perform at 50% will exchange grades rather than converge somewhere in the middle.

It is not difficult to find equal 'flaws' in the FIDE system. Just as with the ECF system they can be extremely inaccurate for rapidly improving (or declining) players. Play too many games and individuals can end up with ratings completely out of line with their performances. In an extreme case this can even be achieved by playing exclusively against very weak players. Considering the volume of games that many people play towards gaining their ECF grade this could be exaggerated if used for ECF domestic grading, which has the advantage of improving its accuracy with number of games played.

Both ECF and FIDE systems work very well for players who are established and of stable strength (and who don't try to manipulate the systems). And less well for everyone else. Little you can do to change that.

Were you aware that the ECF Council has just voted that you have to become a member (at a cost of £18) from September 2012?

No one ever did explain to me why my k factor is 15 despite my first ever rating being 2430 Perhaps it has to be over 2400 after a certain number of games, maybe 30??