When is the new grading list coming out?

[Neill Cooper]

Noticing that your son's grade had allegedly collapsed to 73, 114 is 41 points above this - so maybe the issues are connected to the 40 point rule in some way.

His grade was worked out from a base of 68 (i.e. his grade before the increment of 5 was added), so the loss to the 114 player scored 58, and the wins against the grade 8 and grade 9 players each scored 78.

[Roger de Coverly]

His grade was worked out from a base of 68

So after publishing a supposed regrading to one hundred and something, they then ignore this completely. If you don't chain in some manner grades from one year to the next doesn't the whole system collapse? Confidence in the system collapses.

[Neill Cooper]

Something has to be done.

Agreed.

I think the problem is with the 'estimating grades for new players'. I don't think either the theory or the code works very well, particularly if someone gets more than 85% or less than 15%. This was not a huge problem when only a few players (say 5%) were new, but with all juniors grades being reset it is now about 30% of players are treated as new. Also the estimation routine should only be iterated a couple of times, not to convergence, as it is the repeated iteration which leads to grossly inflated/deflated grades.

My solution is not to reset all junior grades. Best would be to not reset any of them, or failing that only under 11s.

[Adam Raoof]

For the next Golders Green it seems redundant to keep sections at Open, Under 160, Under 130 and Under 100. For September only I am considering doing what the Berks and Bucks Congress do, which is simply to take all the entries and put them in grading order, then divide into quartiles, and draw the section lines for the day that way.

What do readers think of that idea? It can't deal with weird and wacky grades, nor ungraded players, but for the majority of players the relative positions will remain more or less the same, even if their grades go up.

[John Upham]

If you wish to contact the official in charge :

Manager of Grading & Rating - David Thomas | 115 Haunch Lane | Kings Heath | Birmingham | B13 0PA

Telephone number : see the ECF Yearbook, not published on the ECF site for some reason.

[email protected]

I'm sure David will be happy to discuss the issues...

[Alex Holowczak]

Perhaps the simplest way to grade new players to the grading system would be to "ignore" grading results, and estimate them instead, not based on statistics, but on their apparent strength of play. Suppose you lose against lots of strong players in the UKCC Terafinal (without the junior bonus, because your birthday passed a month before it), then score 40% in an Under 100 congress later in the year after improvement. It's obvious that your grade is something like 60. Given the inflation that the new grades gave, that would maybe go up a bit with the new grades. Well, a friend of mine in that scenario ended up graded 32!

Common sense should suggest that 32 is wrong for the results that were had. Lots of losses to people > 100, and decent results against the people Under 100. Instead, estimating new players actually based on the opinions of the people who know/have played the relevant people may yield more accurate results. There seem to be a lot of examples of people previously mentioned who would have more accurate grades if the people who knew them suggested their playing strength. This could be a role of an arbiter at a tournament that an ungraded player enters.

[Mike Gunn]

What I mean by "the junior problem" is the deflationary affect on grades of a a large group of improving players. This was recognised by Richard Clarke who proposed (even if he did not implement it himself) correcting the grades of junior players. However, it's not just the deflationary effect on all grades it's the annoyance of how undergraded juniors generally hoover up the prize money in graded sections of congresses, rapidplays etc (annoyance to all us weakies who play in those sections, I mean).

The ECF has made a valiant attempt to deal with this but it has backfired and we need to take a step back, probably by not treating all juniors as new players each year (as Neill suggests). Also (based on the discussion above) it seems that the way of bringing in new players may be flawed, but this has been hidden (until this year) because new players are just 5% to 10% of players in adult leagues (based on my experience as a grader).

[Roger de Coverly]

Perhaps the simplest way to grade new players to the grading system would be to "ignore" grading results, and estimate them instead, not based on statistics, but on their apparent strength of play

This is the way it used to be done of course when local graders had a greater input into the calculations. The problem then was that if someone played in more than one event, they could end up with as many estimates of their strength as there were local graders. With better communication between graders, I'd expect this issue could be overcome without much trouble. The other problem was there was a lack of consistent rules across the country particularly implicit or explicit minimum grades.

[Roger de Coverly]

What I mean by "the junior problem" is the deflationary affect on grades of a a large group of improving players. This was recognised by Richard Clarke who proposed (even if he did not implement it himself) correcting the grades of junior players. However, it's not just the deflationary effect on all grades it's the annoyance of how undergraded juniors generally hoover up the prize money in graded sections of congresses, rapidplays etc (annoyance to all us weakies who play in those sections, I mean).

So what's new? The same issues applied 10,20,30 or 40 years ago as well. It was about 40 years ago that a junior increment was first introduced. In fact the number of juniors entering the adult mainstream is a trickle compared with even 20 years ago. All I can see that's different is that the standard of play at the bottom of the pile (as measured by the grading system) is much lower compared to the top quartile than it ever used to be.

I suspect though that it's the estimation process for new players which is bringing them in at too low a level and they aren't that bad.

From a top quartile perspective, you usually need about a 175 standard of play to score 50% in an Open. That's been the same for 20 or 30 years. Deflation would have dropped it down to 150 and inflation would have pushed it up to 200. Neither has happened. That's not to say that standards of play for a given grade haven't improved over the period. So I think you need to know openings or the middle games that arise from openings that bit better. You may also have to be able to win or draw difficult positions against the clock without the help of an adjudicator.

Edit - perhaps I should say that 175 standard of play is "225-50" in other words 50 points below IM/GM standard. As the titled players improve, the next layer down must improve to stay in the same relative position.

[David Shepherd]

The above is an example of how I think the system is not working based roughly on what Neil has said

eg Start grade 124

Games say 9

5 wins say 4 against grade 8 1 against grade 124

3 losses say against grade 124 and 1 against grade 50 (blunder)

Grade after iterative process ((4*58) + (1 *174) + (3*74) + (1*0))/9 =69

Now 40 point rule used grade (4*79) + (1*159) + (4*59)= 79

The player has lost 45 grading points by playing weaker players, and say winning 1 out of 4 of the games he played against similar strength (and one blunder) - he had limited protection from the 40 point rule.


If last sesons grade was considered the resulting grade would be

((4*134) + (1*174) +(3*74)+ (1*34)) = 107

The change in system has cost this player 28 grading points - what has happened was that his grade was 124 but by playing and winning against the lower graded players he received 79 grading points each and not 134.

Please forgive me if there are errors in the above or it repeats what has been said earlier!

[Kevin Thurlow]

"There seem to be a lot of examples of people previously mentioned who would have more accurate grades if the people who knew them suggested their playing strength. This could be a role of an arbiter at a tournament that an ungraded player enters."

as if there weren't enough things for arbiters to do..... Remember most arbiters are weak players who couldn't make a useful estimate anyway.

Roger de C is right - the old time graders actually knew about the players and could make sensible estimates or seek sensible estimates from clubs. Now it is all centrally done and estimates (if any) are made by people who have no idea of who the player is. Say you have a new player who lost to a 180, a 160 and a 130 then disappeared. How do you estimate his grade? A good player watching the games could say whether he is 100 or 10...

[Alex Holowczak]

Yes, I think that's what I was aiming for, rather than arbiter.

This person who I know had the following results in the most recent tournament (based on the newly published grades this year, for last season):
J L
J L
J L
J L
J L
113 W
112 L
J L
102 D

J = Junior, i.e. people who would probably therefore be considered ungraded. Yet, according to the site, these are the grades of said juniors:

136 L
152 L
U L (no grading history)
156 L
U L (104 rapid, rapid grade this year)
113 W
112 L
102 L
105 D

Any ideas how you would get a grade of 32 with those results? The first 5 losses came in the UKCC Terafinal, and the next 4 matches were played in a tournament 6 months later. So it's obvious that there was an improvement in that time. So if you're giving someone a new grade, by guessing, you could see that the standard is ~ 90 on the new grades now. If you're taking the other results into account, that wouldn't be unreasonable. Given it's a new grade, and they would be careful of over-estimation, 90 is perhaps a bit much to expect. So anything in the region of 75 would make sense. But 32!

[Roger de Coverly]

Any ideas how you would get a grade of 32 with those results?

Well it's 1.5 from 9 and there's an unrated opponent in there. It's just more evidence that the estimation program runs amok when confronted with unrated players and "extreme" results. "extreme" just seems to mean something like 1.5 from 9 or 8.5 from 10.

The London League top 50 guy had 9.5 from 10. This is just the other side of the coin.

The problems of negative and zero grades have been an issue for a few years now. Do we now have enough evidence to say it was the estimation program "wot dun it" ?

[Neill Cooper]

136 L
152 L
U L (no grading history)
156 L
U L (104 rapid, rapid grade this year)
113 W
112 L
102 L
105 D

Any ideas how you would get a grade of 32 with those results?

Looks like too many loses. (ps what about round 6 of the Terafinal?)

What you really need is the actual printout, because as I found the grades used are not those published!

The calculation will have ended up something like:
L vs 136 scores 22 (32 -10 as that is less than 136 - 50)
L vs 152 scores 22 (32 -10 as that is less than 152 - 50)
W vs 113 scores 122 (32 + 90, as that is less than 113 + 50)
D vs 105 scores 72 (32 + 40, as that is less than 105)

[I've ignored junior correction)

Once a new player is more than 40 points lower than all opponents (or 40 points greater than all opponents) then the process keeps on looping (but Michael White says it will converge eventually).

[Neill Cooper]

Well it's 1.5 from 9 and there's an unrated opponent in there. It's just more evidence that the estimation program runs amok when confronted with unrated players and "extreme" results. "extreme" just seems to mean something like 1.5 from 9 or 8.5 from 10.

MOST of them were unrated (they were juniors).