New grades (split from Editorial thread)

[TomChivers]

What is the average first grade of juniors after one year of playing?

If this number is used as the estimated grade for all incoming newcomers - instead of 40 - then it should solve the problem.

[Sean Hewitt]

Matt / Matthew / Tom,

Have you examined the evidence from the ECF grading database, done mathematical calculations to quantify the level of deflation - taking the results to produce a proper solution to the problem?

The reason I ask is that I have, Howard has, and so have others. Independently of each other And we have all come to essenitally the same conclusion.

Plucking solutions out of thin air (such as giving all new players a start grade of x, whatever x may be) is just nonsensical. Where's the evidence to support the suggestion?? Do you really expect to give a foreign GM, new to this country a grade of 40, 80 or whatever you happen to fancy today? Its farcical. Not every ungraded player is a beginner, and the fact that the ECF calculates a reasonably accurate estimate for such players is one of the positives of the system. You need to understand that new players ARE NOT THE PROBLEM. Rapidly improving players, WHO ALREADY HAVE A GRADE are. I've demonstrated by example in a previous post why this is the case.

There are two solutions that spring immediately to my mind to ensure that this problem does not re-surface in future years after the grading fix is implemented.

1) First, and preferable, is to publish grades monthly. This means that every month our improving players go up in grade by a little bit, and so playing them later in the year would yield a more appropriate grading return than the current system does. However, I am sure that for logistical and other reasons this cant / wont happen.

2) The other soluton is, if any player shows a grade increase of more than 10 in the season, discard his existing grade and substitute it with the newly calculated one. You then plug this new grade back into the grading calculation and recalculate the seasons grades. This is, in my opinion, the practicable solution.

If the junior in my earlier example was treated in the second manner (instead of his actual 125 grade) all three players would come out with accurate 150 gradings and there would be no deflation.

Points lost to the system are totally IRRELEVANT, because the player is lost to the system also.

Assume we have 100 players in our grading system, and the average grade is 110. How many points are in the system? You dont know, but it doesnt matter!! If Mr Average leaves the system, we now have 99 players, we have less points in the system, but we still have an average grade of 110. But I hear you cry, what if one of these weak players left. Well, if Little Jonny (graded 50) left we of course lose points to the system - but the average grade goes UP to 111!! But that too is irrelevant. The other 99 players still have an accurate grade and will continue to play against each other and generate accurate grades amongst themselves.

Michaels opinion that the growth of weekend congresses in the 1970's led to the growth in junior chess is an equally plausible explanation for the initial cause. It doesnt really matter. The point is that a system that is only re-calculated once per year cannot cope with players who are improviong by large amounts in that period. And when you have 32% of your players in the database as juniors, thats a recipe for grading disaster!

[E Michael White]

Mathew Turner : Changes in the mean grade are not an accurate measure of inflation/deflation which has to be assessed by other methods. Deflation/inflation arises also from the pattern of results. It is possible for a player to increase his grade and cause more apparant inflation than deflation.

If the process followed for newstarters is as described by Richard Hadrell in http://forum.bcfservices.org.uk/viewtopic.php?f=4&t=140 on 22 Nov 07 in reply to mine of 21 Nov, then we should have concerns. Incidentally 59 people simultaneously looked at this forum at that time probably to see Richards reply; this was the record viewing total exceeding recent items about the ECF directors and possibly shows where players main interest lies ie grading.

This iterative process is more likely to be deflationary than inflationary. In addition leaving the grades of graded players constant from one iteration to the next is going to increase the inflation or deflation, because to prevent grading entropy gain/loss those grades would need to float down/up to the calculated values from one iteration to the next.

For an example of how points can be lost or gained see the last two paragraphs of my posting of 21 Nov 2007 in http://forum.bcfservices.org.uk/viewtopic.php?f=4&t=140. The new starter problem is a special case of this.

I guess normally highly active players in most sports and games will be more likely to overperform than underperform so with the ECF averaging formula we can expect deflation. This does not mean the ECF formula is inappropriate for League chess but does mean that regular checks need to be carried out every 2 years or so on the deflation/inflation levels.

[TomChivers]

Hi Sean - I didn't mean to cause you offence. I thought what Matthew wrote was accurate and went from there; obviously I now know that it was not. Personally I support this change and trust the expertise that informs it.

[Matthew Turner]

With all due respect Sean, points lost to the system are important. If a players goes from 40 to 100 and then leaves, the points he has gained will have been taken from other players remaining in the system. I am suggesing that new players (U18) would be given a nominal grade of 80 to initiate grading calculations. Foreign players do not have to given a grade of 80 they can simply have a grade calculated from there national/FIDE rating using the formula FIDE = ECF*8 + 600. I am not claiming that this is as mathematically rigorous as the current system, I am just saying it works. It is effectively how the system worked twenty years ago when deflation was not a problem.

[Matt Harrison]

Sean,

No I haven't looked at all the data, but I have read the background material (including your analysis), and I think that you've convincingly demonstrated that deflation is present in the system and the proposed fixes will address this, taking the graded population as a whole. But if as Howard says, most people's grades will change (assuming the new = old x 0.8+ 50 is a reasonable approximation), then there are going to be a lot of quite strong players who have had a reasonably constant grade for the last n years who will look sceptically at a fix that happens to increase their grade by 10-15 points. My concern is that the vocal opposition from this group over the next 12 months might torpedo the exercise.

I therefore think it's interesting to consider where, how and why the grading deflation has come about. A lot of this is likely to be conjecture and hypothesis, but then could this be tested? You've asserted before that low starting grades aren't the problem - I'd be interested in an explanation of why this is the case. Is there an argument for treating new juniors differently from new adults?

[Matt Harrison]

Sean. Your proposal on how to treat rapidly improving players seems to make a lot of sense. The only downside is that all those people who played that player during the year will have assumed they were at grade x, when they were actually grade x+y, meaning they will get a small boost to their grade beyond that they expected.

This seems a small problem, but I don't know how much negative feedback graders currently get following publication of the new list, or how much new feedback this would generate.

Wouldn't this have to be another iterative process, as these improving players (especially juniors playing lots of congresses) could have played each other during the year.

[Dr Andrew Cula]

Why can't you, as you process the yearly results, calculate internally a grade after every game?

[Sean Hewitt]

Hi Sean - I didn't mean to cause you offence. I thought what Matthew wrote was accurate and went from there; obviously I now know that it was not. Personally I support this change and trust the expertise that informs it.

Absolutely no offence taken. And I apologise if I was a bit brusque! The chess community is very good at taking two halves of a story, adding them together and ending up with exactly the wrong end of the stick!! I happy to justify and prove (where possible) my views on this particular subject, but sometimes it is exasperating when so many have "views" of what is going on - and are happy to forcibly express them - when they have none of the insight that those running the show (and thats not me by the way) have. However, given some of the other topics re the ECF recently, maybe thats just a trait that as chess players we all have!!

Oh, and it made me half an hour setting off for the South Lakes congress. But I'm here now so all is well!

[Sean Hewitt]

Why can't you, as you process the yearly results, calculate internally a grade after every game?

In an ideal world you could and you would. My suggestion of monthly grading lists is essentially the same idea.

But in the world we live in, events are graded by volunteer graders. For the above to work they would have to submit the grading files within say 7 days of the event. Thats a long way from where we are now. But I do think we should try to get 2 grading lists per year, then 4.

[John Upham]

Whilst preparing league results databases for export to the Rating Officers a thought occurred...

Some of new junior players have Date of Birth information to go to the ECF and some don't

AFAIK, those with will benefit from the junior rating bonus and those without won't

So, hard luck on those who have lost to improving juniors w/o date of birth information as they will score 10 less for that game

Now, THAT IS RANDOM!

John

[Sean Hewitt]

Whilst preparing league results databases for export to the Rating Officers a thought occurred...

Some of new junior players have Date of Birth information to go to the ECF and some don't

AFAIK, those with will benefit from the junior rating bonus and those without won't

So, hard luck on those who have lost to improving juniors w/o date of birth information as they will score 10 less for that game

Now, THAT IS RANDOM!

John

Then you should sack your gradings officer - its part of his job to get DOB for new players!!

[John Upham]

Sean,
At the weekend I implemented a facility for the Surrey Border League web site to allow captains to upload directly DoBs for any player.

When this happens a confirmatory email is sent to myself and the League Rating Officer.

In the last 24 hours over 30 DoBs have been entered this way.

Here is a sample email message :

Surrey Border Chess League player XXX, (PlayerID = YYY) has had their DoB updated to 1965-05-11 by ZZZZ (CaptainID = X) on 2008-06-09

Names etc witheld to protect the guilty.

No need for sackings, we just need to work together as a team.

John

[TomChivers]

I blogged this story on Tuesday (link.)

Aside from RdC who also posts here, there didn't seem to be any particularly passionate reactions in the comments or from outside links (eg.)

I think if the ECF stage-manages this correctly (explains it clearly and simply, with a technical report also available to download) then I see little opposition from regular chess players.

[Sean Hewitt]

Sean. Your proposal on how to treat rapidly improving players seems to make a lot of sense. The only downside is that all those people who played that player during the year will have assumed they were at grade x, when they were actually grade x+y, meaning they will get a small boost to their grade beyond that they expected.

Thats true - but it already happens.

I've just played in the South Lakes Congress (and it was nice to meet Michael White there too btw). All my opponents were graded BUT the congress, played in June, will be included in the 2009 grading list (not the 2008 list). Therefore, I wont actually know the grades of my opponents for the purposes of my own grade until August this year!!