Grading algorithm - a query

General discussions about ratings.
Post Reply
Paul McKeown
Posts: 3732
Joined: Thu Apr 12, 2007 3:01 pm
Location: Hayes (Middx)

Grading algorithm - a query

Post by Paul McKeown » Fri Jun 01, 2012 12:56 pm

I understand that when a player has played fewer than thirty games in the list to be graded, the current grading algorithm counts back the required number of games in the player's record to make the total exactly thirty and then calculates the new grade from this selection of thirty games. Is this correct? I understand that previously a weighted average was taken with the games from the season in question and from games from one or two complete previous seasons to make the total at least thirty.

Imagine then the following scenario:

Player X, graded 182, say, plays 25 games at 193 and sees that the 5 last games from the previous list are pretty disastrous and drop his new grade to 189. He resolves to play five games in his club championship in the expectation of scoring a perfect five at 192 by virtue of the forty point rule. That would give him a grade of 193 for the next season. He then thinks it through, though. What happens in the following list if he again fails to play thirty games? The five games are the most recent and will be dragged into the grading calculation for the following list. He imagines a number of possibilities:

a) the 5 games graded at 192 by virtue of the forty point rule are regraded at 202. Bit strange.
b) the 5 games remain graded at 192. Does the system "remember" the original forty point calculation? Again a bit strange.
c) the system forgets the forty point rule and regrades the games at the average of fifty points above the mean grade for the five opponents. Pretty disastrous.

Does anyone know the actual answer?

Alan Walton
Posts: 1393
Joined: Mon Nov 02, 2009 8:33 pm
Location: Oldham

Re: Grading algorithm - a query

Post by Alan Walton » Fri Jun 01, 2012 1:07 pm

I have always thought you calculate it as a weighted average

25 games from the current year

40 games the previous year, but take a 5 game average from these games

That's the way it should be, but I have a sneaky suspicion that they would take the last five dated games from the previous season

Hopefully somebody can clarify my view as well

Paul McKeown
Posts: 3732
Joined: Thu Apr 12, 2007 3:01 pm
Location: Hayes (Middx)

Re: Grading algorithm - a query

Post by Paul McKeown » Fri Jun 01, 2012 1:09 pm

Alan, that's exactly how I think it used to work, but I think the algorithm was changed. I may be wrong.

User avatar
John Upham
Posts: 7162
Joined: Wed Apr 04, 2007 10:29 am
Location: Cove, Hampshire, England.
Contact:

Re: Grading algorithm - a query

Post by John Upham » Fri Jun 01, 2012 1:30 pm

Some time ago I asked HG for details of the algorithm but was met by silence.

Maybe it is wished to keep it secret from its customers?
British Chess News : britishchessnews.com
Twitter: @BritishChess
Facebook: facebook.com/groups/britishchess :D

Mick Norris
Posts: 10310
Joined: Tue Apr 17, 2007 10:12 am
Location: Bolton, Greater Manchester

Re: Grading algorithm - a query

Post by Mick Norris » Fri Jun 01, 2012 1:45 pm

"The grade is calculated by dividing the total number of points scored by the number of games played. If there are 30 or more games in the most recent halfyear, then the grade is based on these games alone. If there are not, it is based on the most recent 30 games played; or on all the games played in the last 36 months if that is less. In no case does calculation go back more than 36 months.

How is “most recent 30” interpreted if a number of games share a date and only some of them are required to make up the 30? Answer: the system takes as many as are required, at the average score for the date."

from http://www.ecfgrading.org.uk/help/
Any postings on here represent my personal views

Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: Grading algorithm - a query

Post by Roger de Coverly » Fri Jun 01, 2012 1:46 pm

Paul McKeown wrote: b) the 5 games remain graded at 192. Does the system "remember" the original forty point calculation? Again a bit strange.
Does anyone know the actual answer?
Without knowing the exact answer, I would expect that the performance of 5 at 192 is remembered. You look it up, rather than revisit it. So it doesn't matter if your last five games are five wins against 152 or below or draws against a 192 field.

Six monthly grading creates anomalies. I'm expecting to re-establish a personal title of "190s player/ used to be a 190s player", last seen in 1976-77. Had annual grading continued my second half 195 ish would have balanced with the first half 175 to leave my grade unchanged.

User avatar
Christopher Kreuzer
Posts: 8781
Joined: Fri Aug 06, 2010 2:34 am
Location: London

Re: Grading algorithm - a query

Post by Christopher Kreuzer » Fri Jun 01, 2012 1:48 pm

Is this theoretical, Paul? Anyone can see that your current published grade is 182 and your last five games from the previous list are here:

http://www.ecfgrading.org.uk/?ref=18601 ... 1222827417

My advice would be to play games and not worry about grades. Though in my case that is pot-kettle-black.

I have some sympathy with those oscillating about boundaries for grade-limited congresses or competitions, but other than that, I think worrying about grades is a bit self-defeating. Though I'd, ahem, like to know the algorithm as well...

Mike Gunn
Posts: 1014
Joined: Wed Apr 11, 2007 4:45 pm

Re: Grading algorithm - a query

Post by Mike Gunn » Fri Jun 01, 2012 1:51 pm

The sad fact is that not many people take much interest in the details of how the grading system works. A couple of years ago I tried to put a motion on the council agenda questioning the new method of calculating junior grades and I couldn't find another organisation to second it. I can't recall a single occasion in the last 2 years when either the board or council discussed grading (apart from the single occasion when I raised it).

I believe what is said above is correct - where someone plays less than 30 games in a grading period, the method is to count back exactly 30 games. I would agree that this is probably less accurate than the old method.

Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: Grading algorithm - a query

Post by Roger de Coverly » Fri Jun 01, 2012 2:01 pm

Mike Gunn wrote: I can't recall a single occasion in the last 2 years when either the board or council discussed grading (apart from the single occasion when I raised it).
http://www.sccu.ndo.co.uk/1011/bcf.htm

Half yearly grading was discussed and approved at the October 2010 AGM. There was a thread on the forum where we tried to establish how it was expected to work, as various people present at the Council meeting had made inconsistent claims as to what had actually been agreed. The main point eventually established is that because of half yearly grades, the July 2012 grades will not be the same as if annual calculations had continued. We don't really know the likely effect as no parallel run for 2011 was ever published.

E Michael White
Posts: 1420
Joined: Fri Jun 01, 2007 6:31 pm

Re: Grading algorithm - a query

Post by E Michael White » Fri Jun 01, 2012 2:34 pm

Mick Norris wrote:In no case does calculation go back more than 36 months
Not wishing to be gradingwise pedantic but where the 40 point rule applies on games over the boundary the calculation takes in grades from more than 3 years ago ie the 192 Paul refers to depended on grades in previous years. This is the source of many anomalies.

Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: Grading algorithm - a query

Post by Roger de Coverly » Fri Jun 01, 2012 3:00 pm

E Michael White wrote:Not wishing to be gradingwise pedantic but where the 40 point rule applies on games over the boundary the calculation takes in grades from more than 3 years ago ie the 192 Paul refers to depended on grades in previous years. This is the source of many anomalies.
Grades or results? The earliest result that can directly appear in July 2012 would be a game played in June 2009. The grade of the opposing player might depend on results going as far back as June 2006. I suppose the grading team must know, or can establish the frequency of the various hits. I'm not sure I've every seen an analysis that out of several hundred thousand games, how many would feature A v A as opposed to E v E. I'd expect more A v A just because they play more games even if the E class outnumber them.

E Michael White
Posts: 1420
Joined: Fri Jun 01, 2007 6:31 pm

Re: Grading algorithm - a query

Post by E Michael White » Fri Jun 01, 2012 3:29 pm

If 2 players have identical results over the last 3 years against the same opponents, the 40 point rule can make their latest grades differ. How much difference depends on their results played 4 or more years ago and the grades of their opponents in those games.

Post Reply