Shouldn't the ECF membership and grading season be the same?

General discussions about ratings.
User avatar
Robert Jurjevic
Posts: 207
Joined: Wed May 16, 2007 1:31 pm
Location: Surrey

Re: Shouldn't the ECF membership and grading season be the s

Post by Robert Jurjevic » Wed Sep 26, 2012 7:00 pm

Roger de Coverly wrote:Any estimates you use for pairings are purely internal and have no effect on the published grades. It's been that way for ten years or so.
Ten years (very good :) )!
Robert Jurjevic
Vafra

Neill Cooper
Posts: 1298
Joined: Sun Feb 24, 2008 4:43 pm
Location: Cumbria
Contact:

Re: Shouldn't the ECF membership and grading season be the s

Post by Neill Cooper » Wed Sep 26, 2012 8:05 pm

ben.graff wrote:If you play more than 60 games a year, isn't the problem that games played in the first half of the season (through to December 31st cut off) somehow count for less? Most team/ season long KOs and other events etc are going to use the July grades. Wouldn't it make more sense for all games played in the year to count towards the July grade? (I'm sure this could be done in a way that would still enable juniors rapid pace of development to be accounted for.)
I agree that all games in the past 12 months should contribute towards your grade published at the end of the 12 months.

User avatar
Robert Jurjevic
Posts: 207
Joined: Wed May 16, 2007 1:31 pm
Location: Surrey

Re: Shouldn't the ECF membership and grading season be the s

Post by Robert Jurjevic » Wed Sep 26, 2012 8:38 pm

Neill Cooper wrote:
ben.graff wrote:If you play more than 60 games a year, isn't the problem that games played in the first half of the season (through to December 31st cut off) somehow count for less? Most team/ season long KOs and other events etc are going to use the July grades. Wouldn't it make more sense for all games played in the year to count towards the July grade? (I'm sure this could be done in a way that would still enable juniors rapid pace of development to be accounted for.)
I agree that all games in the past 12 months should contribute towards your grade published at the end of the 12 months.
I don't see why would this be the case (as now the grading period is 6 months)? This would be the same as to ask (when the grading period was 12 months) that all games in the past 24 months contribute towards one's grade published at the end of the 24 months (even if in the last 12 months one has played 30 games or more). My understanding is that ECF corrects grades taking into account all of the games from a grading period which is taken to be such that enough games have been played in it in order to regard the actual performance results statistically significant. Therefore if enough games have been played in 6 months (current grading period) only those games should be taken into account (I think).
Robert Jurjevic
Vafra

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

Re: Shouldn't the ECF membership and grading season be the s

Post by Roger de Coverly » Wed Sep 26, 2012 10:08 pm

Robert Jurjevic wrote:Therefore if enough games have been played in 6 months (current grading period) only those games should be taken into account (I think).
That is how the ECF is now handling it. The natural consequence is increased volatility. So if you have a poor first half season and a good to outstanding second half, then on the old year-based rules, that would just show up as an average season whereas now your grade might drop by ten points only to recover by twenty.

User avatar
David Shepherd
Posts: 912
Joined: Fri Nov 23, 2007 3:46 pm

Re: Shouldn't the ECF membership and grading season be the s

Post by David Shepherd » Wed Sep 26, 2012 11:00 pm

Statistically the more data that is included the higher the chance of the grades being more "realistic". Using old data however introduces more lag in grades and more chance the grade is out of date and not reflective of current playing strength. Like Neill my instinct is that the last 12 months of games should be included.

Using data from the last 12 months is not the same as using data from the last 24 months (although both could be two grading periods).

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

Re: Shouldn't the ECF membership and grading season be the s

Post by Roger de Coverly » Thu Sep 27, 2012 12:16 am

David Shepherd wrote: Using old data however introduces more lag in grades and more chance the grade is out of date and not reflective of current playing strength.
There's a core fudge at the heart of the Clarke grading method. What it does is measure performance over a period. That's fine and interesting. The fudge is that it then uses that performance as the "estimate of strength" to be used in the next period. At the possible risk of confusing everyone, there are really two measures. The first is the estimate of strength, which arguably Elo based systems may do better at provided they don't hit too much lag. The second is the measure of performance such that if two players face identical fields they are tied on their grade. Clarke systems deliver this where Elo systems don't.

EM White wrote about transformation mathematics recently without detailing what he meant. An observation that can be made about both rating and grading systems is this: - that if on average players meet a field of, on average, the same strength as themselves and if, on average, they score 50%, then their grade or rating will remain unchanged. So how good or how bad the grades or ratings are as predictors and how spread they are, doesn't directly change the grading or rating process in the central case. In practice, there are improving and worsening players and those who score much better or worse than 50%.

Kevin Thurlow
Posts: 5803
Joined: Wed Apr 30, 2008 12:28 pm

Re: Shouldn't the ECF membership and grading season be the s

Post by Kevin Thurlow » Thu Sep 27, 2012 8:15 am

"Mike Gunn grades all of the Surrey internal stuff wearing his hat as Grading Officer for Surrey. As a result, Surrey send in all the internal games played by clubs within their territory, and the county is assigned halfresults counting from their games. "

Not true - Redhill now send their own in after some results were not sent in by Surrey and after SCCA charged us twice for the same grading one year!

I'm not sure why Robert is getting worried about people playing twice. At Redhill, you could theoretically play someone 3 times in the championship/handicap and twice in the knockout (if a replay were needed). You could also play them twice in the RP event. It doesn't seem to cause a problem.
"Kevin was the arbiter and was very patient. " Nick Grey

Alex Holowczak
Posts: 9085
Joined: Sat May 30, 2009 5:18 pm
Location: Oldbury, Worcestershire
Contact:

Re: Shouldn't the ECF membership and grading season be the s

Post by Alex Holowczak » Thu Sep 27, 2012 9:10 am

Kevin Thurlow wrote:"Mike Gunn grades all of the Surrey internal stuff wearing his hat as Grading Officer for Surrey. As a result, Surrey send in all the internal games played by clubs within their territory, and the county is assigned halfresults counting from their games. "

Not true - Redhill now send their own in after some results were not sent in by Surrey and after SCCA charged us twice for the same grading one year!
OK, but in general, Surrey do this with most clubs in Surrey.

User avatar
Robert Jurjevic
Posts: 207
Joined: Wed May 16, 2007 1:31 pm
Location: Surrey

Re: Shouldn't the ECF membership and grading season be the s

Post by Robert Jurjevic » Thu Sep 27, 2012 11:38 am

Roger de Coverly wrote:
David Shepherd wrote: Using old data however introduces more lag in grades and more chance the grade is out of date and not reflective of current playing strength.
There's a core fudge at the heart of the Clarke grading method. What it does is measure performance over a period. That's fine and interesting. The fudge is that it then uses that performance as the "estimate of strength" to be used in the next period. At the possible risk of confusing everyone, there are really two measures. The first is the estimate of strength, which arguably Elo based systems may do better at provided they don't hit too much lag. The second is the measure of performance such that if two players face identical fields they are tied on their grade. Clarke systems deliver this where Elo systems don't.
In my opinion measuring a performance over a period of time may not be such a bad idea. The reason is that one needs statistically significant data of measurement and one cannot do that in no time. The longer the period (and the more active the players are) the more statistically significant the measured data will be.

FIDE IMHO has a genuine problem in assessing for how much to change ratings based on results which are not statistically significant (9 tournament games or a single game).

If you know that you have statistically significant measurement data (which I think Élo estimated to be around 30 games or more) the situation ought to be simple. You correct the grades so that the new grades match the performance in the period from which the statistically significant data were obtained! Obviously you cannot do that for statistically insignificant measurement. It would be wrong IMHO to claim that my new grade is 277 because I drew a single game against Luke McShane (my grade would have been increased for 277-121=156 grading points). OK you could have said let us increase my grade for 156/30=5.2 grading points (which would perhaps be fairer as we assume a nonexistent grading period in which I played 30 games), but still we could have increased my grade for 156/60=2.6 grading points (which would be equivalent of me playing 60 games in the grading period).

Let us assume that I have played 60 games in a grading period (one of which is a draw against Luke McShane, then that game would have gained me 2.6 grading points). If somebody was calculating my grade grading after every game (giving me 1/60th of that what would have been the grade change based only on a single game insignificant statistical measurement) then the result would be obviously different than if I was graded as ECF does now (not changing my grade in between the games). It is true that as time passes my chess abilities may change but one cannot obtain a statistically significant measurement in no time. If I count 30 games for the sake of statistical significance of the measured data then my order of play should not matter and if I drew against Luke McShane in the first or the lat game should make no difference. Of course if my chess ability changes during a grading period we would obtain only an average estimate of my chess ability in that period, but FIDE even does not know how long is the grading period (or to be more precise there is no one) nor how many games have each player played in it (it is unclear, at least to me, if the FIDE rating correction is spot on as is should have been, FIDE may well so to speak stretch or shrink the ratings).

The best one can do IMHO is I guess to make as short as possible a grading period still to get a statistically significant measurement in it (say if on average players play 60 games per year the grading period could be 6 months).

Please allow me to take this opportunity and draw you attention to this (if nothing it somewhat clarifies what could be meant by Élo attribute)...
http://www.ecforum.org.uk/viewtopic.php ... 94#p101494
Last edited by Robert Jurjevic on Thu Sep 27, 2012 8:16 pm, edited 1 time in total.
Robert Jurjevic
Vafra

Alex Holowczak
Posts: 9085
Joined: Sat May 30, 2009 5:18 pm
Location: Oldbury, Worcestershire
Contact:

Re: Shouldn't the ECF membership and grading season be the s

Post by Alex Holowczak » Thu Sep 27, 2012 1:55 pm

Robert Jurjevic wrote:The best one can do IMHO is I guess to make as short as possible a grading period still to get a statistically significant measurement in it (say if on average players play 60 games per year the grading period could be 6 months).
The mean number of games played per year is, apparently, somewhere in the region of 10. The variance is huge, ranging from 1 to 200+. This is why you count back up to 3 years to get your 30 games; if the mean is 10 games per year, you should have played 30 in that period.

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

Re: Shouldn't the ECF membership and grading season be the s

Post by Mike Gunn » Thu Sep 27, 2012 4:40 pm

Just to add one small detail to the above.

The ECF now demands a date for each game and in the few remaining number of cases where we don't have them (Ashtead, one other club and the Surrey Individual Tournament) I have to use my best endeavours to work out what they are. I'm currently don't have the records with me, but I have the feeling that for Ashtead all games in a six month period went in as if they were played on a single date. In the future we really do need to record the date for each game because of the way the ECF now calculates grades, particularly those who play less than 30 graded games in 6 months (i.e. most players).

User avatar
Robert Jurjevic
Posts: 207
Joined: Wed May 16, 2007 1:31 pm
Location: Surrey

Re: Shouldn't the ECF membership and grading season be the s

Post by Robert Jurjevic » Thu Sep 27, 2012 4:56 pm

Mike Gunn wrote:The ECF now demands a date for each game and in the few remaining number of cases where we don't have them (Ashtead, one other club and the Surrey Individual Tournament) I have to use my best endeavours to work out what they are. I'm currently don't have the records with me, but I have the feeling that for Ashtead all games in a six month period went in as if they were played on a single date. In the future we really do need to record the date for each game because of the way the ECF now calculates grades, particularly those who play less than 30 graded games in 6 months (i.e. most players).
Thanks for that info Mike. I will keep my internal records of the Ashtead game dates (may not be 100% accurate but should be pretty close). Neil Hayward (the author of the Tournament Director program for Windows) has told me that he intend to include option of recording individual game dates in his program. Neil told me that he is not sure if the ECF 'low level programming file' format has changed to accommodate for the individual game dates. I hope he will contact somebody in ECF and sort the things out. It would be nice if the Tournament Director's 'Submit for ECF grading' feature would work with dates for individual games (providing the file format supports that). http://www.tournamentdirector.co.uk/
Robert Jurjevic
Vafra

Neill Cooper
Posts: 1298
Joined: Sun Feb 24, 2008 4:43 pm
Location: Cumbria
Contact:

Re: Shouldn't the ECF membership and grading season be the s

Post by Neill Cooper » Thu Sep 27, 2012 5:06 pm

Robert Jurjevic wrote: Therefore if enough games have been played in 6 months (current grading period) only those games should be taken into account (I think).
But many grade limited competitions, including the ECF County and national club championships, only use the July grade to determine eligibility. So I think the July figure should represent all your games over the previous season as being the best estimate of your playing ability over the previous 12 months.

Sean Hewitt
Posts: 2193
Joined: Sat Mar 10, 2012 8:18 pm
Contact:

Re: Shouldn't the ECF membership and grading season be the s

Post by Sean Hewitt » Thu Sep 27, 2012 6:01 pm

Robert Jurjevic wrote:Neil told me that he is not sure if the ECF 'low level programming file' format has changed to accommodate for the individual game dates. I hope he will contact somebody in ECF and sort the things out.
For as long as I've been grading games (since 2004) the ECF grading file format has facilitated dating each game individually. Good to know that this software is so up to date!

User avatar
Robert Jurjevic
Posts: 207
Joined: Wed May 16, 2007 1:31 pm
Location: Surrey

Re: Shouldn't the ECF membership and grading season be the s

Post by Robert Jurjevic » Thu Sep 27, 2012 6:17 pm

Sean Hewitt wrote:
Robert Jurjevic wrote:Neil told me that he is not sure if the ECF 'low level programming file' format has changed to accommodate for the individual game dates. I hope he will contact somebody in ECF and sort the things out.
For as long as I've been grading games (since 2004) the ECF grading file format has facilitated dating each game individually. Good to know that this software is so up to date!
May I ask if there is a spec on the format of the file? Thanks.
Robert Jurjevic
Vafra

Post Reply