Minimum Grades

General discussions about ratings.
Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: Minimum Grades

Post by Roger de Coverly » Thu Nov 01, 2012 6:56 pm

Paul Dupré wrote:I bet 20 or 30 people thought that was a good idea. It's laughable that anyone would come up with this idea, let alone vote for it.
I don't think anyone was given the chance to vote for it. If I recall the sequence of events, the idea of treating all juniors as if they were new players was produced like a rabbit from a hat in August 2009 when the new grades were already overdue. No one ever knew quite why, but earlier versions of the iteration produced some really wild results for players scoring close to 100%, or even 0%. For adult players, the new grades seemed mostly in the right order and by adding 20 points to the average player looked quite flattering.

Paul Dupré wrote: Live grades are essential in any well run Federation. With monthly as a minimum. Yes, minimum. After each event is preferable.
You need to refine your thinking as to what you mean by a "live grade". Chess Scotland works out a "what your rating would be if we published a new list", but don't actually use it for anything. As a consequence of only updating the grades used for calculation once a year, they have the same issues as the ECF with potentially obsolete junior ratings corrupting the entire system. FIDE by contrast have monthly updates and the monthly updated rating is used in the next calculation period. They have to have careful rules about which list applies to which event and cutoffs for when data can be processed. Like Scotland, they also show pre-publication, the effect on the next rating. Unlike the ECF system, which attempts to grade every game, Elo based systems will exclude games against new players. This removes one of the complexities of publishing frequent lists.

Paul Dupré
Posts: 331
Joined: Fri Jul 11, 2008 11:04 pm
Location: Sutton, Surrey

Re: Minimum Grades

Post by Paul Dupré » Thu Nov 01, 2012 7:09 pm

Roger de Coverly wrote: You need to refine your thinking as to what you mean by a "live grade". Chess Scotland works out a "what your rating would be if we published a new list", but don't actually use it for anything. As a consequence of only updating the grades used for calculation once a year, they have the same issues as the ECF with potentially obsolete junior ratings corrupting the entire system. FIDE by contrast have monthly updates and the monthly updated rating is used in the next calculation period. They have to have careful rules about which list applies to which event and cutoffs for when data can be processed. Like Scotland, they also show pre-publication, the effect on the next rating. Unlike the ECF system, which attempts to grade every game, Elo based systems will exclude games against new players. This removes one of the complexities of publishing frequent lists.
Oh, I've already got an answer to that.

Players should have every game graded. However, 'New Players' and when I say new players, I don't mean somebody who has lost their grade, I mean completely new, should NOT have a published grade until they have played at least 16 games, like in Wales. As, soon as 16 games have been completed they can get a published grade. Also, I like the way FIDE show next period Individual Calculations November 2012. Oh, it changed from when I last looked!!! Ah, yes New publication today! How annoying.

So, what I'm looking for is the best of all worlds...Why can't I have it...
Any postings on here represent the truth, and nothing but the truth, so help me God,
...and by the way the world is flat.

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

Re: Minimum Grades

Post by Roger de Coverly » Thu Nov 01, 2012 7:19 pm

Paul Dupré wrote: Players should have every game graded. However, 'New Players' and when I say new players, I don't mean somebody who has lost their grade, I mean completely new, should NOT have a published grade until they have played at least 16 games, like in Wales.

Rating can be a double sided process. If you don't give a new player a rating until they've played n games, then you aren't able to give their opponents credit for the result either. There's an underlying principle of Elo style systems, that games rated v un-rated don't count for the rated player except in a few specific examples. The Clarke systems as used by the ECF and BCF before it, attempt to grade every game by every player, which means you need an estimation or initial grade setting process for players who don't have grades you can look up.

Andrew Bak
Posts: 835
Joined: Sun Jan 02, 2011 11:48 am
Location: Bradford
Contact:

Re: Minimum Grades

Post by Andrew Bak » Thu Nov 01, 2012 7:49 pm

Paul Dupré wrote:Players should have every game graded. However, 'New Players' and when I say new players, I don't mean somebody who has lost their grade, I mean completely new, should NOT have a published grade until they have played at least 16 games, like in Wales. As, soon as 16 games have been completed they can get a published grade.
Why 16 games? Why not 10, or 30? Is this an arbitrary choice or is there some underlying mathematical reason behind it?

Paul Dupré
Posts: 331
Joined: Fri Jul 11, 2008 11:04 pm
Location: Sutton, Surrey

Re: Minimum Grades

Post by Paul Dupré » Thu Nov 01, 2012 8:32 pm

Roger de Coverly wrote:Rating can be a double sided process. If you don't give a new player a rating until they've played n games, then you aren't able to give their opponents credit for the result either. There's an underlying principle of Elo style systems, that games rated v un-rated don't count for the rated player except in a few specific examples. The Clarke systems as used by the ECF and BCF before it, attempt to grade every game by every player, which means you need an estimation or initial grade setting process for players who don't have grades you can look up.
OK, now we are just talking semantics. Games against un-graded players can be shown as awaiting verification OR whatever you like. The point is, every situation can be covered. Saying you won't start something until every problem is ironed out is a very defeatist attitude.

Basically, I know it can be done. So, the only reason I can see is laziness on behalf of the ECF. Like every government, they like spending money on "Worthwhile Projects", instead of required projects.

No need to re-invent the wheel.
Any postings on here represent the truth, and nothing but the truth, so help me God,
...and by the way the world is flat.

Paul Dupré
Posts: 331
Joined: Fri Jul 11, 2008 11:04 pm
Location: Sutton, Surrey

Re: Minimum Grades

Post by Paul Dupré » Thu Nov 01, 2012 8:33 pm

Andrew Bak wrote:Why 16 games? Why not 10, or 30? Is this an arbitrary choice or is there some underlying mathematical reason behind it?
Yes, it's the number of white pieces on a chess board....OR black pieces if you're that way inclined.
Any postings on here represent the truth, and nothing but the truth, so help me God,
...and by the way the world is flat.

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

Re: Minimum Grades

Post by Roger de Coverly » Thu Nov 01, 2012 8:39 pm

Paul Dupré wrote: OK, now we are just talking semantics. Games against un-graded players can be shown as awaiting verification OR whatever you like.
Practicalities surely. Elo based systems don't have any problem with discarding results against players without a rating. If you want to include such games in a recalculation for a rated player, then you need to know what to include. Online servers get around this problem by giving all new players an arbitrary rating with K factors that minimise the effect on the rated players. But that means you don't have rated and unrated but rated and provisionally rated. If you want near real time ratings, you cannot hold up the processing for a new player to play another 15 games.

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

Re: Minimum Grades

Post by Alex Holowczak » Thu Nov 01, 2012 8:42 pm

Andrew Bak wrote:Why 16 games? Why not 10, or 30? Is this an arbitrary choice or is there some underlying mathematical reason behind it?
30 was chosen because there is a 95% confidence in the grade being correct. Further explanation can be found here: http://en.wikipedia.org/wiki/Confidence_interval

10 is another confidence interval, but I can't remember off-hand exactly what it is. It was reduced to 9 because there was enough demand for it to be reduced to 9. The various grading categories correspond with other confidence intervals, but I can't remember what they are.

The maths behind this is sound in theory: Because you calculate a grade by adding stuff up and dividing it by a number, i.e. you find the mean, the more games you play, the more accurate your grade is.

For example:

If I have 900 points over 9 games, I have a grade of 900/9 = 100. If I then beat a 100 in the last game in the period, my grade goes up to 1050/10 = 105

If I have 9000 points over 90 games, I have a grade of 9000/90 = 100. If I then beat a 100 in the last game in the period, my grade goes up to 9150/91 = 101, assuming you round up to the nearest whole number.

The result of one game makes less difference to your grade, so the figure of 100 (or 101) is far more accurate, so you can say statistically that you have more confidence in its accuracy.

Paul Dupré
Posts: 331
Joined: Fri Jul 11, 2008 11:04 pm
Location: Sutton, Surrey

Re: Minimum Grades

Post by Paul Dupré » Thu Nov 01, 2012 8:46 pm

Roger de Coverly wrote: If you want near real time ratings, you cannot hold up the processing for a new player to play another 15 games.
I agree, but I don't care what solution you come up with, whether it's a continuous Clarke Grading System or an Elo Rating System (preferable). I just want action - and I don't see it.

The iteration idea was a backward step NOT a forward step.
Any postings on here represent the truth, and nothing but the truth, so help me God,
...and by the way the world is flat.

Paul Dupré
Posts: 331
Joined: Fri Jul 11, 2008 11:04 pm
Location: Sutton, Surrey

Re: Minimum Grades

Post by Paul Dupré » Thu Nov 01, 2012 8:51 pm

Alex Holowczak wrote:30 was chosen because there is a 95% confidence in the grade being correct. Further explanation can be found here: http://en.wikipedia.org/wiki/Confidence_interval

10 is another confidence interval, but I can't remember off-hand exactly what it is. It was reduced to 9 because there was enough demand for it to be reduced to 9. The various grading categories correspond with other confidence intervals, but I can't remember what they are.

The maths behind this is sound in theory: Because you calculate a grade by adding stuff up and dividing it by a number, i.e. you find the mean, the more games you play, the more accurate your grade is.

For example:

If I have 900 points over 9 games, I have a grade of 900/9 = 100. If I then beat a 100 in the last game in the period, my grade goes up to 1050/10 = 105

If I have 9000 points over 90 games, I have a grade of 9000/90 = 100. If I then beat a 100 in the last game in the period, my grade goes up to 9150/91 = 101, assuming you round up to the nearest whole number.

The result of one game makes less difference to your grade, so the figure of 100 (or 101) is far more accurate, so you can say statistically that you have more confidence in its accuracy.
Stop trying to big up how the "successful ECF grading system" works. It's obvious that the latest "improvement" was a MASSIVE disaster.
Any postings on here represent the truth, and nothing but the truth, so help me God,
...and by the way the world is flat.

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

Re: Minimum Grades

Post by Roger de Coverly » Thu Nov 01, 2012 9:00 pm

Alex Holowczak wrote: The various grading categories correspond with other confidence intervals, but I can't remember what they are.
The criterion for publication has been weakened over time in the interests of having some sort of number, whether reliable or not for as many players as possible.

The preface to the 1988 list for instance says:-
The minimum condition for having an official BCF Grade has been reduced to 15 games in three years, including 5 in the most recent.

Estimated Grades are shown for players who do not have a BCF Grade but have played 10 or more games in the last three years.
Earlier there had been a minimum standard of 30 games over two years with at least 20 in the most recent season, which is the current "B" condition. I don't recall when the "C" condition of 30 games in three seasons with 10 in the most recent was implemented, but it appears that the D and E categories date back to 1988. Graders would always have had access to a master list with everyone on it, but I don't know the extent to which they would ignore unpublished grades and treat players with changed standards of play as new.

Andrew Bak
Posts: 835
Joined: Sun Jan 02, 2011 11:48 am
Location: Bradford
Contact:

Re: Minimum Grades

Post by Andrew Bak » Thu Nov 01, 2012 10:40 pm

Alex Holowczak wrote:
Andrew Bak wrote:Why 16 games? Why not 10, or 30? Is this an arbitrary choice or is there some underlying mathematical reason behind it?
30 was chosen because there is a 95% confidence in the grade being correct. Further explanation can be found here: http://en.wikipedia.org/wiki/Confidence_interval

10 is another confidence interval, but I can't remember off-hand exactly what it is. It was reduced to 9 because there was enough demand for it to be reduced to 9. The various grading categories correspond with other confidence intervals, but I can't remember what they are.

The maths behind this is sound in theory: Because you calculate a grade by adding stuff up and dividing it by a number, i.e. you find the mean, the more games you play, the more accurate your grade is.

For example:

If I have 900 points over 9 games, I have a grade of 900/9 = 100. If I then beat a 100 in the last game in the period, my grade goes up to 1050/10 = 105

If I have 9000 points over 90 games, I have a grade of 9000/90 = 100. If I then beat a 100 in the last game in the period, my grade goes up to 9150/91 = 101, assuming you round up to the nearest whole number.

The result of one game makes less difference to your grade, so the figure of 100 (or 101) is far more accurate, so you can say statistically that you have more confidence in its accuracy.
Thanks for the link to the "confidence interval" article - I'd never heard of this concept before, whilst asssuming that such a thing must exist in statistics somehow!

I just thought 16 was an odd number (even though it's even :roll:) - it seems that this was more or less an arbitrary choice.
Paul Dupré wrote:Stop trying to big up how the "successful ECF grading system" works. It's obvious that the latest "improvement" was a MASSIVE disaster.
A massive disaster in the sense that it provides a reliable rating system to estimate the strength of players who play games under the auspices of the ECF?

Paul Dupré
Posts: 331
Joined: Fri Jul 11, 2008 11:04 pm
Location: Sutton, Surrey

Re: Minimum Grades

Post by Paul Dupré » Thu Nov 01, 2012 11:34 pm

Andrew Bak wrote:
Paul Dupré wrote:Stop trying to big up how the "successful ECF grading system" works. It's obvious that the latest "improvement" was a MASSIVE disaster.
A massive disaster in the sense that it provides a reliable rating system to estimate the strength of players who play games under the auspices of the ECF?
What are you chatting about?

Really, you're about 3 days behind... catch up now, before you say anything else dopey.
Any postings on here represent the truth, and nothing but the truth, so help me God,
...and by the way the world is flat.

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

Re: Minimum Grades

Post by Roger de Coverly » Thu Nov 01, 2012 11:55 pm

Andrew Bak wrote: A massive disaster in the sense that it provides a reliable rating system to estimate the strength of players who play games under the auspices of the ECF?
Paul's point is that he considers it a desirable feature of a rating or grading system that it should do something so that players can see the effect of their recent results on their relative standing or ranking. It would have been a challenge to do this with the ECF Clarke based system because of the issue of new players. This was compounded by the decision to treat under eighteens as new players regardless.

How does the Yorkshire system handle the "new players" problem?

Paul Dupré
Posts: 331
Joined: Fri Jul 11, 2008 11:04 pm
Location: Sutton, Surrey

Re: Minimum Grades

Post by Paul Dupré » Fri Nov 02, 2012 1:09 am

Roger de Coverly wrote: Practicalities surely. Elo based systems don't have any problem with discarding results against players without a rating. If you want to include such games in a recalculation for a rated player, then you need to know what to include. Online servers get around this problem by giving all new players an arbitrary rating with K factors that minimise the effect on the rated players. But that means you don't have rated and unrated but rated and provisionally rated. If you want near real time ratings, you cannot hold up the processing for a new player to play another 15 games.
OK, I prefer Elo ratings if there's an option. But, as I said before there is no reason why the ECF can't do what they want with there own system. So, put all "new players" into the system, just without any official rating against the match until the new player is rated. And, retrospectively rate that game, like they do now.... After all, there is no need to hold up monthly calculations for the sake of the odd game. No need to discard games, just rate them later.

I'll tell you a major problem with the current ECF grading system: Foreign players with FIDE ratings. Take for example my win against a spanish player rated 2227. New Player to England maybe, but not new player. ECF conversion 200+, anyway I should get 250+ grading points. But, because I beat him he withdraws and is never seen again. So, I get 174 + 50 points. Unbelievable, I lost 30 points because he lost. How is that right?

Effectively, I got 224 points for beating a 200+ player. What annoyed me the most was that my next grade was only based on 32 games, 181 (my highest ever) should have been 182.
Any postings on here represent the truth, and nothing but the truth, so help me God,
...and by the way the world is flat.

Post Reply