6 monthly grades

General discussions about ratings.
Steve Rooney
Posts: 427
Joined: Tue Jan 13, 2009 2:36 pm
Location: Church Stretton

Re: 6 monthly grades

Post by Steve Rooney » Wed Jan 26, 2011 11:21 am

Richard Bates wrote:
Sean Hewitt wrote:
Steve Rooney wrote:And I'd second that, it can't make sense to fiddle around further with the ECF grading system when there is a perfectly good international standard that we could change to. (But I seem to recall reading about issued with some league time controls?)
The issue around league time controls means that FIDE won't rate such events. But, that does not prevent the ECF from swiching it's grading system form the Clarke system to an ELO based methodology. It just means that players will have two ratings as now ; one from FIDE and and one from the ECF, but the ECF one will also be a four digit ELO number.
I doubt replacing the ECF system with an ELO system is simply a matter of 'borrowing' FIDE's methodology, without any extra research. The effects of the FIDE system on a player pool stretching down to near beginner level is still developing and unknown. It is by no means certain that FIDE's K-factors would be at all appropriate to England where people in general play far more games. And it certainly wouldn't stop people complaining about the "under-rated juniors" problem!
Accepting that there is more to be done that just lifting the ELO calculations, it still seems to make more sense to investigate that route than fiddle with our 'little England' system any more.

Mick Norris
Posts: 10329
Joined: Tue Apr 17, 2007 10:12 am
Location: Bolton, Greater Manchester

Re: 6 monthly grades

Post by Mick Norris » Wed Jan 26, 2011 11:31 am

In the US they have a 4 figure number, which I think is normally higher than ELO

Is it more confusing to be, say, ECF 175 and FIDE 2000 than to be, say, ECF 2050 and FIDE 2000?

If you are going to change the methodology, I think it is less confusing to move to a 4 figure system

At present, trying to compare 175 for a year 2010 grade with 175 in, say, year 2006 confuses some people (and some of them run tournaments)
Any postings on here represent my personal views

Richard Bates
Posts: 3338
Joined: Fri Nov 14, 2008 8:27 pm

Re: 6 monthly grades

Post by Richard Bates » Wed Jan 26, 2011 11:35 am

Steve Rooney wrote:Thanks for the clarifications, guys. I will try to pay more attention in future! Having two ratings based on the same (or very similar methodology) still seems preferable. What players are interested in is equivalence which is much easier to assess if both are 4-number ELO grades.
Well only partly, in the sense that it is marginally easier to calculate FIDE = ECF+50, than ECF = ECF*8 +650 or whatever. Equivalence is only ever a guesstimate based on observation anyway. Just because the calculation methods are similar it doesn't follow that the two ratings can be compared directly.

If all you are concerned about is that both have 4 digits then there is nothing about the ECF system that says it has to have 3 digits - you could have the same system replacing +50,0,-50 with +400,0,-400. I would argue that it is actually better having the ECF grade 'look' different since it arguably avoids confusion (for English players - clearly for foreigners it is pretty difficult to work out what strength an ECF grade 'indicates'). Assuming the calculation method remains distinct, of course.
Last edited by Richard Bates on Wed Jan 26, 2011 11:38 am, edited 1 time in total.

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

Re: 6 monthly grades

Post by Alex Holowczak » Wed Jan 26, 2011 11:36 am

Richard Bates wrote:
Roger de Coverly wrote:
The international system is well-established and you could take its rules and establish a domestic English equivalent. There's a caveat though that with the extension of the ratings down to 1000, as to whether the K=15/K=30 structure can react quickly enough to improving players where the improving players don't play that many rated games.
And if you set a K-factor too high whilst maintaining 6 monthly lists, you risk "improving players" becoming massively over-rated by playing too many games (although you could introduce a hybrid whereby rating changes are limited in some way by performance over the rating period).
If you copy FIDE's way of doing it, k = 25 for the first 30 games, and k = 15 otherwise, until you get to 2400. I would have thought having juniors at k = 25 until they reach 18 would be a reasonable way around it. (The precise value of k is up for discussion, but I think most would agree on a higher k-rate for juniors.)

Richard Bates
Posts: 3338
Joined: Fri Nov 14, 2008 8:27 pm

Re: 6 monthly grades

Post by Richard Bates » Wed Jan 26, 2011 11:47 am

Alex Holowczak wrote:
Richard Bates wrote:
And if you set a K-factor too high whilst maintaining 6 monthly lists, you risk "improving players" becoming massively over-rated by playing too many games (although you could introduce a hybrid whereby rating changes are limited in some way by performance over the rating period).
If you copy FIDE's way of doing it, k = 25 for the first 30 games, and k = 15 otherwise, until you get to 2400. I would have thought having juniors at k = 25 until they reach 18 would be a reasonable way around it. (The precise value of k is up for discussion, but I think most would agree on a higher k-rate for juniors.)
The issue with improving, as opposed to stable, players is that their grade/rating will be dependent more on the number of games they play than their strength. This can obviously create problems if the differences created are magnified by relatively long rating periods. If two players are rated, say, 1700 but they both perform at 1900 over the rating period, if one plays twice as many games than the other then they will gain double the pts, to the value of 0.25K per game. If one plays 20 games and one plays 40 then the latter will be 150pts higher at K factor of 25 (and will be 'overrated' by 100pts!). That is quite a lot IMO. This of course doesn't happen under the ECF system.

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: 6 monthly grades

Post by Roger de Coverly » Wed Jan 26, 2011 11:48 am

Richard Bates wrote: And if you set a K-factor too high whilst maintaining 6 monthly lists, you risk "improving players" becoming massively over-rated by playing too many games (although you could introduce a hybrid whereby rating changes are limited in some way by performance over the rating period).
If you go through the pain barrier of converting to a domestic Elo, then you should go the extra mile to monthly, bi-monthly or quarterly calculations as well. Mostly it's a problem of the logistics of reporting results in a sufficiently standard, complete and timely form. None of this should be difficult, but getting all arbiters either to abandon pairing cards or at the very least transcribe pairings and results to a program might be hard work. Have all league secretaries abandoned requirements to be sent postcards yet?

Richard Bates
Posts: 3338
Joined: Fri Nov 14, 2008 8:27 pm

Re: 6 monthly grades

Post by Richard Bates » Wed Jan 26, 2011 11:54 am

Roger de Coverly wrote:
Richard Bates wrote: And if you set a K-factor too high whilst maintaining 6 monthly lists, you risk "improving players" becoming massively over-rated by playing too many games (although you could introduce a hybrid whereby rating changes are limited in some way by performance over the rating period).
If you go through the pain barrier of converting to a domestic Elo, then you should go the extra mile to monthly, bi-monthly or quarterly calculations as well. Mostly it's a problem of the logistics of reporting results in a sufficiently standard, complete and timely form. None of this should be difficult, but getting all arbiters either to abandon pairing cards or at the very least transcribe pairings and results to a program might be hard work. Have all league secretaries abandoned requirements to be sent postcards yet?
True, but i'm just making the point that changing to 6 monthly lists isn't really an "opportunity" to be converting to an ELO system. Maybe at best it's an opportunity to move towards improving and automating reporting procedures to the extent that a move towards an ELO system with far more regular lists could be contemplated in future.

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

Re: 6 monthly grades

Post by Alex Holowczak » Wed Jan 26, 2011 12:56 pm

Richard Bates wrote:The issue with improving, as opposed to stable, players is that their grade/rating will be dependent more on the number of games they play than their strength. This can obviously create problems if the differences created are magnified by relatively long rating periods. If two players are rated, say, 1700 but they both perform at 1900 over the rating period, if one plays twice as many games than the other then they will gain double the pts, to the value of 0.25K per game. If one plays 20 games and one plays 40 then the latter will be 150pts higher at K factor of 25 (and will be 'overrated' by 100pts!). That is quite a lot IMO. This of course doesn't happen under the ECF system.
Some Internet correspondence servers with Elo have a Glicko system. I don't understand the ins and outs of the Maths, but the jist is that k varies with activity. If you play more often, k is a smaller number, and vice versa. Assuming some research was done into how k should vary with activity, that could be perhaps one way of doing it. For example, your June 2010 rating could be 2000 with k = 50. Then in the next period, you play your games at k = 50. Your rating might drop to 1900 in January 11, but with k = 20. So you'd effectively have 2 values; rating and k-factor for each person.

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

Re: 6 monthly grades

Post by Alex Holowczak » Wed Jan 26, 2011 12:59 pm

Roger de Coverly wrote:Have all league secretaries abandoned requirements to be sent postcards yet?
The Birmingham League have had e-cards (you download the card and fill it in, before e-mailing the card to the secretary) for years. The postcard option remains, however. I think I'm right in saying most people do it online. Some secretaries don't have computers, and some just find it easier to invest in stamps.

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: 6 monthly grades

Post by Roger de Coverly » Wed Jan 26, 2011 1:35 pm

Alex Holowczak wrote: Some Internet correspondence servers with Elo have a Glicko system. I don't understand the ins and outs of the Maths, but the jist is that k varies with activity. If you play more often, k is a smaller number, and vice versa.
Glicko systems make an assumption which is complete nonsense for national rating systems. They make the arrogant assertion that the only chess anyone plays is measured in their rating system. Consider the counter-example of an extremely active non-English player just playing the one event , say London Chess Classic, Hastings or Gibraltar. Their form in that event would have a disproportionate effect on their rating.

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

Re: 6 monthly grades

Post by Alex Holowczak » Wed Jan 26, 2011 2:50 pm

Roger de Coverly wrote:
Alex Holowczak wrote: Some Internet correspondence servers with Elo have a Glicko system. I don't understand the ins and outs of the Maths, but the jist is that k varies with activity. If you play more often, k is a smaller number, and vice versa.
Glicko systems make an assumption which is complete nonsense for national rating systems. They make the arrogant assertion that the only chess anyone plays is measured in their rating system. Consider the counter-example of an extremely active non-English player just playing the one event , say London Chess Classic, Hastings or Gibraltar. Their form in that event would have a disproportionate effect on their rating.
I'm not sure that's right.

The problem there is caused by something else. In the international list, no one goes straight in at the top; you always go in lower down. By contrast, the national list could get international players to go in at the top.

If we're only talking about rapidly improving juniors, I think you could safely assume that they weren't at the top, but were indeed at the bottom.

Under the current system, form plays a big part if you play very few games. E.g. Jonathan Mestel's grade went up 26 points last year, but he only played 9 games in the last three years. I would put this down to form, and not playing enough games. The players at Gibraltar often get very high ECF grades. While what you describe will happen, I'm not convinced it's a problem caused just because it's Glicko.

I'm sure as a way around it, k could vary on rating as well as activity. If a strong international player only plays in Gibraltar, and they play in no other event where their ECF grade takes priority over their FIDE-rating, then does it matter? The average league or congress - which is who the grades are really designed for - will remain completely unaffected.

Sean Hewitt

Re: 6 monthly grades

Post by Sean Hewitt » Wed Jan 26, 2011 3:06 pm

Richard Bates wrote: If two players are rated, say, 1700 but they both perform at 1900 over the rating period, if one plays twice as many games than the other then they will gain double the pts, to the value of 0.25K per game. If one plays 20 games and one plays 40 then the latter will be 150pts higher at K factor of 25 (and will be 'overrated' by 100pts!). That is quite a lot IMO. This of course doesn't happen under the ECF system.
Actually Richard, this does happen in the ECF system. In the scenario you describe the first player will have his grade 'watered down' by his previous years grade whereas the second player will have a grade based solely on this years results.

That does not invalidate the point you make, about ELO. And the the longer the rating periods the more accute the problem is.
Roger de Coverly wrote: If you go through the pain barrier of converting to a domestic Elo, then you should go the extra mile to monthly, bi-monthly or quarterly calculations as well. Mostly it's a problem of the logistics of reporting results in a sufficiently standard, complete and timely form. None of this should be difficult, but getting all arbiters either to abandon pairing cards or at the very least transcribe pairings and results to a program might be hard work.
Absolutely right.

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

Re: 6 monthly grades

Post by Alex Holowczak » Wed Jan 26, 2011 3:25 pm

Sean Hewitt wrote:
Roger de Coverly wrote: If you go through the pain barrier of converting to a domestic Elo, then you should go the extra mile to monthly, bi-monthly or quarterly calculations as well. Mostly it's a problem of the logistics of reporting results in a sufficiently standard, complete and timely form. None of this should be difficult, but getting all arbiters either to abandon pairing cards or at the very least transcribe pairings and results to a program might be hard work.
Absolutely right.
I missed this earlier. Since I've been on the control team in Birmingham based events, i.e. the rapidplay and Warwickshire Championship,I've brought my laptop and used Tournament Director. We still do everything with pairing cards, but we have it on the computer explicitly for grading. The main barrier to doing that before is that the Chief Arbiter uses Ubuntu, and none of the software capable of producing a grading file runs on any Linux-based system. I have a laptop with Windows though, so that problem was negated. Since using TD, the grader of those events estimated that it reduced his time commitment by about a week. Now all he does is check the last round results; the rest are checked during the play (unless it's rapidplay, when there's no time, so the checking is done afterwards).

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: 6 monthly grades

Post by Roger de Coverly » Wed Jan 26, 2011 5:29 pm

Alex Holowczak wrote:I'm not sure that's right.

The problem there is caused by something else. In the international list, no one goes straight in at the top; you always go in lower down. By contrast, the national list could get international players to go in at the top.
Not really my point. I think it's only the Australians who use Glicko for their national rating system. They have problems ( or rather their players have problems) because if you become inactive, you can gain or lose 300 points in your come-back tournament if you have a poor (or really good) result. So you are a lifetime 2000 player, make a temporary retirement and comeback and find you are 1700. Good for sandbagging in rating restricted tournaments perhaps but doesn't inspire much faith in the notion that the rating system is supposed to rank players by strength. You might not even have been inactive - just not playing in Australia.

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

Re: 6 monthly grades

Post by Alex Holowczak » Wed Jan 26, 2011 5:40 pm

Roger de Coverly wrote:
Alex Holowczak wrote:I'm not sure that's right.

The problem there is caused by something else. In the international list, no one goes straight in at the top; you always go in lower down. By contrast, the national list could get international players to go in at the top.
Not really my point. I think it's only the Australians who use Glicko for their national rating system. They have problems ( or rather their players have problems) because if you become inactive, you can gain or lose 300 points in your come-back tournament if you have a poor (or really good) result. So you are a lifetime 2000 player, make a temporary retirement and comeback and find you are 1700. Good for sandbagging in rating restricted tournaments perhaps but doesn't inspire much faith in the notion that the rating system is supposed to rank players by strength. You might not even have been inactive - just not playing in Australia.
In which case, you can tweak the k-factor accordingly. Arguably k is too high if your rating can change by 300 points in, say, a 5-round weekender. If you had it such that:
A grade: k = 15
B grade: k = 20
C grade: k = 25
D grade: k = 30
E grade: k = 35
Otherwise: k = 40
Then you might negate the problem. I don't know if k = 40 is too high, but if so, find out the maximum practical k, and allocate a k to each class such that they're roughly equally distributed.