Six monthly grades

General discussions about ratings.
David Pardoe
Posts: 1225
Joined: Mon Apr 16, 2007 11:29 pm
Location: NORTH WEST

Re: Six monthly grades

Post by David Pardoe » Sun Jul 24, 2011 1:03 am

Paul Cooksey wrote:
Alex Holowczak wrote:
Paul Cooksey wrote:I assume ECF is more volatile than Elo for more frequent lists. Has this been discussed already?
Alex Holowczak wrote:By whom?
I meant on the forum, I was inviting a link to a previous thread if there was one.
Alex Holowczak wrote:It has been discussed by the grading team. They're aware of these issues.
http://ecforum.org.uk/viewtopic.php?f=2 ... 7f6#p66137

PS Players should take a close look at the new gradings...out today...!! Most players grades move by very small amounts, even over a whole season. I`m guessing mine will move by about 8 points (south) this season.
BRING BACK THE BCF

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

Re: Six monthly grades

Post by Roger de Coverly » Sun Aug 21, 2011 11:27 am

There's an update on the proposals at
http://www.sccu.ndo.co.uk/grad.htm

I have reservations about the count back method proposed as this can mean that some events feature in more than one published list, whilst others only appear once. By way of an example, consider a player who plays forty games a season, five a month each month from September to April. On the proposed system, his A grade in the August 2012 list would contain, working backwards Apr 2012, Mar 2012, Feb 2012, Jan 2012, Dec2011, Nov 2011. On the February 2013 list, his grade would contain Dec 2012, Nov 2012, Oct2012, Sep2012, Apr 2012, Mar 2012. So the Apr and Mar 2012 results are counted twice. The Dec and Nov 2012 results are likely to reappear in the August 2013 list.

On average across all players you might expect the effect to be neutral. Presumably though it increases the likelihood that an individual's grade is biased up or biased down, depending on the results towards the end of a grading period.

Paul Cooksey

Re: Six monthly grades

Post by Paul Cooksey » Sun Aug 21, 2011 11:45 am

Roger de Coverly wrote:Presumably though it increases the likelihood that an individual's grade is biased up or biased down, depending on the results towards the end of a grading period.
I agree, it seems very likely we will see more volatility as a result.

I wonder if it would improve the system to vary the number of games used in count back? (A similar concept to k, reducing count back for juniors for example).

John Hodgson
Posts: 73
Joined: Thu Jul 29, 2010 10:13 am

Re: Six monthly grades

Post by John Hodgson » Tue Sep 13, 2011 6:02 pm

Can someone explain, to a non-statistician as myself, the following:

Where a player plays less than 30 games in a grading period (in the new case, six months), why is it a correct methodology to include the last 'x' games from the previous period (to bring the total to 30)rather than the average of the grading results in the previous grading period (again, to bring the total to 30)?

And, in passing, why 30? If more than 30 results are available in a single period the system uses them, but it restricts calculations to 30 results if they fall in multiple periods. Why 30 in one case, but not the other?

Say that I play 27 games in six months (realistic for me, as I play 50-60 games a year). I won my last three games the previous period (but lost the previous three). This increases my new grading, but if I reverse the sequence of results in the previous period this gives a quite different result. How can this be correct? Surely an average is better?

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

Re: Six monthly grades

Post by Roger de Coverly » Wed Sep 14, 2011 12:06 am

John Hodgson wrote:Can someone explain, to a non-statistician as myself, the following:

Where a player plays less than 30 games in a grading period (in the new case, six months), why is it a correct methodology to include the last 'x' games from the previous period (to bring the total to 30)rather than the average of the grading results in the previous grading period (again, to bring the total to 30)?
You may be familiar with the method as described at http://www.sccu.ndo.co.uk/grad.htm
Standard will now fall in line with Rapid in all respects. This will involve a change in the way Standard games are brought forward from earlier periods. If you need n games to make up your 30, you would expect them to be the n most recent games. And indeed they are, in Rapidplay.
This was an approach first proposed around fifteen years ago or more. I'm not aware of any theoretical work which demonstrated that using the most recent 30 gave a better estimate of strength and performance than using, say, the 25 from the most recent period with the last 5 averaged in. As suggested the "most recent 30" creates a bias in favour of results at the end of a rating period.

Ola Winfridsson
Posts: 324
Joined: Fri Aug 14, 2009 12:26 pm

Re: Six monthly grades

Post by Ola Winfridsson » Wed Sep 14, 2011 9:22 am

Won't a performance-based grade (such as the ECF system), rather than a cumulative rating (like the Elo system), nearly always be a far less accurate prediction of future performance, almost regardless of the number of games played in each individual grading/rating period? (Unless you've been relatively inactive for a goodish while, in which case all reliable predictions go out the window anyway?)

User avatar
Greg Breed
Posts: 723
Joined: Thu Apr 05, 2007 8:30 am
Location: Aylesbury, Bucks, UK
Contact:

Re: Six monthly grades

Post by Greg Breed » Sat Apr 28, 2012 12:32 pm

Alex Holowczak wrote:
Paul Cooksey wrote: I can only speak for myself. On hearing this was happening, I was quite pleased. I like the idea of more responsive grading, and hope six monthly is a stepping stone to monthly. I imagine grading is quicker and easier than it has ever been and continues to improve.

But I do have some doubts about whether ECF could handle monthly, if it needs a 30 game period for reasonable accuracy. If it can great, otherwise there is Elo. I'll concede it is not a discussion likely to fascinate the average man in the street. But I though on a chess forum in a thread about grading it might be worth talking about.
That's probably fair enough. I'm just used to several people questioning why we're having it, when the time to do so was 12 months ago...

You're right that the grading system works better when it has more games per player in it. Grades are a mean, so this is fairly intuitive.

You don't need 30 games in the period; you can go back over previous periods to get to 30. You can theoretically have games rolling in and out as months go by if we have monthly grading lists.

Some graders don't publish the dates of games, or if they do they get the wrong dates. I suspect there will be complaints when all the information is published online on July 23/24, assuming this will have the new layout with more information on it provided by Carl. There will be questions along the lines of why their league games don't have colours when their congresses do. The reason of course is that the grader hasn't supplied them, which is possibly because he wasn't supplied that information in the first place.

My personal preference is that we switch to an Elo-based system.
For the first time in it's history, my local league Hillingdon District Chess League (HDCL) has been provided with a website! This is a smashing website used by many other local leagues and is created by John Upham (Many thanks John!). It not only records results, updates league tables and Knock-Out competitions, but provides grading information for the grader and for those viewing the website. I love it.

Why can't something like this be purchased or leased by the ECF and dispensed to all leagues nationally? It is after all in everyone's interest to get reliable and up-to-date data on player performances. This would surely be more convenient and simpler. I for one can attest to the massive difference it has on a grader's job. I previously had to record and compile all results in a spreadsheet at the end of the season. I even had to take time off work to do it and that was just for a small league with 15-20 teams in 3 divisions! Now it is all automated and the Chief graders receive everything in a nice timely manner already to enter into their system.
Hatch End A Captain (Hillingdon League)
Controller (Hillingdon League)

Post Reply