FIDE and Glicko

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

FIDE and Glicko

Post by Roger de Coverly » Mon Apr 16, 2012 12:03 pm

A report from Stewart Reuben at the ECF site about potential changes to international rating.
http://www.englishchess.org.uk/?p=18494

Elo v Glicko
http://www.englishchess.org.uk/wp-conte ... Glicko.pdf

To my mind there's a very obvious flaw in Glicko based systems, which would apply if FIDE tried to introduce one.
Glicko features wrote:Rating deviation increases due to inactivity and decreases after game results are submitted.
What that means is that if you were performing at around 180 over a season in domestic events and around 2100 in rated events, that if you played a domestic event every week and a FIDE event every two months, you would be treated as relatively inactive for the FIDE event. So any fluctuations in results would lead to a magnified change in rating. Thus is you play in two consecutive events, with a performance of 2200 in the first and 2000 in the second, under Glicko you get a higher rating than if you perform at 2000 in the first and 2200 in the second.

I would also say that I'm not convinced that rating systems should be designed around supposed predictive ability. I would suggest the lesser aim of ranking players in the correct order. In other words, the number one ranked player is the one you expect to make the best score against a field of players of comparable strength without expecting precision as to how much his score will exceed that of the second ranked player.

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

Re: FIDE and Glicko

Post by Alex Holowczak » Mon Apr 16, 2012 12:36 pm

Roger de Coverly wrote:Thus is you play in two consecutive events, with a performance of 2200 in the first and 2000 in the second, under Glicko you get a higher rating than if you perform at 2000 in the first and 2200 in the second.
In England, we think the juniors are under-rated relative to their ECF grade. If we accept that, and accept that this is because juniors play less FIDE-rated chess than ECF-graded chess, and that they improve their ability more quickly than their rating can catch up, wouldn't such a change to the FIDE system be desirable?

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

Re: FIDE and Glicko

Post by Roger de Coverly » Mon Apr 16, 2012 2:11 pm

Alex Holowczak wrote: If we accept that, and accept that this is because juniors play less FIDE-rated chess than ECF-graded chess, and that they improve their ability more quickly than their rating can catch up, wouldn't such a change to the FIDE system be desirable?
An assumption for juniors that if they haven't played in an event subject to your rating system for a period of time, then their standard of play has changed is probably valid. I believe the Scottish system has a hack whereby if you play 200 Elo points above your previous rating, you are fed back in as if you are a new player.

I don't think it's valid for adults and wrong to assume that just because a player hasn't played in games subject to your rating system, this somehow makes the estimate of strength established by your system less valid. You don't know that the player hasn't been active somewhere else.

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

Re: FIDE and Glicko

Post by Alex Holowczak » Mon Apr 16, 2012 6:07 pm

Roger de Coverly wrote:I don't think it's valid for adults and wrong to assume that just because a player hasn't played in games subject to your rating system, this somehow makes the estimate of strength established by your system less valid. You don't know that the player hasn't been active somewhere else.
If we assume that adults have a fairly constant ability, and don't really improve (to the extent that juniors do) for much of their adult lives, does this make much difference? If their ability remains the same, the likelihood of them playing a tournament with a TPR 200 points higher than their previous tournament is unlikely, and so all that will happen is that the rating fluctuates more wildly than under Elo, but averages out to about the same number over the long-term?

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

Re: FIDE and Glicko

Post by Roger de Coverly » Mon Apr 16, 2012 6:16 pm

Alex Holowczak wrote:If their ability remains the same, the likelihood of them playing a tournament with a TPR 200 points higher than their previous tournament is unlikely, and so all that will happen is that the rating fluctuates more wildly than under Elo, but averages out to about the same number over the long-term?
TPR can easily fluctuate by 25 ECF points from one five round tournament to the next. Every win is worth 10 points, so scoring 2.5 against an average 180 field is 180 whilst 3.5 is 200. Under Glicko the equivalent of the K factor increases with supposed inactivity, so the first tournament after a presumed break has a bigger effect if you out perform or under perform.

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

Re: FIDE and Glicko

Post by Greg Breed » Fri Dec 14, 2012 2:24 pm

So FIDE has recently gone to monthly rating updates. I think this is a step in the right direction. Wouldn't it be good if we had something similar for the ECF? "Yes!" I hear you cry, "but how?" Well we need to have a way of submitting results nationally that is simple and effective. How hard can this be in this day and age?
Hatch End A Captain (Hillingdon League)
Controller (Hillingdon League)

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

Re: FIDE and Glicko

Post by John Upham » Fri Dec 14, 2012 3:06 pm

Greg Breed wrote:So FIDE has recently gone to monthly rating updates. I think this is a step in the right direction. Wouldn't it be good if we had something similar for the ECF? "Yes!" I hear you cry, "but how?" Well we need to have a way of submitting results nationally that is simple and effective. How hard can this be in this day and age?
It's not hard at all.

The ETTA have released a free League Management application for any league that wishes to use it.

The results server knows about all the leagues that submit results to it.

From this a national ranking system is planned.

The software development and management is funded by advertising and sponsorship.

Replace ETTA with ECF in the above.

Of course it will never happen for chess in England as the body stopping progress will see to that: ECF Council.
British Chess News : britishchessnews.com
Twitter: @BritishChess
Facebook: facebook.com/groups/britishchess :D

Brian Valentine
Posts: 574
Joined: Fri Apr 03, 2009 1:30 pm

Re: FIDE and Glicko

Post by Brian Valentine » Fri Dec 14, 2012 4:48 pm

I agree with Roger that the system that optimises result prediction might not be the most suitable. However the first paper shows that some tests have been done on what happens to rankings and the inflation of top players.

I understand that there is a potential problem with apparent inactivity. Nevertheless all the top 5 Kaggle contestants recognised that "activity" was an influence on the results.

While more work needs to be done appraising Sticko, I think the key question is whether players would prefer 3-4% improvement in predictability of results in return for the increased complexity.

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

Re: FIDE and Glicko

Post by Roger de Coverly » Fri Dec 14, 2012 6:25 pm

John Upham wrote: From this a national ranking system is planned.
Planned is the operative word. For all your denigration of the chess world and claims as to how table tennis does it better, it is worth noting that the BCF had a national ranking system from the early 1950s, so that's approaching sixty years ahead of the table tennis world.

David Lettington
Posts: 141
Joined: Fri Apr 06, 2007 10:37 pm

Re: FIDE and Glicko

Post by David Lettington » Sun Dec 16, 2012 7:13 pm

Roger de Coverly wrote: it is worth noting that the BCF had a national ranking system from the early 1950s, so that's approaching sixty years ahead of the table tennis world.
Yes, but that is quite a long time ago, so it's not unreasonable to expect significant developments to have occurred since then.

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

Re: FIDE and Glicko

Post by Roger de Coverly » Sun Dec 16, 2012 7:43 pm

David Lettington wrote:Yes, but that is quite a long time ago, so it's not unreasonable to expect significant developments to have occurred since then.
John Upham seemed to think that monthly grading would be rejected out of hand by the ECF voting membership. I don't think it would, but anyone proposing such a change would have needed to have worked out the finer detail as those with many years experience of grading systems would expect well thought out answers to some of the more difficult decision points. Yorkshire, for example, quote a new grade every time a new result is received, but the calculation base is only updated once a year. FIDE quote a new rating every month with a week or two's lag but don't go back in time (usually) when a late result is received. The ECF quote a new grade every six months and unlike Yorkshire rebase every grade. Also unlike FIDE, results of games played right up to nominal cutoff still count, so there's a waiting period whilst the up to date grade is being calculated.

At least twice in my playing lifetime, well intentioned upgrades to the grading process have come close to wrecking the entire system and at the very least called the integrity of the process into question. In both cases, the failure was not of BCF/ECF Council or Board asking too many questions, but too few.

Post Reply