FIDE Delaying/Maybe Abandoning Introducing New K Factors

[James Coleman]

http://www.fide.com/component/content/a ... e-k-factor

[Sean Hewitt]

That's one hell of a U-turn!

[Kevin Thurlow]

"During the 10-day working visit of Deputy President Georgios Makropoulos in Singapore with General Secretary Ignatius Leong...."

And the rest of the agenda was.....?

[Ben Purton]

What else is there to do in Thailand.....


Youd hope along to Cambodia where there is some heavy History.

Ben

[Richard Bates]

Voice of Reason

The proposals are a load of mathematical nonsense. They are based on an assumption that the majority of players in the FIDE list are significantly mis-rated and therefore need a higher K-factor to more rapidly reach their "correct" level (due to the constraining effect of more frequent rating lists). They take no account of the reality of most people's ratings being a pretty accurate reflection of their strength, with the result that all that happens is the size of the margin of error is increased.

[Roger de Coverly]

Stewart Reuben published a useful summary of the (proposed) changes at

http://www.bcf.org.uk/organisation/fide ... _mar09.htm

with the actual (draft) regulations at

http://www.bcf.org.uk/organisation/fide ... _mar09.doc

From the summary

3. The k factors are being changed. K=20 for players 2400 or higher, even if their rating later slips below 2400. K=30 for all other players. K is the multiplier. For example, if a 2400+ player were expected to score 0.38, but won the game, he would gain 0.62x20 = 12.4. For those under 2400 the gain will be 18.6. This is the first time these k values have been changed since Professor Elo introduced the system in 1970. Then there was only one rating list per year and far fewer events. When k=20, a rating changes over in 38 games. When it is 30 the number of games is 25.

Expanding this comment, the 0.38 expectation arises (according to the tables in the regulations) for a difference in ratings of 84-91. So in the example it's a 2400 player against 2491. If the 2400 plays 38 such games and scores 50% (k=20) , his new rating is 2400 + 38 * (0.62-0.5) * 20 = 2491. At k=10, it would take 76 games.

If I've understood the debate on the relationship between k factor and frequency of rating list correctly, let's assume we have a k of 10 and 76 games being played by our 2400 player to reach 2491. If we introduce a new rating list half way through the period, then after 38 games our 2400 player has reached 2400 + 38 * (0.62-0.5) * 10 = 2446. His score expectation against 2491 players is now 0.56 from the 40-46 table. So after another 38 games, his rating is 2446 + 38 * (0.56-0.5) *10 = 2469. He would now need to play 75 games to reach 2491. The calculation is 2446 + 75 * (0.56-0.5) * 10 = 2491.

So that's the argument in favour of the k factor doubling coupled with the publication frequency change that it enables players whose strength has changed to more rapidly reach their true level.

In his note, Stewart also observes

2. The lists will appear every two months. It was agreed in 1999 eventually to have monthly lists. FIDE have been working towards this slowly for fear of overburdening the system.

In a way, FIDE already have "up to date" ratings since the summation on line at the ratings site of the previous quarterly list and the results processed so far gives the current position. The communication benefits of more frequent lists have already been achieved.

It's obvious though that for players whose strength is variable rather than changing, that more frequent lists and higher k factors increase volatility. For example if you play 2 tournaments in a quarter and score 2 above your expectation in the first and 2 below in the second, then on the existing system you will be unchanged. If there is a fresh rating list between the tournaments then you will appear to gain or lose points. For K=30, you gain 60 points and then give them back in the following list.

There's a fear of course that higher k factors make ratings manipulation easier. There's a suggestion that if you are intending to play a lot of games to try and reach a 2400 or 2500 target, that it may help to have a low initial rating so as to score "more" points for every game.

I dare say that those really in the know were aware that the k factor change was possible at Dresden. The rest of us though only became aware of the change after rumours of Dresden decisions started circulating. Acceptance of the changes would have been greater had the issue been exposed for debate well before Dresden. Did the ECF board for example discuss whether they endorsed the change prior to the Olympiad?

Stewart also notes

1. The new rating floor will be 1200. This means that any English player graded above 75 should be able to aspire to a FIDE Rating. Ratings lower than 1400 will only start to appear from the 1 September 2009 list. It was agreed in 1999 to go down to 1000 in due course. FIDE have been working slowly towards this goal for fear of destablising the system.

At some stage FIDE are going to have to worry about junior increments and rapidly improving players. If you have 8 year olds coming in at the bottom at around 1200, is the rating system going to be able to keep up if they reach a 2000 + standard within 5 years? Higher k factors will help there of course.

[Roger de Coverly]

According to chessbase at http://www.chessbase.com/newsdetail.asp?newsid=5527, the doubling of the K factor has been abandoned for the time being. 6 rating lists a year and a "400 point" rule remain.