Elo rating issue

General discussions about ratings.
Post Reply
Brian Valentine
Posts: 574
Joined: Fri Apr 03, 2009 1:30 pm

Elo rating issue

Post by Brian Valentine » Fri Feb 26, 2016 10:33 am

I'm hoping that someone with a mathematical back ground and Elo methodology can check out my maths attached.

This comes from an investigation of national rating systems using Elo. There are a number of variations from FIDE for new entrants and improvers. I have been fumbling through some theory which might help sorting out the best in class variations.

If I've got it right then there might be problems if the number of games in a batch is large. Given I was surprised by this, I was hoping to get some confirmation of the result, in case it becomes important.

Brian
Attachments
Elo160226.pdf
(71.28 KiB) Downloaded 181 times

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

Re: Elo rating issue

Post by Roger de Coverly » Fri Feb 26, 2016 11:16 am

Brian Valentine wrote: This comes from an investigation of national rating systems using Elo.
Aren't there two very similar distributions that can be used? The one in your formula looks to be the USCF version, whilst FIDE's one is slightly different. They are very similar in the 0 to 200 range, but diverge a bit at the extremities.

What you are picking up is the "over-correction" effect. That's been known to those seeking a ratings target for many years. What you have to do is to improve above your current rating without playing that much. You then play an intensive number of games in as short a period as possible and hope your results reflect your new strength. Your rating then overshoots, with the effect of briefly reaching your target rating.

Using the entirely natural process of playing a lot of chess in the school summer holidays and less so at other times of the year, Juniors are now able to get a more realistic International rating by exploiting this effect, particularly with K=40.

NickFaulks
Posts: 8453
Joined: Sat Jan 02, 2010 1:28 pm

Re: Elo rating issue

Post by NickFaulks » Fri Feb 26, 2016 12:30 pm

Brian,

You are completely right. For the rapid and blitz lists there is a special adjustment to ensure that "a" cannot exceed 1. It would have been nice to apply this to the main list at the same time, but this will have to wait for a regulation change this year, for implementation in 2017. Meanwhile, I check every rating list for any egregious examples of excessive rating changes. Mercifully, there has so far been only one.
If you want a picture of the future, imagine a QR code stamped on a human face — forever.

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

Re: Elo rating issue

Post by Brian Valentine » Fri Feb 26, 2016 3:16 pm

Nick and Roger,

Thanks for the comments. I now see 12.3 of the FIDE rating regulations should have alerted me. Furthermore the USCF have a restriction, loosely described as 800/games played, and ChessScotland sets K as 800/n.

I shall go away and see if this revelation (to me) matters.

Brian

NickFaulks
Posts: 8453
Joined: Sat Jan 02, 2010 1:28 pm

Re: Elo rating issue

Post by NickFaulks » Fri Feb 26, 2016 4:20 pm

I don't know what purpose 12.3 serves, but people seem to complain when you suggest taking it out.
If you want a picture of the future, imagine a QR code stamped on a human face — forever.

User avatar
IM Jack Rudd
Posts: 4818
Joined: Tue Apr 17, 2007 1:13 am
Location: Bideford
Contact:

Re: Elo rating issue

Post by IM Jack Rudd » Fri Feb 26, 2016 6:49 pm

NickFaulks wrote: You are completely right. For the rapid and blitz lists there is a special adjustment to ensure that "a" cannot exceed 1. It would have been nice to apply this to the main list at the same time, but this will have to wait for a regulation change this year, for implementation in 2017. Meanwhile, I check every rating list for any egregious examples of excessive rating changes. Mercifully, there has so far been only one.
Be careful with this. I can see game-throwing becoming a big thing if you do introduce that adjustment.

NickFaulks
Posts: 8453
Joined: Sat Jan 02, 2010 1:28 pm

Re: Elo rating issue

Post by NickFaulks » Sat Feb 27, 2016 9:07 am

Can you explain that, Jack?
If you want a picture of the future, imagine a QR code stamped on a human face — forever.

Post Reply