Page 2 of 3

Re: FIDE rating calculations

Posted: Fri Jun 17, 2011 7:18 am
by Paul Dargan
I thought that FIDE recalculated ratings for players every 30 games if they played more than this number within the same rating period?? exactly to try and prevent this kind of effect?

Or did this stop when they moved to rating each game and more frequent lists?

Re: FIDE rating calculations

Posted: Fri Jun 17, 2011 8:27 am
by Roger de Coverly
Paul Dargan wrote:I thought that FIDE recalculated ratings for players every 30 games if they played more than this number within the same rating period?? exactly to try and prevent this kind of effect?
National rating systems may do this, but never FIDE. The international nature of the process and issues such as Norm qualification make it imperative that ratings are always well defined. In other words, there is always a rule which enables a tournament or league organiser to determine a rating.
Paul Dargan wrote:Or did this stop when they moved to rating each game and more frequent lists?
Game by game rating didn't change very much. They used to add up the score for the tournament, work out your average opposition and then look up how much you were expected to score against that field. Now they work out what you are supposed to score against each opponent. It works slightly better where you face players 400 points away from you, since you can then truncate the expected score to the result for +/- 400.

Re: FIDE rating calculations

Posted: Mon Jun 20, 2011 11:15 am
by John Upham
My understanding of these calculations is poor.

For example, at the recent e2e4 Gatwick Major

A 1789 player achieves a TPR of 2719 with the following results :

He draws with the Bye
Beats a 1989
Looses to a 1489
Looses to a 1933
draws with someone with a rating of 0

Keep doing this over a few majors and...

So, clearly my understanding is badly wrong. Can someone help please?

Re: FIDE rating calculations

Posted: Mon Jun 20, 2011 11:27 am
by Roger de Coverly
John Upham wrote:My understanding of these calculations is poor.

For example, at the recent e2e4 Gatwick Major

A 1789 player achieves a TPR of 2719 with the following results :

He draws with the Bye
Beats a 1989
Looses to a 1489
Looses to a 1933
draws with someone with a rating of 0

Keep doing this over a few majors and...

So, clearly my understanding is badly wrong. Can someone help please?
You need to check whether all these ratings were FIDE. For pairing purposes, they would have converted ECF grades to a FIDE equivalent. These were shown with a * on the pairing lists. The TPR of 2719 is consistent with the only rated result being a win against a player of 1989.

FIDE's implementation of TPRs gives extremely high results for 100% scores. An alternative method, the 400 rule, would give a TPR of 2389. The basic problem being that any player between 2389 and 2719 has a near 100% expectation of winning against a player of 1989, so you cannot infer much about the strength of the winner.

Re: FIDE rating calculations

Posted: Mon Jun 20, 2011 12:21 pm
by Sean Hewitt
Thanks Roger. Spot on!

Re: FIDE rating calculations

Posted: Tue Jun 28, 2011 10:29 pm
by Jonathan Bryant
Christopher Kreuzer wrote:...I'm actually slightly disillusioned with the way the FIDE rating system works ... there are too many under-rated players around for the system to work properly.

...

When I first talked to various people about the FIDE rating system (this was some 10 years ago, when it was much harder to get an initial rating), they all said that the initial rating you get is critical.

I've now played 8 games against elo rated opponents (4 at e2e4 Sunningdale and 4 at e2e4 Gatwick) so I think I now need just one more game to get an official elo rating. I think I'd be quite unlucky to go through the whole of the Benasque Open next month without playing at least one person with an elo rating so I'm guessing I'm going to get that first elo rating pretty soon.

I must admit I'm not sure I understand the process that FIDE use to work out initial ratings.

If you take me games from Sunningdale I scored 3.5/4 against relatively weaker opposition and apparently that gave me a TPR of 1936. At Gatwick I scored 1.5/4 against relatively stronger opposition for a TPR of 2007. Quite a difference.

The funny thing is, if you convert the elos to ECF and work out TPR based on the 'add 50 for a win then take the average' system, my TPRs work out to 194 and 168. That is, FIDE system says Gatwick was much the better tournament for me but the ECF system says that my better tournament by quite a margin was Sunningdale.

Under the ECF system my overall 'grade' for the 8 games would be 181 or so. That's an elo equivalent of around 2095. I'm guessing my *actual* elo for these games will be nothing like that. As much as 100-150 points less in fact.


The ECF might well be rather generous for me for Sunningdale but the elo system does seem rather stingy. [EDIT: It's interesting that the two systems work out about the same for Gatwick but for Sunningdale, when I one three games against (slightly) lower-rated opposition than myself, the elo system is considerably less impressed with the result than the ecf system). If there really are a lot of under-rated elo folk knocking around then this sort of thing might be one of the reasons why.


I've also heard the thing about your first rating being critical. Whether it's actually true or not I don't know. Perhaps it's received wisdom? I suppose if I keep playing rated games then even if I am under-rated at first my actual elo will catch up sooner or later for the reasons outlined further up the thread.

Re: FIDE rating calculations

Posted: Tue Jun 28, 2011 10:41 pm
by Alex Holowczak
Jonathan Bryant wrote:I must admit I'm not sure I understand the process that FIDE use to work out initial ratings.

If you take me games from Sunningdale I scored 3.5/4 against relatively weaker opposition and apparently that gave me a TPR of 1936. At Gatwick I scored 1.5/4 against relatively stronger opposition for a TPR of 2007. Quite a difference.
For your 3.5/4, your rating would be: Average Opponent Rating + 1.5*25. 1.5 is the number of points above the 50% score you are. 25 probably is a random number. Hence your rating for that event is average opponent rating + 37.5, rounded to the nearest integer.

For your 1.5/4, you scored less than 50%. 37.5%, to be precise. By looking it up in a look-up table, scoring 0.375 (rounded up to 0.38) means you get your average opponents' rating minus 87. See the table on this page.

So this is where your part ratings come from.

Re: FIDE rating calculations

Posted: Tue Jun 28, 2011 10:55 pm
by Jonathan Bryant
Alex Holowczak wrote:You've scored 3.5/4 v 1898 and 1.5/4 v 2094. In total, that would be 5/8 v 1996. If you managed that in one tournament, you'd have a rating of 2021 over your 8 games. The way you've done it, you have part ratings of 1936 and 2007 at 4 games each, which would work out to be 1972 over 8 games. So your rating is 49 points lower than it would have been had you scored 5/8 in one tournament, rather than do it in two separate tournaments.
Thanks for that. Is there a logical reason that playing 8 games in one tournament should be considered 'better' than 4 games each in two tournaments? I mean, it's not like I had a choice. e2e4 events are good but I think even Sean would struggle to pair me against elo rated opposition eight times in a seven round event.

Re: FIDE rating calculations

Posted: Tue Jun 28, 2011 11:00 pm
by IM Jack Rudd
Alex Holowczak wrote: You've scored 3.5/4 v 1898 and 1.5/4 v 2094. In total, that would be 5/8 v 1996. If you managed that in one tournament, you'd have a rating of 2021 over your 8 games. The way you've done it, you have part ratings of 1936 and 2007 at 4 games each, which would work out to be 1972 over 8 games. So your rating is 49 points lower than it would have been had you scored 5/8 in one tournament, rather than do it in two separate tournaments.
Note to everyone reading this: this is no longer the way the system works. The way it now works does give Jonathan an effective 2021 over 8 games.

Re: FIDE rating calculations

Posted: Tue Jun 28, 2011 11:04 pm
by Sean Hewitt
Alex Holowczak wrote:
Jonathan Bryant wrote:I must admit I'm not sure I understand the process that FIDE use to work out initial ratings.

If you take me games from Sunningdale I scored 3.5/4 against relatively weaker opposition and apparently that gave me a TPR of 1936. At Gatwick I scored 1.5/4 against relatively stronger opposition for a TPR of 2007. Quite a difference.
For your 3.5/4, your rating would be: Average Opponent Rating + 1.5*25. 1.5 is the number of points above the 50% score you are. 25 probably is a random number. Hence your rating for that event is average opponent rating + 37.5, rounded to the nearest integer.

For your 1.5/4, you scored less than 50%. 37.5%, to be precise. By looking it up in a look-up table, scoring 0.375 (rounded up to 0.38) means you get your average opponents' rating minus 87. See the table on this page.

So this is where your part ratings come from.

You've scored 3.5/4 v 1898 and 1.5/4 v 2094. In total, that would be 5/8 v 1996. If you managed that in one tournament, you'd have a rating of 2021 over your 8 games. The way you've done it, you have part ratings of 1936 and 2007 at 4 games each, which would work out to be 1972 over 8 games. So your rating is 49 points lower than it would have been had you scored 5/8 in one tournament, rather than do it in two separate tournaments.
That's not correct Alex. Once a player has his nine games for his first rating, the games are lumped together and considered as if they had been played in one event.

Also, part ratings have to be achieved in units of three games or more. So playing one rated player in your next event is no use, Jonathan needs to play three.

Re: FIDE rating calculations

Posted: Tue Jun 28, 2011 11:11 pm
by Alex Holowczak
Oh, OK. Sorry Jonathan, and everyone else. When was that changed? :?

The reason I hadn't noticed the change: Mine works by doing 2/9 v 1880 = 1660 and (1689*6 + 1602*3)/9 = 1660. :oops:

Re: FIDE rating calculations

Posted: Tue Jun 28, 2011 11:43 pm
by IM Jack Rudd
The two calculations are unlikely to give significantly different answers if both partials were above 50% or both partials were below 50%. It's where one was above and the other was below that it really matters.

Re: FIDE rating calculations

Posted: Wed Jun 29, 2011 12:51 am
by Roger de Coverly
Sean Hewitt wrote: Also, part ratings have to be achieved in units of three games or more. So playing one rated player in your next event is no use, Jonathan needs to play three.
Whilst it's very plausible that in a 9 round international at least 5 opponents will be rated, I had this idea that once you have the minimum 1/3 from the first part rating, that every result after that will count. Stewart has hunted down and destroyed most of the incentives to withdraw from tournaments to preserve your rating.

Re: FIDE rating calculations

Posted: Wed Jun 29, 2011 9:42 am
by Jonathan Bryant
IM Jack Rudd wrote:
Alex Holowczak wrote: You've scored 3.5/4 v 1898 and 1.5/4 v 2094. In total, that would be 5/8 v 1996. If you managed that in one tournament, you'd have a rating of 2021 over your 8 games. The way you've done it, you have part ratings of 1936 and 2007 at 4 games each, which would work out to be 1972 over 8 games. So your rating is 49 points lower than it would have been had you scored 5/8 in one tournament, rather than do it in two separate tournaments.
Note to everyone reading this: this is no longer the way the system works. The way it now works does give Jonathan an effective 2021 over 8 games.

That will be the spanky new Arbiter Qualification kicking in Jack. Thanks to you and everbody else for your responses.


{PS: I still think FIDE's evaluation of wins against lower rated opponents is a bit mean spirited}


J

Re: FIDE rating calculations

Posted: Wed Jun 29, 2011 10:30 am
by Alex Holowczak
Jonathan Bryant wrote:{PS: I still think FIDE's evaluation of wins against lower rated opponents is a bit mean spirited}
I think they're trying to guard against someone scoring 100% v a really weak field. If I had 9/9 v 1500, then my rating could be 2300, since they assign +800 for 1.00. Instead, they would do 4.5*25 = 112.5, so my rating would be 1613.

I don't understand why they don't do what they do for scoring below 50% for above 50%. Why not just do what they do for below 50% for above 50%, and just cap the +/- number at 400 points? So if you got 9/9 v 1500, you get 1900. That way you get the benefit for your wins, but you don't get a silly too-high rating.