ECF grades compared with FIDE ratings

[John McKenna]

RdC>Performances in Gibraltar may be worth a look as it features English club players, but not enough of them that they play amongst themselves for at least half the games as would be the case at Hastings...<

Hastings Masters finished earlier this month and began with accelerated pairings.
Gibraltar, which looks much more in need of such a device, has not it seems.

At Gib, 1st round pairings like -

79 205 Hermes Geoff R 1998 0 0 IM Porper Edward 2446 79
(80 80 IM Tania Sachdev 2442 0 0 Reppen Ellisiv 1995 206)
81 207 Spanton Tim R 1995 0 0 IM Munguntuul Batkhuyag 2438 81

aren't going to contribute much to the sum of rating-grading knowledge.

Incidently, in round 2, Porper drew with Short from a pawn down position, while Rapport tried for over 120 moves to beat Sachdev in a rook and knight vs. rook endgame. While the former result may be a tad lucky, by some accounts, for IM Proper the latter may have happened because Richard was so enamoured of Tania that he simply did not want the game to end.

[Christopher Kreuzer]

I wonder if FIDE ratings fluctuate more at lower levels? My rating graph can be seen here:

http://ratings.fide.com/id.phtml?event=470589

Entered the rating list in 2007 at 1946. Peaked at 1977 in November 2012. Lowest ratings have been 1907 and 1908 in September 2009 to May 2010 and September 2013 to January 2014. That is a fluctuation of about 70 rating points over about 7 years, with a total of 120 games (though 205 appear in my games list, I think there is a reason for that). Those games are mostly from three sources: 4NCL, playing in ICCD events abroad, and e2e4 congresses.

I wonder if that sort of grading history is typical. Using the conversion formulas, the 1907 to 1977 range corresponds to about 157 to 169, which sounds about right. Maybe I've been expecting too much...

[MartinCarpenter]

My impression (never checked) was that they don't vary much, simply because of the combination of the conservative update function and only playing a few games.

Hence the performance in the first grade has a tendency to dominate rather and that's obviously far from always representative.

[Sean Hewitt]

My impression (never checked) was that they don't vary much, simply because of the combination of the conservative update function and only playing a few games.

Hence the performance in the first grade has a tendency to dominate rather and that's obviously far from always representative.

IOh how I wish that we true. I'm currently rated 150 points below my peak rating

[Christopher Kreuzer]

Oh how I wish that we true. I'm currently rated 150 points below my peak rating

Still, nice picture...

[Jonathan Bryant]

My impression (never checked) was that they don't vary much, simply because of the combination of the conservative update function and only playing a few games.

Hence the performance in the first grade has a tendency to dominate rather and that's obviously far from always representative.

IOh how I wish that we true. I'm currently rated 150 points below my peak rating

I'm currently 120 points below my peak rating which was my first rating (and about 170 points below my 'conversion rating')

[IM Jack Rudd]

I'm 179 points below my peak.

[MartinCarpenter]

Well you're playing quite a bit of FIDE rated chess

The interesting bit is people just playing 4NCL N/Div 3. So say 6/7 games a year.

Even the ECF system would take a long time to drop the initial grading result then and I'd got the impression that FIDE was a bit more conservative. Fundamentally a fairly impossible problem for a grading system to be accurate with so little data of course.

[Jonathan Bryant]

The interesting bit is people just playing 4NCL N/Div 3. So say 6/7 games a year.

Even the ECF system would take a long time to drop the initial grading result then and I'd got the impression that FIDE was a bit more conservative. Fundamentally a fairly impossible problem for a grading system to be accurate with so little data of course.

I wonder if this would explain the difference in experiences that we've found comparing London tournaments (for me ) and 4NCL for you.

Perhaps we've just been talking about two different things? Clearly, it's a given that no system has any chance of being accurate based on a very small number of games. This is very much not the same problem that I've encountered.

[Christopher Kreuzer]

Perhaps we've just been talking about two different things? Clearly, it's a given that no system has any chance of being accurate based on a very small number of games. This is very much not the same problem that I've encountered.

Are you sure? What number of games over what period is your FIDE rating based on, and how does this compare to your ECF grade. I know the ECF grade is technically only the last 30 games and the system is blind to games before that cut-off point, but how many games on average have gone into each calculation of your ECF grade over the years? For me, for my FIDE rating it is 120 games over 7 years, so about 16-17 games a year. For my ECF grade over the same period (2007-2013) it is 515 games over 7 seasons, so about 73 games a year on average. It may seem obvious, but that seems to be the main reason for many of the discrepancies between ECF grades and FIDE ratings. I agree with what you say about the effect of beating unrated players, but it is precisely because those (in many cases) count for ECF grading that you have divergent results between the two systems.

[Jonathan Bryant]

Perhaps we've just been talking about two different things? Clearly, it's a given that no system has any chance of being accurate based on a very small number of games. This is very much not the same problem that I've encountered.

Are you sure? ... I agree with what you say about the effect of beating unrated players, but it is precisely because those (in many cases) count for ECF grading that you have divergent results between the two systems.

Chris, Sorry I may be suffering from Sunday evening brain death, but I'm not quite following what you're asking. You seem to be questioning my conclusion that the problems with the FIDE system in London are not caused by randomness generated by small sample size but then in the bit after I've inserted a "..." you seem to agree with me that it isn't.


Situation A (what MC describes): a lot of players have elo ratings based on a handful of games. The ratings are therefore unreliable with many significantly over-rated compared to ECF conversion and many significantly under-rated. The only factor that's relevant here is that n=small number and in the long-term the situation would resolve itself if those players played more games.

Situation B (what I'm describing): players who play many elo rated games but something is going on which results in nearly everybody being significantly under-rated compared to ECF conversion. Various factors could be relevant here (i'm suggesting non-rating of a good chunk of games as well as other things e.g. under-rated juniors) but one which is not is randomness in ratings caused by n=low number. We know this because over-rateds compared to ECF conversion are rare to non-existent.

In situation B, in the long-term things will not resolve themselves if the players played more games. On the contrary the problem would probably get worse.


Two different things.


This, in any event, is my contention. (With the proviso that I'll believe that I"ve just been very unlucky if and when somebody gives me a convincing reason to believe that).

[John Hickman]

Chess players are far too hung up on ratings. Just play

[Christopher Kreuzer]

Situation A (what MC describes): a lot of players have elo ratings based on a handful of games. The ratings are therefore unreliable with many significantly over-rated compared to ECF conversion and many significantly under-rated. The only factor that's relevant here is that n=small number and in the long-term the situation would resolve itself if those players played more games.

Situation B (what I'm describing): players who play many elo rated games but something is going on which results in nearly everybody being significantly under-rated compared to ECF conversion. Various factors could be relevant here (i'm suggesting non-rating of a good chunk of games as well as other things e.g. under-rated juniors) but one which is not is randomness in ratings caused by n=low number. We know this because over-rateds compared to ECF conversion are rare to non-existent.

In situation B, in the long-term things will not resolve themselves if the players played more games. On the contrary the problem would probably get worse.


Two different things.

Ah, right. I get you now. My thought was that if I had played 73 FIDE rated games a year, my FIDE rating might be different. Do you think 16-17 FIDE rated games a year is enough to establish a stable (or rather, accurate) rating? I think someone (Leonard Barden?) mentioned a figure of around 50 per year for juniors wanting to progress. I suppose I am comparing with those who play 9-10 ECF games a year with all (or most) of those also FIDE rated. I think that if people play more FIDE rated games, it can resolve itself, but it depends where you play those games. How many chess players from the UK travel regularly to the continent to play FIDE rated chess, and how often to people come here from continental Europe? Is the flow mostly from continental Europe to here? (You might also want to include influxes from Eastern Europe, Russia, USA, Canada, and other Anglophone countries.) I suppose someone with a lot of time on their hands could analyse the FIDE rating list and tournament results...

[Sean Hewitt]

...over-rateds compared to ECF conversion are rare to non-existent.

Doesn't this simply point to the ECF conversion formula being wrong?

[Jonathan Bryant]

...over-rateds compared to ECF conversion are rare to non-existent.

Doesn't this simply point to the ECF conversion formula being wrong?

I don't think so. Clearly the conversion formula isn't working at the moment, but if it was just a case of finding the right formula then we'd all be in the same order on both lists wouldn't we? And we're not.

E.g. there's a junior who plays in nearly every event that I play in and we have similar ECF grades. But my rating is significantly higher than his. Similarly, I recall Peter Lalic's ECF grade rushing past mine but it being a good year or two before he overtook me on elo rating (by which time he was miles ahead of me on ECF).


@John: Strictly speaking I'm hung up on the rating system rather than my rating per se, but no doubt your suggestion is equally applicable.

@Chris: I'm not sure about numbers required. I suspect it's the wrong question anyway. If the system is inherently deflationary then it doesn't matter how many games you play.
Out of interest, why do you think the system will correct itself? The factors that are causing the deflation seem pretty permanent to me. And my personal situation is that as I play more and do better my FIDE is rising a bit, but my ECF is going up even more so relatively speaking the situation is indeed getting worse.