New grades (split from Editorial thread)

[Ian Kingston]

Is there any chance that people will move this discussion to the right forum?

[Roger de Coverly]

For instance, calculate yourself two average grades, one from your games against opponents lower rated than you, one for your games against opponents higher rated than you. If the latter is higher than the former, then your grading history will tend to confirm what the ECF have found in their analysis.

Sure. I actually did this in quarters for last season's results

under 151 151 to 175 176 to 201 and over 201

This was over 81 games which had an aggregate result of 174

The results were respectively 164 (23 games), 171(31 games), 192 (16 games) and 176 (11 games) which shows an under performance against lower rated players and an over performance against higher rated, but not against highest rated.

I don't see the leap of logic that these results should restore a grade (190) which I last held for one season over 30 years ago.

My observations are that it might just be down to style ( if you play "correctly" you improve your results against strong opposition but could concede more draws to the lower rated) or that the linear model underlying the ECF system is itself suspect.

I'm not aware that the ECF actually carried out such an analysis. If they did, it hasn't been published. What I think they did was to use the "new player" estimation recursion on everyone and then compared it to the actual grades.

[IM Jack Rudd]

Is there any chance that people will move this discussion to the right forum?

Done.

[IM Jack Rudd]

Hmmm. I meant to split this thread early on, not move the entire thread across.

[Sean Hewitt]

I'm not aware that the ECF actually carried out such an analysis. If they did, it hasn't been published. What I think they did was to use the "new player" estimation recursion on everyone and then compared it to the actual grades.

They did. They also asked me to investigate. I published my results. You know this, because you have previously read and commented on the results!!

[Sean Hewitt]

They were apparently concerned about player confidence in the system.

and rightly so.

Can you demonstrate deflation by reference to Tim's grading history - or mine for that matter?

It would be totally nonsensical to try to prove or disprove an axiom based on one or two players individually, when there are more than 30,000 live players in the ECF database. I would also need all of your individual results, and those of every one of your opponents, and their opponents - and so on. The evidence of deflation only exists when player’s records are compared against each other.

Turning career 130 players into 150's or 175's into 190's is overnight inflation to my mind. What's likely to happen is that the top of the food chain ( the Hebden's of this world) will gain points as they restore their historic differences to lesser beings. Equally the newly created 150's will be fed off by the fake 190's.

That's rather disingenuous. The idea is to take career 150 players, whose grade has eroded through the effects of grading deflation, and restore them to their rightful level. The evidence is incontrovertible, for those interested enough to examine it.

That's actually a decision to introduce inflation because you would get the same effect by leaving 100 players alone, taking 15 points off the 175's and 30 off the 250.

Nearly right. It doesnt introduce inflation, it fixes deflation.

You are right that the effect would be the same by leaving lower rated players alone and lopping points off the higher rated players. But the evidence suggests that the higher graded players (210 plus) have not suffered deflation. Why would you want to reduce their grade when the evidence suggests you shouldn't? More pertinently, the "old" ECF / FIDE conversion formula works pretty well for players above the 210 level (because these players haven't suffered much deflation). It falls over when applied to lower graded players (FIDE 2000 used to be ECF 175. Now, a 150 ECF player would expect to achieve such a rating). Or maybe this is just a coincidence?!

Once the grades are fixed the conversion formula needs changing back. It probably needs to become FIDE = ECF*8+650 (or 700) to account for the FIDE inflation - BUT IT MUST MUST BE FIDE = ECF *8 + x

[Sean Hewitt]

Can you demonstrate deflation by reference to Tim's grading history - or mine for that matter?

You can do little tests for this sort of thing yourself if you keep records of your games.

For instance, calculate yourself two average grades, one from your games against opponents lower rated than you, one for your games against opponents higher rated than you. If the latter is higher than the former, then your grading history will tend to confirm what the ECF have found in their analysis.

(In my case there is a 22 point difference between the two averages described above, the latter indeed higher than the former. So my personal experience strongly agrees with what the ECF have found.)

Hooray. Someone who examines the evidence - even though one persons results neither prove nor disprove anything. But if you did this for a large sample, it would be another proof. I'll do it this evening for the Leicester League and post the results tomorrow!

[Tim Spanton]

Hooray. Someone who examines the evidence - even though one persons results neither prove nor disprove anything. But if you did this for a large sample, it would be another proof. I'll do it this evening for the Leicester League and post the results tomorrow!

How does this prove deflation? Anything one player gains or loses is lost or gained by another player, so the grading average remains the same, which shows there is neither deflation nor inflation.

[IM Jack Rudd]

Hooray. Someone who examines the evidence - even though one persons results neither prove nor disprove anything. But if you did this for a large sample, it would be another proof. I'll do it this evening for the Leicester League and post the results tomorrow!

How does this prove deflation? Anything one player gains or loses is lost or gained by another player, so the grading average remains the same, which shows there is neither deflation nor inflation.

Er, no. True for an Elo-style system, but not true in the ECF system: if there is a significant correlation between form and activity in an ECF system, then the average grades can change by a lot.

(See the other megathread for Michael White's example of this, in which a group of six players, playing nobody but each other, can all go from 100 to 200+ in four years.)

[Roger de Coverly]

I would also need all of your individual results, and those of every one of your opponents, and their opponents - and so on. The evidence of deflation only exists when player’s records are compared against each other.

I thought the main grading database was the one being analysed and that would have all the results. The current ECF grading website allows individual histories to be seen. For example I've been about 30-40 points below the former international director for the last 15 years and Peter is now just one point lower than he was in 1994. Both of us have played more than 30 games every season.

The evidence is incontrovertible, for those interested enough to examine it.

I'm sorry but taking one year's results and drawing a few graphs is unconvincing. Justifing results by comparison to the international list is dubious since the international ratings in the 2000 - 2200 range are or were inflated by the very nature of the rules and tournaments required to get on the list.

[Tim Spanton]

Hooray. Someone who examines the evidence - even though one persons results neither prove nor disprove anything. But if you did this for a large sample, it would be another proof. I'll do it this evening for the Leicester League and post the results tomorrow!

How does this prove deflation? Anything one player gains or loses is lost or gained by another player, so the grading average remains the same, which shows there is neither deflation nor inflation.

Er, no. True for an Elo-style system, but not true in the ECF system: if there is a significant correlation between form and activity in an ECF system, then the average grades can change by a lot.

(See the other megathread for Michael White's example of this, in which a group of six players, playing nobody but each other, can all go from 100 to 200+ in four years.)

But the point is average grades have NOT been changing ...

[Ian Kingston]

Is there any chance that people will move this discussion to the right forum?

Done.

Thanks!

[Sean Hewitt]

Hooray. Someone who examines the evidence - even though one persons results neither prove nor disprove anything. But if you did this for a large sample, it would be another proof. I'll do it this evening for the Leicester League and post the results tomorrow!

How does this prove deflation?.

Because the lower graded player is performing better than the higher graded player it shows [anecdotally] that the grades have become stretched. Given that we know this phenomanon does not occur at the very highest end of the grading scale the stretching is downwards (deflation) rather than upwards (inflation.

Anything one player gains or loses is lost or gained by another player, so the grading average remains the same, which shows there is neither deflation nor inflation.

No, it doesn't!!! Rather than going into the maths and the algebra of the thing, let me give you a simple if exagerated example.

Three players, A, B and C play games amongst themselves whilst stuck on their proverbial desert island. They are graded 170 ; 145 ; 120. Their [mean] average is 145.

We know from ECF data that players who are graded 25 points higher score 67%.

Player A and Player B play 100 matches and the score is 67 : 33 (67%)

Player B and Player C play 300 matches. The score is 201 : 99 (67%)

The results are sent to the ECF for grading (!)

Player A is now graded 162

Player B is now graded 141

Player C is now graded 128

The mean average of the group is now 143.67, not 145. DEFLATION!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

[Sean Hewitt]

I thought the main grading database was the one being analysed and that would have all the results.

Nope. Events have only had to submit full results since 2005. Before that they just submitted summaries know as bundles which cannot be analysed to that extent.

[Sean Hewitt]

I thought the main grading database was the one being analysed and that would have all the results.

Nope. Events have only had to submit full results since 2005. Before that they just submitted summaries know as bundles which cannot be analysed to that extent.

The current ECF grading website allows individual histories to be seen. For example I've been about 30-40 points below the former international director for the last 15 years and Peter is now just one point lower than he was in 1994. Both of us have played more than 30 games every season..

My mistake. Your comparison of two players is clearly more accurate than the 6,302 players and their 61,375 games that I examined.

I'm sorry but taking one year's results and drawing a few graphs is unconvincing. Justifing results by comparison to the international list is dubious since the international ratings in the 2000 - 2200 range are or were inflated by the very nature of the rules and tournaments required to get on the list.

There's none so blind as he who will not see. 61,000+ games is more than enough of a sample to draw meaningful conclusions. The graphs demonstrate visually for the non-mathematician (such as yourself perhaps?) the formulae that I calculated - not the other way around. If you read what I have said, I have not justified my work by referencing the FIDE conversion formula. Rather, I actually said

If we apply the correction showed above then the conversion formula

ELO = ECF*8+600

produces a good approximation for players above 150 (ie above current ECF130). Below that level the approximation is not so good, but this is perhaps explained by the surprisingly high ELO ratings of players below ECF130 which on average is 1913.

In other words, I am commenting that the fix to the grading deflation which I calculated just happens to fix the FIDE conversion formula as well. This could be a coincidence. But I suspect not