What are the new conversion formulae?

[Paul McKeown]

1. between ECF grades and FIDE ratings
2. between ECF grades and general national rating systems
3. are there any suggested national variations on these conversion formulae and do they apply throughout the grading scale

[Roger de Coverly]

1. between ECF grades and FIDE ratings
2. between ECF grades and general national rating systems

Elo= 8*"New ECF" + 650 has been quoted without distinction. It's about right in the 2000-2200 range because international Elo is perhaps 100 points inflated in that range and "new" ECF about 50. The Scots though don't accept the formula because they consider "new" grades to be inflated relative to their domestic Elo particularly at the lower ends of the scale.

To make matters more complicated, Yorkshire have their own "private" grading system.http://www.chessnuts.org.uk/ny5/
No changes to its methodology have been made.

[Dean Madden]

http://grading.bcfservices.org.uk/help.php#elo

has

ECF x 8 + 650 = FIDE
ECF x 8 + 600 = National Elo

[Andy Howie]

The Scots though don't accept the formula because they consider "new" grades to be inflated relative to their domestic Elo particularly at the lower ends of the scale.

Damn right, just because you are buggering up your grading system, shouldn't mean that we should do the same to ours. Lets use my grade as a simple example 1363. My ECF grade is 120. Converting using x8 + 600, you get 95. Interestingly my ECF grade last year (prior to the fudge factor) was 90!

Looking at the people I have played in congresses in England, their grades have gone up pretty much the same as my ECF grade so this means using the formula, I would be beating a 1560, which although nice, would give me a massive boost in grading performance that is not realistic.

Here is something to think about. Any Scot coming down to play in an ECF tournament, usually enters using the CS grade, Why? Well for one it is more accurate as it is based on more games. This will have the benefit for most of us as we will be able to play in tournaments 20 - 30 points below our actual strength. I predict England is going to be a popular venue for Scots this year

[Roger de Coverly]

Damn right, just because you are buggering up your grading system, shouldn't mean that we should do the same to ours.

Did you ever have anybody in Scotland plot rating difference against result and test how well or badly it fitted the Elo percentages? If you did, I wonder whether it looked the same as the English one. In any event, you didn't decide that it would be a good idea to reduce everybody's grade to 80% of its previous value ( and add something back so it doesn't look totally stupid)

[Andy Howie]

actually yes, a statistical analysis was carried out in the last 2 years on our grades and they were found to be sound. Certainly was not what I expected.

I have a feeling that the reason for this is we have a drift calculation for players that play more than 30 games a season. I suspect that keeps it in check

[Roger de Coverly]

actually yes, a statistical analysis was carried out in the last 2 years on our grades and they were found to be sound.

Did the analysis ever make it to a website? I was vaguely aware that some sort of investigation had been done but not what it had established. In fact for players rated in both systems it tends to contradict the ECF analysis since previously you felt that SCO = 8 * ENG + 600 was fine.

[Brian Valentine]

Roger,

The report on the Scottish Rating system can be found here:

http://www.chessscotland.com/grading/gr ... ssep08.htm

Andy says it is sound, even with evident stretch!

[Roger de Coverly]

Andy says it is sound, even with evident stretch!

I did spot this comment though

However the discrepencies for larger grade differences are I believe not due to errors in the method of calculation but to the assumption that the ELO curve is valid

The shape is quite similar to the Grist graph - but they came to the opposite conclusion about the need to change anything. One of the comforts is that zero grade difference = 50% score

There's frequent reference to a 200 -up rule - this is a fudge which deals with rapidly improving players.

They didn't look at the grades as an historic series though - To my mind, the stability of that series is where you should test for inflation or deflation. If you start with an unstretched system and then stretch it, does it stabilise or continue stretching? If say you artificially increased every grade by 20%, do they stay at that level, reduce back towards zero or continue expanding.

[Brian Valentine]

I'm sure the Scots will continue to improve their monitoring.

I just commend them for publishing something; transparency should lead to a constructive debate. We are still waiting to see the modelling the grading team did to justify the the ecf change.

[Roger de Coverly]

We are still waiting to see the modelling the grading team did to justify the the ecf change.

I'm unconvinced there was much. They drew a graph, noted that it didn't produce the results they expected and that if they took 80% of the original values, the fit would be better. Realising that everyone would laugh if they made Mark and Keith 190 players, they added back about 40 to 50 points.

[Brian Valentine]

There may be several separate reasons for grades to inflate, deflate or stretch out, but the main one is very simple: if we don't play lots of games against opponents who also play lots of games, the list is bound to become gradually stretched. This has been demonstrated by mathematical modelling, but the results cannot easily be summarised in a meaningful way. We are looking at ways of making this information available to those who are interested.

This is from the report on the ecf website and suggests there is something else. It's just that my modelling of the old approach does not get to the same place! Furthermore what I have put on the forum suggests 1) Stretch must always be within the system (but may need controlling) and 2) the new junior solution does increase stretch.

[Ben Purton]

Basically

180= 2100
200= 2200
215=2300

In my view.

[Roger de Coverly]

There may be several separate reasons for grades to inflate, deflate or stretch out, but the main one is very simple: if we don't play lots of games against opponents who also play lots of games, the list is bound to become gradually stretched. This has been demonstrated by mathematical modelling, but the results cannot easily be summarised in a meaningful way. We are looking at ways of making this information available to those who are interested.

This is from the report on the ecf website and suggests there is something else. It's just that my modelling of the old approach does not get to the same place! Furthermore what I have put on the forum suggests 1) Stretch must always be within the system (but may need controlling) and 2) the new junior solution does increase stretch.

I remember that as well, but it just seemed hand-waving to me. I don't see how the grades of E players make any real difference to those of A players because you never play them ! You could eliminate E players at the stroke of a pen by increasing the publication threshold. Once upon a time it used to be 20 in two years.

The issue of the internal inflation or deflation has been raised because of the lack of zero sum when players with different activity levels meet. It's an issue of sorts but the calculation of average grades doesn't seem to point in any particular direction.

There was a statement "stronger players are more active". Actually it's the other way round, A rated players have higher average grades - the obvious point is that if you play more you get better. There seems to be an upper bound though, those at the top of the "most played" list mostly seem to keep the same grade.

[Brian Valentine]

A little care is required here since about 70% of non home players are category E and on average these players are pretty strong. Given the tournaments you play in Roger, I think you will come across them. Otherwise I agree.