Rough Guide to new grades

[Paul McKeown]

You're half right there. The cause is actually well known,

I disagree. There has been a lot of speculation and a variety of theories have been advanced, some of which have been "proved" by those proposing them. None of them have been subjected to any truly independent peer review.

And that astonishes me. The most valuable service provided by the English Chess Federation is the grading system. Any responsible change process would have subjected any proposals to rigorous, independent audit - professional, paid for and independent of the chess world.

[Mike Gunn]

If a grading system was 100% reliable as a predictor, no grade would ever change.

Except for lighthousekeepers, of course.

[Roger de Coverly]

Except for lighthousekeepers, of course.

There's an easy answer to the lighthouse keeper problem.

Keeper A - grading 130 plays a 30 game match against keeper B - grading 160. They score 50%. If you publish a grading list after 30 games, they swop grades. If you publish the grading list after 15 games, then player A is 50% * 160 + 50% * 130 ( on the 30 game rule). Player B is of course 50% * 130 + 50% * 160. So both are 145.

Those who think "grades are deflated" have to have an answer to the Arkell (and Turner) hypothesis. Simply that top players have got better. Those who have kept pace with "top" players have maintained their grades. Those who haven't have lost points. The probability is that the gulf between the GM and the "average" player has widened as has the gulf between the "average" player and the newcomer to competitive chess. The effect of the ECF changes is to knock something like 30 points off the distance between the grades of Arkell, Turner, Hebden etc. and the "average" player. I don't think it's justified.

[Paul McKeown]

Those who think "grades are deflated" have to have an answer to the Arkell (and Turner) hypothesis. Simply that top players have got better. Those who have kept pace with "top" players have maintained their grades. Those who haven't have lost points. The probability is that the gulf between the GM and the "average" player has widened as has the gulf between the "average" player and the newcomer to competitive chess. The effect of the ECF changes is to knock something like 30 points off the distance between the grades of Arkell, Turner, Hebden etc. and the "average" player. I don't think it's justified.

Roger, I couldn't agree more. Barking mad, the whole thing. Completely divorced from reality.

There's an easy answer to the lighthouse keeper problem.

Keeper A - grading 130 plays a 30 game match against keeper B - grading 160. They score 50%. If you publish a grading list after 30 games, they swop grades. If you publish the grading list after 15 games, then player A is 50% * 160 + 50% * 130 ( on the 30 game rule). Player B is of course 50% * 130 + 50% * 160. So both are 145.

True. Do you propose a weighted average between two lists? Or which exact mechanism? If one literally chunked up groups of 15 games within each given list, this wouldn't work for lighthouse keepers who played each other 10 times every season. Or do you believe, as I do that those pesky lighthouse keepers are more of a theoretical amusement, rather than a real issue?

[Roger de Coverly]

Or do you believe, as I do that those pesky lighthouse keepers are more of a theoretical amusement, rather than a real issue?

Of course. The point is that the frequency of publication and the minimum number of games to qualify for a grade that "forgets" the previous experience are parameters of the system.

Elo methods would also converge the lighthouse keepers to the same rating. How many games this takes is a function of the k factor. There's a lot of stuff about k factors etc at http://www.chessbase.com/newsdetail.asp?newsid=5527

[IM Jack Rudd]

Elo ratings would only converge the lighthouse keepers to each others' grades if they played few enough games. If they played enough, they'd end up with their grades getting further away from each other.

[Richard Bates]

Those who think "grades are deflated" have to have an answer to the Arkell (and Turner) hypothesis. Simply that top players have got better. Those who have kept pace with "top" players have maintained their grades. Those who haven't have lost points. The probability is that the gulf between the GM and the "average" player has widened as has the gulf between the "average" player and the newcomer to competitive chess. The effect of the ECF changes is to knock something like 30 points off the distance between the grades of Arkell, Turner, Hebden etc. and the "average" player. I don't think it's justified.

Although i have also advanced this theory, I must admit i have my doubts.

Firstly from a theoretical point of view one needs to explain why the "deflation" only seems to have materialised in the last decade, whereas the theory would seem to suggest that it should have been happening since the start of the grading system. One could try and link in the development of computer databases as something which have aided the top players more than the weaker, but it seems a bit tenuous. One is also going to struggle to convince me that noted theoreticians messrs Arkell and Turner have exploited computer technology to its fullest extent!

Secondly on an anecdotal basis my grade has I think clearly stayed pretty much the same (within a normal distribution, and cumulative variances caused by the 40pt rule) for over a decade. (217,221,220,221,218,214,216,220,226,225,216,209,221,222). I don't believe that i have made any material improvement in strength over that time (arguably the reverse) and yet i appear to have suffered no adverse consequences. Although like i say, isolated and anecdotal, I don't really see why i should be so unaffected whereas players much further down the food chain, who have no direct contact with these "improving top players" are declining rapidly.

[Matthew Turner]

Richard et al,
Let me try to give an explanation why deflation has occured in the last. I believe the problem is the greater use of computing in the process of grading. Imagine a club competition with two new players A and B and two established players graded 40 and 100

A B 40 100
A X 0 1 0.5
B 1 X 0 0
40 0 1 X 0.5
100 0.5 1 0.5 X

What used to happen was that the local estimated a grade for new players let's say 80.

Player A now gets 30+90+100 = 220 grading points
Player B get 130 +(-10) + 50 = 170

However, (as I understand it) the computer now awards new players a grade to calculate their other results on
Player A win against 40 = 90 plus draw aginst 100 = 100 therefore performance = 95
Player B loss against 40 = -10 plus loss against 100 = 50, therefore performance = 20

For player A's loss against B he can only receive a maximum of 90 less than his 'grade' so gets 5 points. Therefore his total grading points are 185 - a loss of thirty-five.
For player B's win aginst A he can only get a maximum of 90 plus his grade so 110, meaning his total point are 150 - a loss of twenty.

There is nothing wrong with computers, but you do have to understand what they are doing.

[E Michael White]

…"the resulting grades would be between 50 and 150 as long as there is no 40 point rule"……How on Earth do you manage to come up with that proposition?

The ECF formula applied in large all-play-all tournaments produces this result. Whatever the starting grades the performance results are always within approximately 50 points of the mean Tournament grade, provided there is no 40 point rule.

[Paul McKeown]

The ECF formula applied in large all-play-all tournaments produces this result. Whatever the starting grades the performance results are always within approximately 50 points of the mean Tournament grade, provided there is no 40 point rule.

…"the resulting grades would be between 50 and 150 as long as there is no 40 point rule"……How on Earth do you manage to come up with that proposition?

The ECF formula applied in large all-play-all tournaments produces this result. Whatever the starting grades the performance results are always within approximately 50 points of the mean Tournament grade, provided there is no 40 point rule.

Resulting in an absurdity. Happily the real world is not capable of producing such. What intrigues me is the sort of throught process that would consider such an outcome desirable.

[Roger de Coverly]

What used to happen was that the local estimated a grade for new players let's say 80.

What used to happen is that there was in effect an unofficial and locally based minimum grade. Depending on the grader, new players were assigned a grade of somewhere between 0 and 100 unless their performances indicated they had played at a strong level previously. With the national database, this was abandoned and the initial grading estimate was based on the actual performances. Howard Grist described it somewhere on this forum. Whilst this is "more accurate" it does have the disadvantage of removing the quantitative easing ( printing grading points) effect of graders estimates.

It's lead to the appearance of negative grades and may have dropped the grades of the "average" player a point or two. I don't think there's yet any impact at 150 plus.

There's a case for a minimum grade for new entrants - most online servers do this.

I don't believe that i have made any material improvement in strength over that time (arguably the reverse) and yet i appear to have suffered no adverse consequences. Although like i say, isolated and anecdotal, I don't really see why i should be so unaffected whereas players much further down the food chain, who have no direct contact with these "improving top players" are declining rapidly.

I suppose we should ask how much preparation a 4NCL and London League board 1 does before the game. Anything greater than zero is more than the "average" player. At these sort of events, it's possible to keep up with theory just by looking around the playing room.

[Richard Bates]

Richard et al,
Let me try to give an explanation why deflation has occured in the last. I believe the problem is the greater use of computing in the process of grading. Imagine a club competition with two new players A and B and two established players graded 40 and 100

A B 40 100
A X 0 1 0.5
B 1 X 0 0
40 0 1 X 0.5
100 0.5 1 0.5 X

What used to happen was that the local estimated a grade for new players let's say 80.

Player A now gets 30+90+100 = 220 grading points
Player B get 130 +(-10) + 50 = 170

However, (as I understand it) the computer now awards new players a grade to calculate their other results on
Player A win against 40 = 90 plus draw aginst 100 = 100 therefore performance = 95
Player B loss against 40 = -10 plus loss against 100 = 50, therefore performance = 20

For player A's loss against B he can only receive a maximum of 90 less than his 'grade' so gets 5 points. Therefore his total grading points are 185 - a loss of thirty-five.
For player B's win aginst A he can only get a maximum of 90 plus his grade so 110, meaning his total point are 150 - a loss of twenty.

There is nothing wrong with computers, but you do have to understand what they are doing.

That all may be true Matthew, not that i can be bothered to try and understand it, but it seems a somewhat different argument to the "top players are getting better" argument advanced in your name above.

[Richard Bates]

I suppose we should ask how much preparation a 4NCL and London League board 1 does before the game. Anything greater than zero is more than the "average" player. At these sort of events, it's possible to keep up with theory just by looking around the playing room.

I would be surprised if i do more preparation than the "average player". And certainly not more than my average opponent which is surely more relevant to looking at the changes of my grades in isolation. Not that it's really relevant to the question of whether one has got stronger over the last decade - you'll just have to take my word for it that i almost certainly haven't. The game isn't solely about avoiding catastrophe in the opening.

[Ian Thompson]

Richard et al,
Let me try to give an explanation why deflation has occured in the last. I believe the problem is the greater use of computing in the process of grading. Imagine a club competition with two new players A and B and two established players graded 40 and 100

A B 40 100
A X 0 1 0.5
B 1 X 0 0
40 0 1 X 0.5
100 0.5 1 0.5 X

What used to happen was that the local estimated a grade for new players let's say 80.

Player A now gets 30+90+100 = 220 grading points
Player B get 130 +(-10) + 50 = 170

However, (as I understand it) the computer now awards new players a grade to calculate their other results on
Player A win against 40 = 90 plus draw aginst 100 = 100 therefore performance = 95
Player B loss against 40 = -10 plus loss against 100 = 50, therefore performance = 20

For player A's loss against B he can only receive a maximum of 90 less than his 'grade' so gets 5 points. Therefore his total grading points are 185 - a loss of thirty-five.
For player B's win aginst A he can only get a maximum of 90 plus his grade so 110, meaning his total point are 150 - a loss of twenty.

There is nothing wrong with computers, but you do have to understand what they are doing.

This explanation doesn't convince me there has been deflation in the past.

First, its based on the assumption that in the days when graders had to estimate the grades of ungraded players and use them in their calculations, overall they overestimated the true strength of those players. What evidence is there to support this?

Second, if we assume that graders did overestimate the strength of ungraded players in the past, that would have lead to inflation in the grading system while they were doing it. Now they no longer have to do this estimate, there is no longer inflation due to it. That's left us with an inflated grading system.

Third, take your example of 2 graded and 2 ungraded players. Before the event there were 2 graded players and the average grade of all graded players was 70. After the event there were 4 graded players and the average grade of all graded players was 63 (see http://www.ecforum.org.uk/viewtopic.php?f=4&t=470 for the calculation of this). If we change the results so that B beats the player graded 40 instead of losing, then, after the event, the average grade of all graded players is 77. One set of results causing the overall average grade to go down and another causes it to go up. What evidence is there to show that, in reality, either of these scenarios occurs more often than the other?

[Matthew Turner]

Roger de Coverley wrote
Whilst this is "more accurate" I am not sure that there is any evidence for this, just because a computer works something out doesn't make it more accurate, it all depends on what it has been told to work out. Do you believe that it is better to calculate a foreign player's strength on 4 games in the 4NCL or by doing a conversion from their FIDE rating? The ECF grading computer thinks the former.