Database Upgrade

Request amendments or pass comments on this free service.
User avatar
Carl Hibbard
Posts: 6028
Joined: Fri Dec 08, 2006 8:05 pm
Location: Evesham

Database Upgrade

Post by Carl Hibbard » Tue Mar 03, 2009 9:30 pm

The online system has been upgraded to the latest database for the fifth release of this year
Cheers
Carl Hibbard

Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: Database Upgrade

Post by Roger de Coverly » Wed Mar 04, 2009 12:10 am

On a very brief test, they've changed the "new" grades. I've had a point removed from 180 to 179.

It looks as if they've tuned down the revaluation of the juniors.

If there's a download available, it might be worth testing the full extent of the changes to "new" grades. Obviously it's a total secret as to what alterations have been made. They never have published the "new" 2007 equivalents so we can see whether the "new" 2008 are grossly inflated or just minimally inflated.

User avatar
Carl Hibbard
Posts: 6028
Joined: Fri Dec 08, 2006 8:05 pm
Location: Evesham

Re: Database Upgrade

Post by Carl Hibbard » Wed Mar 04, 2009 7:02 am

Roger de Coverly wrote:Obviously it's a total secret as to what alterations have been made.
It's not a "secret" but I am not sure what changes were made just an error in the calculations which has now been corrected

Richard or Howard will hopefully login and answer your concerns more fully
Cheers
Carl Hibbard

Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: Database Upgrade

Post by Roger de Coverly » Wed Mar 04, 2009 9:53 am

I am not sure what changes were made just an error in the calculations which has now been corrected
Comparing v3 (last August) with v5 (current), it's apparent that the revaluation of juniors has been scaled back with evidently a knock-on effect on adult grades. One could speculate that the junior increment had been double-counted or the estimation process for new players revisited.

It doesn't say much for the credibility of the whole exercise that the junior grades were queried last year and defended as correct by the advocates of the changes.

User avatar
John Upham
Posts: 7162
Joined: Wed Apr 04, 2007 10:29 am
Location: Cove, Hampshire, England.
Contact:

Re: Database Upgrade

Post by John Upham » Wed Mar 04, 2009 11:12 am

Carl Hibbard wrote:
Roger de Coverly wrote:Obviously it's a total secret as to what alterations have been made.
It's not a "secret" but I am not sure what changes were made just an error in the calculations which has now been corrected

Richard or Howard will hopefully login and answer your concerns more fully
Carl,

I assume that the "error in the calculations" would not have been corrected had the feedback from this place and elsewhere been so strong?

I await full transparency on these matters!
British Chess News : britishchessnews.com
Twitter: @BritishChess
Facebook: facebook.com/groups/britishchess :D

Howard Grist
Posts: 83
Joined: Tue Dec 12, 2006 1:14 pm
Location: Southend-on-Sea
Contact:

Re: Database Upgrade

Post by Howard Grist » Wed Mar 04, 2009 2:22 pm

There have been two alterations made to the calculations of new grades for this database upgrade.

Firstly, I was taking the grading performance of juniors including the junior enhancement over the period 2003-6 to be their performance over the same period without junior enhancement (e.g if an adult played a junior in 2003 and that junior was graded 100 in 2006 then I was assuming that the adult played against someone graded 100 in 2003) Correcting this error had a general effect of reducing the grades of juniors.

Secondly the Junior Enhancements have changed. These are now as published on the new grades page http://grading.bcfservices.org.uk/newgrades.php. This has the effect of increasing the grades of juniors under 14, and decreasing the grades of older juniors - if you're over 14 then you no longer get an enhancement.

The overall impact is that the mean junior grade has increased, but if you're the average adult player looking at your opponents, you'll probably think that the reverse has happened.

The effect on adult players is minimal.
Former ECF Grading System Programmer

User avatar
John Upham
Posts: 7162
Joined: Wed Apr 04, 2007 10:29 am
Location: Cove, Hampshire, England.
Contact:

Re: Database Upgrade

Post by John Upham » Wed Mar 04, 2009 2:37 pm

Howard,

is the junior bonus the same (for a particular age) regardless of whether the junior is improving or not?

i.e. would a recreational player aged 13 get the same bonus as say Felix Jose Ynojosa-Aponte?

Should the bonus be a function of the slope of their periodic performance change therefore?
British Chess News : britishchessnews.com
Twitter: @BritishChess
Facebook: facebook.com/groups/britishchess :D

Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: Database Upgrade

Post by Roger de Coverly » Wed Mar 04, 2009 3:43 pm

(e.g if an adult played a junior in 2003 and that junior was graded 100 in 2006 then I was assuming that the adult played against someone graded 100 in 2003) Correcting this error had a general effect of reducing the grades of juniors.
Well that's a pretty small error of 50 points or so in the 2003 calculation.

What about improving adult players? Do they have the same (inflationary) effect?

How about generic inflation? When I tested the grades for people in both the 2007 and 2008 list, the mean as in sum(grades)/count(people) was almost the same for both the 2007 and 2008 calculations. Does the same apply to "new" grades?
If you're over 14 then you no longer get an enhancement.
So for players 14 and over, the "new" grades are asserting that then can score 50% against adults of the same "new" grade. Is there any empirical testing of this, say against this season's results?

Sean Hewitt

Re: Database Upgrade

Post by Sean Hewitt » Thu Mar 05, 2009 9:33 am

In my opinion, the whole concept of "junior enhancement" is flawed, simply because juniors improve at different rates even if they are the same age. The analysis I did back in 2006 showed this conclusively.

The most accurate way to grade games against juniors (and any other rapidly improving player) is to treat them as a new players each year, work out their grading performance and then use the performance to grade the games of their opponents. Thus a 100 junior who improves a bit might be graded as a 110 whilst a player who improves alot might be graded as a 130. This is better than graidng both as if they were 120.

Howard Grist
Posts: 83
Joined: Tue Dec 12, 2006 1:14 pm
Location: Southend-on-Sea
Contact:

Re: Database Upgrade

Post by Howard Grist » Fri Mar 06, 2009 12:59 am

John Upham wrote:is the junior bonus the same (for a particular age) regardless of whether the junior is improving or not?

i.e. would a recreational player aged 13 get the same bonus as say Felix Jose Ynojosa-Aponte?

Should the bonus be a function of the slope of their periodic performance change therefore?
The junior bonus depends solely on age. It is used to indicate how much a player will improve over the coming 6 (for rapid play) or 12 (for standard play) months. Probably the best indicator of how much a player improves is the amount of chess he plays over this period compared to how much he played previously. As this is not something known to the grading system, I think this is the best that can be done.
Roger de Coverly wrote:What about improving adult players? Do they have the same (inflationary) effect?
One of the grading system's assumptions is that an adult player's playing strength remains, on the whole, constant. There will be improving players over the initial 2002/6 period that will have an (inflationary) effect, there will also be players whose playing strength is deteriorating and that will have a deflationary effect.
Roger de Coverly wrote:How about generic inflation? When I tested the grades for people in both the 2007 and 2008 list, the mean as in sum(grades)/count(people) was almost the same for both the 2007 and 2008 calculations. Does the same apply to "new" grades?
The mean new grade of adult players with an A-C category standard play grade in both 2007 and 2008 was 141.05 in 2007 and 141.11 in 2008 For old grades, the equivalent figures are 121.24 in 2007 and 120.52 in 2008.
Roger de Coverly wrote:So for players 14 and over, the "new" grades are asserting that then can score 50% against adults of the same "new" grade. Is there any empirical testing of this, say against this season's results?
Analysis showed that Juniors aged 15-17 had an actual grading performance against adults in 2006/7 0.2 grading points higher per game played than they did in 2005/6. This is too small to give any increment for.
Sean Hewitt wrote:In my opinion, the whole concept of "junior enhancement" is flawed, simply because juniors improve at different rates even if they are the same age. The analysis I did back in 2006 showed this conclusively.

The most accurate way to grade games against juniors (and any other rapidly improving player) is to treat them as a new players each year, work out their grading performance and then use the performance to grade the games of their opponents. Thus a 100 junior who improves a bit might be graded as a 110 whilst a player who improves alot might be graded as a 130. This is better than graidng both as if they were 120.
Difficult to argue with this, except that if you do calculate juniors' grades as you are suggesting, then no-one who's played a junior will be able to predict what their new grade will be which strikes me as a major drawback to this idea.
Former ECF Grading System Programmer

Richard Bates
Posts: 3338
Joined: Fri Nov 14, 2008 8:27 pm

Re: Database Upgrade

Post by Richard Bates » Fri Mar 06, 2009 8:02 am

Howard Grist wrote: Difficult to argue with this, except that if you do calculate juniors' grades as you are suggesting, then no-one who's played a junior will be able to predict what their new grade will be which strikes me as a major drawback to this idea.
I don't see why this is necessarily a disadvantage. What real value is there in being able to predict your grade, unless you intend to use it it order to manipulate the outcome?

In fact the entire system should be changed as follows:

Step 1 - "end of year grades" calculated as happens presently
Step 2 - Average grade for the year established as the difference between start of year grade, and end of year grade
Step 3 - New grade calculated on the basis of numbers produced by step 2.

Bit of extra work obviously... ;)

Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: Database Upgrade

Post by Roger de Coverly » Fri Mar 06, 2009 8:09 am

The mean new grade of adult players with an A-C category standard play grade in both 2007 and 2008 was 141.05 in 2007 and 141.11 in 2008 For old grades, the equivalent figures are 121.24 in 2007 and 120.52 in 2008.
I've highlighted the word adult because I was testing against all players. The implication of what you say is that the mean for all players including juniors has increased. On old grades it was static. Thus I think the new rules for junior increments are injecting inflation into the system. Historically I don't think even the prodigies like Short, Hodgson etc. managed a 150 standard of play whilst well under 10.

So I conclude you've added 20 points to the system and introduced rules which seem to be inflationary. Why is it supposed this will not filter up to the top players?

Has any work been done to show the effects of

(a) the recalculation with the existing junior increment rules
(b) just the new junior increment rules against the published list - eg rebase the 2007 list with the new increments and rerun 2008 - see which players it most affects.

I thought the original idea of this exercise was to recalculate adult grades and fit junior grades around them, not to revisit how tournaments for the under 10s are rated.

Analysis showed that Juniors aged 15-17 had an actual grading performance against adults in 2006/7 0.2 grading points higher per game played than they did in 2005/6. This is too small to give any increment for.
Is that a calculation on "old" grades, "new" grades or both? Given the excess revaluation in the first version of new grades, you might not get the same results. So given that 15 -17 year olds are more likely to have played overrated under 14s, it seems plausible that they will be overrated on "new" grades by comparison with adults of the same nominal grade.

Sean Hewitt

Re: Database Upgrade

Post by Sean Hewitt » Fri Mar 06, 2009 8:49 am

Howard Grist wrote:
Sean Hewitt wrote:In my opinion, the whole concept of "junior enhancement" is flawed, simply because juniors improve at different rates even if they are the same age. The analysis I did back in 2006 showed this conclusively.

The most accurate way to grade games against juniors (and any other rapidly improving player) is to treat them as a new players each year, work out their grading performance and then use the performance to grade the games of their opponents. Thus a 100 junior who improves a bit might be graded as a 110 whilst a player who improves alot might be graded as a 130. This is better than graidng both as if they were 120.
Difficult to argue with this, except that if you do calculate juniors' grades as you are suggesting, then no-one who's played a junior will be able to predict what their new grade will be which strikes me as a major drawback to this idea.
You're right - that's the drawback. However, I don't know how important this is to the majority of players. It isn't at all important to me (so I'm not best placed to judge), and I would sacrifice it for the sake of grading accuracy. Of course, the same scenario already applies when playing an ungraded player, and we seem happy enough to forgo players predicting their new grade then!

Howard Grist
Posts: 83
Joined: Tue Dec 12, 2006 1:14 pm
Location: Southend-on-Sea
Contact:

Re: Database Upgrade

Post by Howard Grist » Sat Mar 07, 2009 1:22 am

Richard Bates wrote:I don't see why this is necessarily a disadvantage. What real value is there in being able to predict your grade, unless you intend to use it it order to manipulate the outcome?
Personally I like to see how I'm doing during the season.
Richard Bates wrote:In fact the entire system should be changed as follows:

Step 1 - "end of year grades" calculated as happens presently
Step 2 - Average grade for the year established as the difference between start of year grade, and end of year grade
Step 3 - New grade calculated on the basis of numbers produced by step 2.
This probably has some mathematical sense to it, however it does remove the possibility of even checking that your grade calculation is correct. For instance if I play an improving 100, their grade at step 1 comes out at 120, so I am credited with playing someone graded 110. However this player is now re-graded in step 3 and due to the 40 point rule benefit of his extra 10 points comes out at 122. How on earth do I know that he was assumed to be 110 for my game?
Roger de Coverly wrote:I've highlighted the word adult because I was testing against all players. The implication of what you say is that the mean for all players including juniors has increased. On old grades it was static. Thus I think the new rules for junior increments are injecting inflation into the system.
Yes, the grade of all players in the system with grades in both years he system has increased. This is exactly what I would expect to happen. You have even identified the cause of this increase - it is the junior increment. However the central principle of the ECF grading system is that the pool of adult players is constant. If this happens, then you can't say that inflaton is taking place.
Roger de Coverly wrote:Has any work been done to show the effects of

(a) the recalculation with the existing junior increment rules
(b) just the new junior increment rules against the published list - eg rebase the 2007 list with the new increments and rerun 2008 - see which players it most affects.
I'm not sure what point you're trying to make here. I am unable to calculate grades using the current rules without using junior increments. As there has been no analysis of these conducted why should the existing junior increments be correct? The junior increments are prone to change over time in any case as it is how much an average x-year old player is likely to improve. If the number of junior players doubles, I would expect this to go down.
Roger de Coverly wrote:
Howard Grist wrote:Analysis showed that Juniors aged 15-17 had an actual grading performance against adults in 2006/7 0.2 grading points higher per game played than they did in 2005/6. This is too small to give any increment for.
Is that a calculation on "old" grades, "new" grades or both? Given the excess revaluation in the first version of new grades, you might not get the same results. So given that 15 -17 year olds are more likely to have played overrated under 14s, it seems plausible that they will be overrated on "new" grades by comparison with adults of the same nominal grade.
That was an analysis of new grades. No junior increments were used in the analysis - the point of the analysis was to calculate what the junior increments should be. I calculated grades for the adults based solely on their games with other adults, then calculated grading performances for each junior year range against these adults.
Former ECF Grading System Programmer

Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Re: Database Upgrade

Post by Roger de Coverly » Sat Mar 07, 2009 8:46 am

However the central principle of the ECF grading system is that the pool of adult players is constant.
Why? That is not the principle that has been used in the past. Back in the late eighties, there was concern that the overall pool was inflating. As a consequence, the junior increment was refined. In those days, the grading list published actual performance and the junior effect was taken into account on the following season's rating by use of the formula

win = opponents grade + 50 + junior increment.

In order to support age related increments, the published list had to be changed to include the junior increment.
If this happens, then you can't say that inflation is taking place.
Historically the health of the system was measured from the average grade of all the participants. Are you now saying that we should expect to see the average grade of all players increasing? I think you are also saying that the analyses in the grading lists of the past few years were false. Looking at the report in the 2002-2003 grading list (July 2002), this observes that for the category "played >7 games both years", "All players" went from an average of 120 in the 2000 list to 121 in the 2001 list ( it went up a point). Adults stayed constant at 122, Juniors went up from 98 to 107. Remember this is the played both years category - so there's an age and experience effect in this junior statistic. Across all published players, the average went from 115 to 114. For adults from 118 to 117. For juniors from 86 to 91. The commentary suggests use of the "played > 7 games" as the most "reliable" average as it accounts for 80% of games played.
I'm not sure what point you're trying to make here. I am unable to calculate grades using the current rules without using junior increments.
Is it difficult? You take the 2007 published grades You remove the junior increments which presumably were added at the end after the 2007 performances had been calculated. You either run 2008 with no junior increment at all or with whatever formula you wish to test. You compare the resulting 2008 grades with the "actuals". You've then got some benchmarks for the effect of different formulae. I doubt that the players above 200 would be affected by any change involving under 10 s but the under 100s might well be.
I calculated grades for the adults based solely on their games with other adults, then calculated grading performances for each junior year range against these adults.
That's an artificial construction because there is (or rather was) no such thing as a grade calculated against other adults. The way the junior increment was tested historically was to follow players through the age ranges and see how their grading improved with age and experience. This measures their performance both against other adults and other juniors. The strength of a junior increases not just against adults but also against their fellow and younger peers.

Even before you use a grading system to predict results, is not its purpose to give you player rankings by strength ? Thus you can use it for team board orders, competition and prize eligibility and seeded swiss pairings. Some of us believe that the "new" grades have broken this fundamental ranking principle especially for the younger juniors.

In the "new" grades, the pool of adult players may still be in the right order if a bit compressed. The pool of junior players may still be in the right order if a bit elongated. I just don't think the merge is right though. Reality check - do you really believe in under 10 players of a 150 standard?

I'm aware that the whole exercise was started off by the observation that the formula (BCF*8 +600) = FIDE no longer seemed to work at the sub 2200 level. Did no one in the grading team have sufficient experience of international competition to be aware that inflated ratings in the sub 2200 range was most plausibly down to the effects of the historic 2000 cutoff for international rating combined with the way that seeded pairings in large swisses usually mean that players always play people several hundred points above or below them? You could not infer deflation in the ECF system from this comparison.

Locked