Yet More Deflation

General discussions about ratings.
Ola Winfridsson
Posts: 324
Joined: Fri Aug 14, 2009 12:26 pm

Re: Yet More Deflation

Post by Ola Winfridsson » Tue Sep 07, 2010 7:54 pm

Sean Hewitt wrote:
Paul McKeown wrote:
IM Jack Rudd wrote:The more games a year you play, the less impact any individual result has on your ECF grade. If I play someone of my own grade and beat him, where I play 100 games a year and he plays only 50, this will have cost him 1 grading point but gained me only ½ a grading point.
And more active players are likely to be stronger than less active players, therefore ....
Absolutely right, although this has been well known for some time.
Yes and no. I'm sure we can all agree that the optimum number of games varies greatly between one person and the next. I for one couldn't even contemplate clocking up 100 games a season, although I know a fair few who do. Last season, I played nearly 60 games (from experience I know that 40-45 is my ideal number). From the end of March onwards, I started feeling jaded and played with far less energy and verve than earlier in the season and consequently lost 7 of my last 14 games (the same number I had lost in the previous 45).

So, yes, in sheer number crunching terms, the more you play, the less each individual result matters. However, the total number of games played is also likely to have an effect on the performance itself, both up to (helping to improve the performance because it keeps you sharp) and after a certain point (quite possibly impairing the performance because of tiredness). Obviously with great individual variations.

E Michael White
Posts: 1420
Joined: Fri Jun 01, 2007 6:31 pm

Re: Yet More Deflation

Post by E Michael White » Tue Sep 07, 2010 8:59 pm

Sean Hewitt wrote:E Michael White wrote:
I dont think its been mentioned before but Swiss events with an odd number of players also cause deflation and spread of grades, by small amounts which accumulate.

Hi Michael. That's interesting. Can you tell us why that is?
Hi Sean

The way I look at this is that in a swiss with an odd number of players, those that get the byes play one game less and are at the bottom end of the draw. If they played an extra game as in an even player swiss, they would also play someone at the bottom end. Either the player or their opponent would gain a significant amount on their TPR as the result % would increase. Changing 1 out of 4 to 1.5 out of 5 and leaving the average grade of opponents approximately the same changes their % from 25 to 30 giving a TPR increase of 5 ECF points. Nearly always these missed out results occur at the bottom end of the draw.

Its difficult to verify this numerically or analytically because the other elements of deflation/inflation can swamp it. I was only considering the built in effect of swiss events with byes allocated as they are. I think it would be a fair estimate to say the swiss effect is the same as considering the virtual grade changes on the same group of players assuming they all played the same number of games in a season before the event, as this removes activity based deflation/inflation.
Sean Hewitt wrote:E Michael White wrote:
Some baby boomers born 1945-1950 seem to be returning to play more chess; this is likely to have a short term inflationary effect and may make the current grading fix look more effective than it is in curing deflation.

Can you tell me why this is likely to be inflationary? I can't figure it out this early in the morning!
Here I was thinking that a group of players may start playing again; I have noticed some old names/faces reappearing. The new player process reintroduces players at a more realistic level on average although as someone posted recently apparent anomalies can occur. I guess that that group of players will gradually reduce in grade causing the grades of opponents to rise. If they are particularly active then this will magnify the effect. They don’t have to be very strong players just more active and gradually reducing in strength. Comparing this to the last 5 years, there were fewer players born in the period 1940-1945.
Roger de Coverly wrote:Could you expand this? We know that more active players have a higher average grade than less active players. An obvious conclusion is that if you play a lot, then you get better. I don't think that players playing 5 games a year have much impact on those playing 30 or much more

both

because you are less likely to meet them in an event ( because they don't play much)

and

Because the effect of one individual game is reduced for the more active players.
I think Jack’s explanation is correct. I would add that I believe that the average number of games played is less than 20 so the figures for distortion in one game could be an extra 1 to the winner and -5 to the loser.
Roger de Coverly wrote: More to the point is that the active players mostly play each other, so it doesn't have much effect on the most active players as to what grade the inactive players (5 a year) have.
The most important point is how many games are played where the opponent activity rates are significantly difference. Deflation/inflation emerges on a per game basis and accumulates over the years. Many players who play mostly in their local league support their local congress and not many others. If you as an active player play in those congresses you will meet them although mostly you might expect to meet active players. Last time I looked the average number of games played was just under 20, A graded players counted for only 20% and A,B and C accounted for 50-60%. Those figures suggest to me that there will be many games where a 16 game player plays against a 32 or more game player but without data this cannot be checked.
Brian Valentine wrote: The point is that if both players repeat their activity the next year Jack will contribute 100*.5 points = 50 into the rating pool to be redistributed by results, whereas his opponent will place 50*1 points =50 into the pool. Hence deflation explained by this idea only occurs if activity alters.
Except that we are assuming Jack improves so the reverse does not compensate for distortion.

Bear in mind to cause activity based deflation you do not need to be a very strong player just more active and improving. The link between activity and improvement is intuitively appealing. Some studies have been done to show that the effect is not great but in my view the wrong correlation was looked at. Although activity v player improvement is easier to understand the correlation should be based on games not players. ie the correlation needs to be evaluated between games played by active players versus improvement in strength of the players who played those games thus weighting by games.
Roger de Coverly wrote:Do you have an opinion about changes in skill standards? Take two players who were about equal graded in the 1970s. One retired then and has recently returned but hasn't updated their knowledge. The other has played continuously. Do you think they will both have about the same grade playing today?
If you had asked my that in 1965 I would have said a player could have expected to be initially about 30 points lower in playing strength unless they had previously been very strong ie 215 +. But now I am inclined to think about 15 points initially. I know there are players who have played continuously and dropped by similar or greater amounts but there are also players who are not much lower than before they stopped.

I know different players will record different experience but what seems to me to happen is that predominantly combinative players lose their skill quicker after age 55-60 which would coincide on average with reduction of short term memory skills. Whereas a predominantly positional player utilises more stable longer term memory.

Many say that the opening theory and preparation has advanced so much that a returning player doesn’t have a chance. But those reasons equally favour a returning player as it much easier to look at databases and play on the internet before making a return. In addition a player can wheel out 1960s theory and update it to the confusion of their opponents. Some of the lines I gave up as unsound 30 years ago I have discovered with Fritz 12 are playable, are not in the books and can lead to advantage.

Brian Valentine
Posts: 574
Joined: Fri Apr 03, 2009 1:30 pm

Re: Yet More Deflation

Post by Brian Valentine » Tue Sep 07, 2010 10:24 pm

E Michael White wrote:Brian Valentine wrote:
The point is that if both players repeat their activity the next year Jack will contribute 100*.5 points = 50 into the rating pool to be redistributed by results, whereas his opponent will place 50*1 points =50 into the pool. Hence deflation explained by this idea only occurs if activity alters.Except that we are assuming Jack improves so the reverse does not compensate for distortion.
Michael,
The assumption that Jack improves is new.

E Michael White
Posts: 1420
Joined: Fri Jun 01, 2007 6:31 pm

Re: Yet More Deflation

Post by E Michael White » Wed Sep 08, 2010 9:12 am

Brian Valentine wrote:The assumption that Jack improves is new.
Well not really. The ECF averaging system ensures that a more active player who is improving causes deflation. The conjecture is that greater activity is likely to accompany improvement, at least for players who have not yet reached their peak, which may be a local peak at any time during their playing days. Having peaked if a player reduces their activity enough and declines they will also cause defaltion. Some players seem to opt to play in fewer events if they feel they are not playing well, it may even be forced on them by a team captain, which brings about the second phase of the cycle. I have also noticed a correlation between forum postings on grading and colder weather.

Brian Valentine
Posts: 574
Joined: Fri Apr 03, 2009 1:30 pm

Re: Yet More Deflation

Post by Brian Valentine » Wed Sep 08, 2010 9:41 am

The argument was about Jack being stronger than average, not improving. The extra (not unreasonable) assumption that he is improving does changes the picture. However the discussion can then go on in two directions - is it the aim of the system that other things being equal, that the improvement should lead to an increase in average grades (as measured by the SCCU posting) or should the system maintain some sort of stability?

If the former then one of your other idea on older re-entries could mean that deflation represents the situation.

The mild deflation of the mean is not followed throughout the system: the grade of the 90th (top) percentile has gone up 2 points in the last two years, the median is down about 2 and the 10th percentile is down 6 points.

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Yet More Deflation

Post by Roger de Coverly » Wed Sep 08, 2010 10:30 am

Brian Valentine wrote:
The mild deflation of the mean is not followed throughout the system: the grade of the 90th (top) percentile has gone up 2 points in the last two years, the median is down about 2 and the 10th percentile is down 6 points.
That is consistent with the hypothesis that Mark, Keith, Simon etc. are about 115-130 points better than the "average" player and not, as the graders seem to believe, only about 95-110 points better. Having had their differences to the average eroded by last year's revaluation, they appear to be re-establishing them.

Talking of deflation, has it been noticed that it's getting increasingly difficult for a non-internationally rated player of around 175 standard to get on the FIDE list in the 2050-2150 range? It's obvious enough that this will happen because the international list now includes players of all standards, so it's no longer possible for an international rating to only measure your decent results against high rated players.

Sean Hewitt

Re: Yet More Deflation

Post by Sean Hewitt » Wed Sep 08, 2010 11:15 am

Roger de Coverly wrote: Talking of deflation, has it been noticed that it's getting increasingly difficult for a non-internationally rated player of around 175 standard to get on the FIDE list in the 2050-2150 range? It's obvious enough that this will happen because the international list now includes players of all standards, so it's no longer possible for an international rating to only measure your decent results against high rated players.
Isn't that because 175 ECF is equivalent to about 2050 FIDE? It ought to be pretty difficult to get on the international scale up to 100 points higher surely?

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Yet More Deflation

Post by Roger de Coverly » Wed Sep 08, 2010 11:38 am

Sean Hewitt wrote:
Isn't that because 175 ECF is equivalent to about 2050 FIDE? It ought to be pretty difficult to get on the international scale up to 100 points higher surely?
It used to be very feasible. My first international rating was 2130 in December 1993 at the time when my BCF grade from August 1993 was 176. Other 170/180 players of that era also got 2100 ratings

The first partial from an Open tournament in Italy was 1/5 against a GM, 2 IMs and 2 FMs. If I hadn't beaten the unrated 1800s on the pairing downbounce I wouldn't have "qualified" to play the titled players. The second was in the Isle of Man and included the then British champion. It did also include a female player who was at least 100 points too high from the Polgar revaluation and a couple of other English 170/180s who had also got ratings in the 2100s.

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Yet More Deflation

Post by Roger de Coverly » Wed Sep 08, 2010 4:44 pm

E Michael White wrote:Last time I looked the average number of games played was just under 20, A graded players counted for only 20% and A,B and C accounted for 50-60%. Those figures suggest to me that there will be many games where a 16 game player plays against a 32 or more game player but without data this cannot be checked.
There is a proxy for this - just count the number of games involving the categories of players.

So using the v2 of the 2010 list from the download and standard play only, I get

total number of half games 198997
of which
A category 90991 (44.2%) - average games per player 47
B category 39129 (19.0%) - average games per player 24
C category 35728 (17.4%) - average games per player 15
D category 22843 (11.1%) - average games per player 10
E category 9452 ( 4.6%) - average games per player 6
no category 854 ( 0.4%) - average games per player 4
unpublished 6725 ( 3.3%) - average games per player 3


no category is players who are in the published list for their rapidplay grade but don't have a published standard play grade. Unpublished is the difference between the game count from the download file and the game count from the SCCU website and deduces the unpublished player count from the SCCU website.

So that's A&B 63.2% and the others 36.8%.

The interpretation of this is that if you pick a game at random from the list of graded games, the chance of at least one of the players being A or B category is 63.2%. It's also saying E, no category or unpublished players feature in about 1 from 12 games. If you pick from a list of players , A & B are only 28%. So 28% of the players play 63% of the games.

Brian Valentine
Posts: 574
Joined: Fri Apr 03, 2009 1:30 pm

Re: Yet More Deflation

Post by Brian Valentine » Wed Sep 08, 2010 5:26 pm

Roger de Coverly wrote:unpublished 6725 ( 3.3%) - average games per player 3
This deduction is a good spot!

Although I don't agree the average. There appear to be 11,626 players on the list, so the remainder from the active total per SCCU (12,489) gives 863 players averaging 7.8 games, which is suspiciously high for all unpublisheds to be less than 9 to get into category E.

I'm not sure if the existence of these unpublisheds cause inflation or deflation. As they have no prior grade they should be predominately new juniors. As lower graded players, keeping them out of the list would increase the list mean. On the other hand they are sucking points out of the system so may cause deflation. Alternatively the junior calculation may just keep the balance just right. I'm still trying to get my head round this new revalation!

Ian Thompson
Posts: 3551
Joined: Wed Jul 02, 2008 4:31 pm
Location: Awbridge, Hampshire

Re: Yet More Deflation

Post by Ian Thompson » Wed Sep 08, 2010 5:37 pm

Brian Valentine wrote:Although I don't agree the average. There appear to be 11,626 players on the list, so the remainder from the active total per SCCU (12,489) gives 863 players averaging 7.8 games, which is suspiciously high for all unpublisheds to be less than 9 to get into category E.
Actual numbers from the latest list issued to graders (which is probably slightly different from the published list) are:

1. Players who played at least 1 standard play game last year, but don't have a published standard play grade - 2415 (playing 7528 games in total)

2. Of these, 230 have a published rapidplay grade (and they played 854 standard play games in total).

These are pretty close to Roger's figures.

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Yet More Deflation

Post by Roger de Coverly » Wed Sep 08, 2010 9:51 pm

Brian Valentine wrote:Although I don't agree the average. There appear to be 11,626 players on the list, so the remainder from the active total per SCCU (12,489) gives 863 players averaging 7.8 games, which is suspiciously high for all unpublisheds to be less than 9 to get into category E.
To reconcile the count.

The download file contains 11625 rows being 11624 data items plus a header.

My analysis shows

A 1937
B 1630
C 2453
D 2347
E 1704
no games 1323
games but no grade 230

The SCCU site gives
A 1938 B 1630 C 2453 D 2348 E 1705 adding to 10074. Presumably 3 publication declineds.
adding 2415 gives 12489

I presumed that the 230 were included in the 2415, therefore 2185 unpublished players with 6725 games, so about 3 each.

E Michael White
Posts: 1420
Joined: Fri Jun 01, 2007 6:31 pm

Re: Yet More Deflation

Post by E Michael White » Wed Sep 08, 2010 11:38 pm

Roger de Coverly wrote: So using the v2 of the 2010 list from the download and standard play only, I get

total number of half games 198997
of which
A category 90991 (44.2%) - average games per player 47
B category 39129 (19.0%) - average games per player 24
C category 35728 (17.4%) - average games per player 15
D category 22843 (11.1%) - average games per player 10
E category 9452 ( 4.6%) - average games per player 6
no category 854 ( 0.4%) - average games per player 4
unpublished 6725 ( 3.3%) - average games per player 3 -----

------So that's A&B 63.2% and the others 36.8%.

The interpretation of this is that if you pick a game at random from the list of graded games, the chance of at least one of the players being A or B category is 63.2%. It's also saying E, no category or unpublished players feature in about 1 from 12 games. If you pick from a list of players , A & B are only 28%. So 28% of the players play 63% of the games.
Thats interesting Roger; it gives some indication of which pairings are candidates for producing internal deflation/inflation arising from different activity rates and player improvement. It would be tempting to say that when an A player plays an A player that no Def/Inf will occur but in practice there is still plenty of scope. A players have game totals ranging from 30 to about 200 so two opponents could be significantly different.

B and C players should be treated as 30 games as the download shows games in the current season. In topping up B and Cs to 30 games in the grading formula, the games appear in more than one years calculations. Does this not cause a smaller amount of deflation but in more than one year ?

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Yet More Deflation

Post by Roger de Coverly » Wed Sep 08, 2010 11:43 pm

Brian Valentine wrote:
I'm not sure if the existence of these unpublisheds cause inflation or deflation. As they have no prior grade they should be predominately new juniors. As lower graded players, keeping them out of the list would increase the list mean.
Not necessarily. They can also be "foreign" players ( this concept includes Welsh, Scottish and Irish ) playing less than 9 games. You need 2 weekend tournaments to qualify for an ECF grade. You should also note that the world's highest internationally rated player is presumably also sitting on a 7 game partial ECF grade from last year's London Chess Classic.
Last edited by Roger de Coverly on Thu Sep 09, 2010 12:37 am, edited 1 time in total.

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Yet More Deflation

Post by Roger de Coverly » Wed Sep 08, 2010 11:59 pm

E Michael White wrote:B and C players should be treated as 30 games as the download shows games in the current season. In topping up B and Cs to 30 games in the grading formula, the games appear in more than one years calculations. Does this not cause a smaller amount of deflation but in more than one year ?
Expressed another way, that says for 80% of the games, the maximum contribution of a win v draw result to a grade is 1.67 (50/30). Alternatively if a grade is a hypothetical estimate of strength plus or minus an error function, general reasoning suggests that the error function can be large for players with a small number of games, but it probably doesn't matter because not many games are affected.