Effect of activity on grade changes

[Christopher Kreuzer]

Something I've been curious about, ever since I started (a few years ago) to track changes to my ECF grade more closely during the season, is what effect extremes of activity have on changes in a player's grade.

I'm basing the following on what is stated here:

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

"It [the grading list] includes all players who have had nine or more games recorded on the system in the last three seasons, including at least one in the most recent season. Players not meeting this requirement are ungraded, and do not appear."

The activity levels are:

A - Grade based on 30 or more games in the latest season
B - Grade based on 30 games in the two latest seasons, of which at least 20 were played in the latest season
C - Grade based on 30 games in the latest three seasons, of which at least 10 were played in the latest season
D - Grade based on 15 or more games in the latest three seasons, of which at least 5 were in the latest season
E - Grade based on 9 or more games in the latest three seasons, of which at least 1 was played in the latest season

I've been active enough over the past six grading lists (2005-2010) to have a "A" grade each time (30 or more games), and have at times gone quite a way over that. What I've been wondering is whether playing a large number of games (e.g. 90 games, though this is nowhere near the records) distorts things somewhat?

The effect seems to be that the more games you play in a season (assuming you play a normal range of grades), the less effect games later in the season have on your new grade the next season (it is either somewhat comforting or dispiriting to sit down to the 100th game in a season knowing that the result will affect your new grade by 0.5 grading points, YMMV depending on which way your grade is heading). Another point is that once you pass the figure of 30 games, then all results from previous seasons count for nothing, so if you have a bad (or good) season while only playing 10 games, that can lead to a drastic fall (or rise) in your grade.

Another scenario (for 60+ games) is that you play well in the first 30 games, and then play badly in the last 30 games. The final grade will not indicate your current form, but rather the form over the entire season (because that is what it is designed to do). Talking of form over a season, has anyone noticed whether they play better in different periods of the season? Better at the start or end of the season? Before or after Xmas? Better in league chess than in congresses? Better on different nights of the week?

I suppose it is also possible to calculate what your grade would be for low levels of activity (based on the results of the previous season) if you could be bothered. Do those who have only played nine games over three seasons, really get accurate grades or not - I can think of scenarios where such grades could be very inaccurate. For example, if a player plays just one game in the current season, regardless of how active they have been, their next grade would be based on that game and the last nine games of the previous season (imagine their new grade if they happened to have won all those games)? The effect seems to be if you go inactive on a "high" (winning a few games), then you can get quite a good grade when you return to playing a few years later, but if you go out on a "low" (losing a few games at the end of the season), then the opposite seems to happen (for the first grade, anyway, for later grades it should even up again).

EDIT: I think I see the mistake I made here. Since games are not all dated within a season, the "X or more games" is calculated from games in the current season, plus *all* games from the previous season, and if that doesn't bring the total above 9, 15, or 30 (depending on how many games were played in the current season), then you keep adding entire numnbers of games on from earlier seasons, back for three seasons. So no "part season" ratings carrying forward, it's the whole season carrying forward or nothing, I presume, when bringing results forward to aid calculations.

One historical question I have is how long the "30 games" bit has been in the grading calculations for and whether it is based on the median activity across the entire grading list or not? I do recall seeing something published about this somewhere, but can't remember where. And has there ever been considerations for the very active players of only considering the "last 30 games"? I suppose not, because some games are from leagues and thus you can't (always) date them in order to get a last 30 games.

The main point I wanted to raise though is that for players playing over 50 games a season, the fluctuations in form don't seem to be reflected in the grading system (and indeed it can't because of the problem of dating some league games and other grading oddities). Should a grade reflect current form rather than form over a season?

[Sean Hewitt]

All true Christopher. If you want a system that is more reflective of current form, it's called ELO!!

[Peter Rhodes]

I agree with your sentiments Christopher, ELO is a far superior system used around the entire world.

I don't understand the full details but we seem to have a team of graders who know better than the rest of the world. That's why we have this peverse system in place.

[Roger de Coverly]

Should a grade reflect current form rather than form over a season?

I would argue that two quantities are being measured or estimated. One is an estimate of player strength and the other is current performance. Player strength could be defined as "if A is higher in the ratings than B, then we expect A to score more points against a set of opponents than B"

Elo style systems work by measuring the performance in a recent game or games and using it to adjust the rating brought forward. The ECF system works by calculating performance over a time period (a year or more) and using it as the estimate of strength for the next period.

Arguably a more sophisticated method of rating would supply two numbers, a slowly changing estimate of player strength and a measurement of performance over a period. Depending on the period chosen, the performance measured may or not be volatile.

The higher levels of international titles (IM/GM) already require the player to demonstrate IM/GM performance over a discrete number of events(norm requirements) and also to elevate his or her rating to the required minimum.

The 30 game number in the ECF system dates from the origins of the system in the fifties. It can be derived as a standard statistical result - the workings are on this forum. It's the number of games you need to be confident in a statistical sense that A's grade exceeding B's by 8 points is enough to assert that A should get better results than B against the same field of players.

[Richard Thursby]

Judging from the information supplied on my ECF grading printout, the way games from previous seasons are incorporated into your current grade has changed recently. If you play fewer than thirty games in a season, they used to make it up to 30 by taking an average of games played, but they now use the latest games played in a season. To give an example, if you play 20 games in each of two consecutive seasons, it used to be that the new grade would be

((points in games this season) +(points in games in previous season)*10/20)/30

Whereas now it is
((points in games this season)+(points in last 10 games of previous season))/30

[Just for clarification, by points I mean, for each game you get your opponents grade (assumed to be 40 different from your own if it is more than 40) plus or minus 50 depending on if you win or lose. Naturally the final value is rounded to an integer. No doubt grading people who inhabit this forum can point out any errors in what I have said.]

Some years back I used the latest games to calculate my grade and got the wrong answer. Having queried this with the grader, I checked the grading list, realised the mistake and then got the correct answer.

This change appears to have coincided with the introduction of the new grades, since in my 2008 grade an average is used and in 2009 the latest games are used.

[Christopher Kreuzer]

Judging from the information supplied on my ECF grading printout, the way games from previous seasons are incorporated into your current grade has changed recently. If you play fewer than thirty games in a season, they used to make it up to 30 by taking an average of games played, but they now use the latest games played in a season. To give an example, if you play 20 games in each of two consecutive seasons, it used to be that the new grade would be

((points in games this season) +(points in games in previous season)*10/20)/30

Whereas now it is
((points in games this season)+(points in last 10 games of previous season))/30

[Just for clarification, by points I mean, for each game you get your opponents grade (assumed to be 40 different from your own if it is more than 40) plus or minus 50 depending on if you win or lose. Naturally the final value is rounded to an integer. No doubt grading people who inhabit this forum can point out any errors in what I have said.]

Some years back I used the latest games to calculate my grade and got the wrong answer. Having queried this with the grader, I checked the grading list, realised the mistake and then got the correct answer.

This change appears to have coincided with the introduction of the new grades, since in my 2008 grade an average is used and in 2009 the latest games are used.

Oh. If this is true, then it is now possible to play (and win) a single game in one season, and add that to the nine games you won in a row at the end of the previous season (it could happen) and shoot up the grading list (even if it is "only" an 'E' grade). Hopefully those who calculate the grades will reveal how it is done (and note that there is a line saying that large increases in grades are looked at, but I don't think it applies here, as the converse could also happen - you lose your last ten games in a row and due to inactivity meaning only those games are counted, sink way down the grading list). My caveat about dating the games played still applies, though. I am not convinced that all leagues report their games in such as way to allow such dating, but if they now do, that is good!

[Christopher Kreuzer]

I agree with your sentiments Christopher, ELO is a far superior system used around the entire world.

I don't understand the full details but we seem to have a team of graders who know better than the rest of the world. That's why we have this peverse system in place.

I find an advantage of the ECF system over the ELO system is that I can actually work out my grade change under the ECF system. I've tried using online ELO calculators to work out my FIDE grading changes, and it seems my latest FIDE rating change did approximate to assuming everyone's grade stayed the same during the Sunningdale Major I played in at the beginning of September. But I had assumed that to be accurate, I needed to calculate the changes in my opponents grade between rounds (and hence the change in their opponents grades, and so on). From the tournament cross-table, could I have worked out my change in FIDE rating easily and quickly?

http://www.e2e4.org.uk/sunningdale/Sept2010/index.htm
http://ratings.fide.com/card.phtml?event=470589
http://ratings.fide.com/calculator_rtd.phtml

Rating: 1910
TPR: 2205
W-We total: 1.55

I assume W-We is the critical figure, and that I use that together with the "k" value of 15 to calculate the change (my calculations estimated that the change would take me to from 1910 to 1933, which turns out to have been correct).

The next thing I want to try and work out is whether my rapidplay grade (currently 148 E from the July 2010 list) will overtake my slowplay grade for the first time ever! I played 18 games in the recent 4NCL rapidplay for a performance of 159, so I'm guessing I need to either work out how many rapidplay games I played last year (one, I think, clearly less than 5) and possibly the year before that (not a clue, but it should be on the grading printout that I have somewhere and should be attached to an e-mail somewhere as well) to work out my new rapidplay grade. Maybe I should just play another 12 rapidplay games before the January 2011 rapidplay grading list comes out to get an X grade (30 games in past half-list) - I presume any events before Xmas will count for the January 2011 list.

I also want to try and sort out the conversions from FIDE to ECF, as I've heard that FIDE ratings are now slightly different from ELO ratings (at least at my level), so the conversion figures are a bit confusing.

At the bottom of this page:

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

It gives what looks like updated conversion figures, so with ECF 160 being equivalent to FIDE 1930, it looks like my FIDE (1933) and ECF (159) ratings agree somewhat again (which is rarely the case, I've heard). The next thing is to work out how to win these games I keep losing or drawing. Maybe not calculating rating changes will help!

[Roger de Coverly]

But I had assumed that to be accurate, I needed to calculate the changes in my opponents grade between rounds (and hence the change in their opponents grades, and so on).

Why assume complexity where none exists? I agree that on line servers will recalculate the rating after every game but the international Elo system locks the rating for calculation purposes at the published rating designated for the date of the tournament. Most tournament cross table software in common use will display the W-We value round by round - the only caveat is to check whether it's been updated to move the 350 cutoff to 400. If you want to do it in your head, you could memorise the factors in the FIDE rating regs alternatively remember that if you play someone 200 points higher your score expectation is 0.24 and if you play someone 400 points higher it's 0.11. You can interpolate between these for an approximate result for the actual rating.

(edit) and to state the obvious it's 0.50 for someone of your exact rating.

Link to FIDE rating system description and factors
http://www.fide.com/fide/handbook.html? ... ew=article

[Roger de Coverly]

Rating: 1910
TPR: 2205
W-We total: 1.55

I assume W-We is the critical figure, and that I use that together with the "k" value of 15 to calculate the change (my calculations estimated that the change would take me to from 1910 to 1933, which turns out to have been correct).

That's the precise calculation. You multiply the (W-We) by your personal K factor (10/15/25) to get the rating change. So you can do it in your head provided you can multiple 15 by 15.5. So it's 10% of 15 squared (225) rounded up a bit - so 23.

[Christopher Kreuzer]

the international Elo system locks the rating for calculation purposes at the published rating designated for the date of the tournament.

Thanks. Is it the same for leagues? For example, will the 4NCL use ratings from the November 2010 list or the September 2010 list? I would assume the former, as it doesn't start until November, but teams will have been registered when only the September 2010 list was available.

Most tournament cross table software in common use will display the W-We value round by round - the only caveat is to check whether it's been updated to move the 350 cutoff to 400. If you want to do it in your head, you could memorise the factors in the FIDE rating regs alternatively remember that if you play someone 200 points higher your score expectation is 0.24 and if you play someone 400 points higher it's 0.11. You can interpolate between these for an approximate result for the actual rating.

(edit) and to state the obvious it's 0.50 for someone of your exact rating.

Link to FIDE rating system description and factors
http://www.fide.com/fide/handbook.html? ... ew=article

Ooh, thanks, but I want to be able to concentrate on the chess without getting too absorbed in the intricacies of the rating system...

[IM Jack Rudd]

Leagues can choose one of two options: either they use the ratings at the date the season started, or they use the ratings in force when the game was played.

[Roger de Coverly]

Thanks. Is it the same for leagues? For example, will the 4NCL use ratings from the November 2010 list or the September 2010 list? I would assume the former, as it doesn't start until November, but teams will have been registered when only the September 2010 list was available.

The 4NCL seems to use a different system every season. The third alternative beyond November 2010 is to use the list when the games are played. The captains agreed last season that board orders would be based on the rating list from time to time. Whether ratings for calculation purposes will be based on September or November, I wouldn't know.

[Christopher Kreuzer]

Thanks. Is it the same for leagues? For example, will the 4NCL use ratings from the November 2010 list or the September 2010 list? I would assume the former, as it doesn't start until November, but teams will have been registered when only the September 2010 list was available.

The 4NCL seems to use a different system every season. The third alternative beyond November 2010 is to use the list when the games are played. The captains agreed last season that board orders would be based on the rating list from time to time. Whether ratings for calculation purposes will be based on September or November, I wouldn't know.

Oh, so it is wait and see, then (I suppose I could ask in the other thread I started). I take it "from time to time" has a specific meaning here?

[Roger de Coverly]

I take it "from time to time" has a specific meaning here?

On the precedent of last season, each weekend will have a rating list associated with it for board order purposes. The May weekend last season used the 1st March list rather than the 1st May list so that match captains could determine team composition and board orders in advance without having to worry about late rating changes.

[Peter Rhodes]

I find an advantage of the ECF system over the ELO system is that I can actually work out my grade change under the ECF system. I've tried using online ELO calculators to work out my FIDE grading changes...

Why not just calculate it yourself - this is the formula :

=((((10^((DIFF)/400))+1)^-1)+RES-1)*2*KFACTOR

where DIFF is the difference between your grade and your opponents.
RES is the Result (WIN=1, DRAW=0.5, LOSS=0)
KFACTOR - consult FIDE rules, but I think 16 is still used.


I keep a spreadsheet of all my results. If you want me to send you a copy, send me a PM with your email.