Grades Grades are used within chess to help determine fair and predictable pairings in competitions, to measure progress, and as one indicator to team managers that a player might be asked to play for a county team. Note that this page was updated on 13/10/2009 and reflects the significant changes to the grade calculation methods introduced at the start of the previous season (2008-09).
Berkshire Junior Chess have been running our own grading system for several seasons. Player rankings, based on these Berkshire Junior grades are also published in the Year Book.
Grading systems This website displays 4 different grades for players, and for event results. The grade types are: Berkshire Junior, FIDE, ECF Standard Play, and ECF Rapidplay (ECF = English Chess Federation). On a player detail page, all four grades may be displayed if the player has gained a grade of each type. A results page will only display grades of one type.
ECF Grades ECF Standard Play grades are assigned to players if they have played matches where the time limit is not less than one hour for the initial 30 moves (and, of course, where the match result has been submitted to the ECF for grading purposes). A Standard Play grade can be gained by playing in certain county matches (e.g. under 18 matches), in the adult Berkshire League, or in some tournaments such as the Berks and Bucks Congress. At the start of the 2009-10 season, 32 Berkshire Juniors had a standard play grade.
ECF Rapidplay grades are calculated from the results of games of less than one hour for the initial time period, but no faster than 15 minutes sudden death. There are currently 45 Berkshire Juniors with a rapidplay grade.
Berkshire Junior Grades As of September, 2004, the ECF (BCF as was at the time) no longer grade junior events for free. It was estimated that to carry on submitting results at the same rate as the 2003-04 season would cost us £800! Instead of increasing tournament entry fees to cover ECF grading charges, we decided to calculate our own grades, and call them "Berkshire Junior Grades". Further, we decided to adopt a modified version the FIDE grade calcuation methods, which are different from the methods used for ECF grades.
At the start of the 2009-10 season, there were 1,198 players with a Berkshire Junior Grade (453 are active, having recorded some sort of result in the last year or so). The grades have become the standard measure used to pair players at the popular Saturday afternoon events. We publish end-of-season ranking lists based on these grades, but only team managers can access the current point-in-time rankings.
Grade Calculation The nitty-gritty part. How are tournament results turned in to Berkshire Junior grades? First, all match results for a tournament are entered into the website database (this level of detail also allows the website to display a "crosstab" section on a results page). The actual calculations take place in three steps:
- Calculate new grades for graded players:
- For players who already have a grade, the average of their opponents' grades are used to calculate an expected score. Note that results against ungraded players are not included at this stage.
- If the opponent's grade is more than 350 points higher or lower than the player's grade (at the start of the event), then it is adjusted such that the difference is exactly 350, before the averaging calculation is done. This prevents players from making large grade gains by losing to a very highly graded opponent within a set of otherwise roughly "expected" results. However it can also result in an already highly graded player gaining further grade points by beating many comparitively very weak opponents.
- The difference between a player's expected score, and the actual score obtained against graded opponents is then used to calculate an increase or decrease in the player's grade.
- The player's grade is adjusted in proportion to the difference between their expected and actual scores. This algorithm is similar to the one used by FIDE.
- This new grade which is considered more accurate than the starting grade is only used for the following step.
- There must be at least 5 graded players in a tournament in order to estimate initial grades for ungraded players. If this minimum condition is met, a straight line best fit between grade and points scored is calculated using the newly calculated graded player grades from step 1. A simple least squares linear regression is used initially, but if the highest or lowest grades are far from this line, they are removed from the value set, and the regression is re-calculated. The minimum requirement for 5 graded players does not include anyone removed from the value set as described.
- The straight line is then used to estimate initial grades for all ungraded players. No grade is calculated is the player scores 0 points (byes do not count toward the score for grading purposes).
- The calculation as described above for graded players in step 1 is now performed for both graded and ungraded players. For the former, their starting grade is used (not the grades calculated in step 1, which were for use in step 2 only); for the latter, their estimated initial grade is used.
- Unlike in step 1, results between graded and ungraded players are included. This is a change from the initial method used as it was found that low-graded, but still reasonable, players got no benefit from winning against ungraded players, but lost points for losing to those above them.
- The minimum published grade for any player is 300. If the calculation for an ungraded player yields a value below 300, no grade is published - if the player enters another tournament, it will be as an ungraded player. If a newly calculated grade for an existing graded player drops below 300, it is set to 300. Note however that estimated initial grades below 300 are still used when calculating grade changes for others. You can see these sub-300 grades if you hover over a crosstab entry which includes a player who did not reach the 300 threshold at the end of the event.
- A tournament bonus may be added to all entrants' grades. This is intended to reflect an expected improvement in playing standard simply through the practice gained by competitive play. The bonus varies between event types and also varies between seasons. As an example, it is currently set at 5 for Saturday afternoon events, but it is larger for the more challenging events such as the Berkshire Junior Congress.
Some general comments on the overall system:
- Grades change continuously throughout the year, after each event for which the above calcuations are performed. Since young players often improve rapidly, this system gives them some immediate feedback on how well they are doing.
- The system aims to have a median grade of 500 within the active playing population for any given season. To achieve this, the event bonuses can vary from one season to the next, and there is also the capability of having an end-of-season bonus (which is awarded at the first event of the next season as a personal bonus). For 2008-09, there was a 10 point adjustment to "correct" the below target grade average at the end of 2007-08. For 2009-10, there is no season bonus, and the event bonuses have been reduced. It wasn't so much that we over-compensated the previous season, it was that the change to including results against ungraded players had a larger impact than was anticipated.
- Although grades are calculated and published as soon as results become available, they are subject to change if results later arrive for an earlier event (grade calculations are strictly chronological), or if a player is not recognized when loading the results into the database (usually through a different spelling of their name), resulting in them being treated as a new ungraded player, rather than an existing graded one.
- Grades are only calculated for results at Berkshire events. Although the database includes results for "external" events, these are not used for Berkshire grading purposes. On the other hand, some non-Berkshire players have Berkshire grades if they play in Berkshire events on a regular basis.
- For the two Junior Congresses, only results between Berkshire players are used - visitors are not graded.
- It is quite possible for an event to be included in both the Berkshire and ECF grading systems. The two Congresses are an obvious example.
- Berkshire Junior grades are not intended to be the same as, or even to tend to approach over time, either FIDE grades, or ECF grades (after applying a conversion function). They are useful only in comparing Berkshire players.
- Grades are most "accurate" for those players who have taken part in several events. Indeed, the end-of-season ranking lists require at least three results for a player to be included. Initial grades in particular have been known to be somewhat inaccurate in the past, although this is becoming less common.
The actual formulae used to perform the "calculations" mentioned above are given below. (note: * is used for multiplication, and ** is the exponent (to the power of) symbol):
- For a player with no Berkshire Junior grade, but with a ECF rapidplay grade:
BG = (RPG * 8) + 600
where BG = Berkshire Junior grade, and RPG = Rapidplay grade.
- Estimated Score:
ES = GM / (1 + 10**( (AOG - BJG) / 400 ) )
where GM=graded matches; ES = expected score; BJG = Berkshire Junior grade; AOG = average opponent grade;
- Change of grade:
GC = K * (AS - ES)
where GC = grade change; K = 40 (the progression coefficient); AS = actual score (against graded opponents); ES = expected score.
For players who are ungraded at the start of an event, a value of 50 is used for K.