Included players and games

The current statistics  include players rated 2500 or higher (in the fighter rating list 2600 or higher), in games against opponents rated 2400+. The historic statistics for the period since 1972  include players rated 2400+ in games against opponents rated 2300+, the women's statistics the current top-30 (04/2004 FIDE rating list) plus some selected players, in games against opponents rated 2000+ . The minimum number of games for inclusion is 50 in most cases, with the following exceptions:

Annual  statistics for classical games:         20      (10 for current year)
Overall statistics for rapid games:                20
Annual statistics for rapid games:                10
Overall statistics for blitz games:                 10
Historic statistics up to 1950:                        20

The main sources for the games are The Week In Chess and ChessCollect.

Fighter Rating

The "Fighter Rating" is based on a "Short Draw Factor" (SDF), which is an adaptation of the "Chicken Factor" conceived by Mig Greengard (check out his site ChessNinja and his Mig on Chess articles on ChessBase; there's a definition of the original Chicken Factor in Mig on Chess #116.). The more often a player agrees to short draws, the higher this factor will be. For my list, the SDF is calculated as follows:

In the final position of each drawn game, the value of each player's material is counted (with the usual scale of Q=9, R=5, B=3, N=3, P=1). The number of moves is then subtracted. For example, in a draw in 20 moves with two pawns and one knight exchanged, the result is 34 (1*9 [Q]+2*5 [R]+2*3 [B]+1*3 [N]+6*1 [P])-20=14.  If less than 45 half-moves have been played, there's an additional penalty of 3 points per half-move under 45 (but not more than 45 points). So in this case (draw in 20 moves=40 half-moves) this penalty is 15, making the total factor 29. The Elo difference between the players (with an addition of 50 points to the rating of white) is divided by 8, then added to the factor for white and subtracted for black. If white is rated 2700 and black 2670, this means a (2700+50-2670)/8=10 points bonus for black and penalty for white, adding up to a final result of 39 for white and 19 for black.  A negative factor (corresponding to a "fighting" draw) is neglected, while positive factors are summed up for each player and divided by the overall number of games to obtain the average SDF.

Playing Style

The "Sacrifice Percentage" is the percentage of games a player won after being down on material for at least six consecutive half-moves (disregarding exchanges during that period), relative to that player's overall number of games.

For the "Opening Draw Factor", an average precentage of draws is calculated for each each opening variation  (according to ECO from A00 to E99, you can find this list under opening statistics.). Then for each player the average draw percentage of the used openings is calculated.

The endgame performance is the Elo performance for all games that lasted for 50 moves or longer, in which no player was up on material by more than a pawn at move 40.

Improved Elo Rating

One of the problems of the Elo system used for the official FIDE ratings is that the ratings are rather static for someone who plays infrequently. Thus, I am using a more flexible method in my ratings. The k-factor (the constant that determines how quickly the ratings change, which is 10 in the FIDE ratings) is calculated individually for each player, depending on the number of recent games. The following formula turned out to optimize the predictions of future results based on the ratings:

k = 120/gm,

with gm being equal the number of games played in the last two years, but not larger than 150 or smaller than 70. This means k can vary between 8 and 17.1.

Note that my rating list is based on the game downloads  from TWIC, which are bound to contain less games than are included in the FIDE ratings. The game database should be almost complete for the top-15 or so, for other players my ratings may  not be completely accurate in case some results are missing.

Rapid Chess Rating

The rating is calculated according to the Elo formula, but with a k-factor of 20 instead of 10, making the rating twice as "dynamic" to account for the smaller number of games. It uses games played since 2000, both at rapid chess tournaments and in tiebreaks of classical events. As an initial value, the FIDE rating at the time of a player's first rated rapid game is used. All players with at least 20 rapid games against 2400+ rated opposition since 2000 are included.

Back to main page