Friday, March 21, 2008

USCL Ratings (with 2007 Final List)

I have independently created a United States Chess League rating system that takes into several elements that other rating systems do not. Most importantly, the USCL ratings take into account that players are part of a team, and their strategy about how to conclude their individual games depend on the games situations of their teammates.

Here is an overview of the USCL rating system.

1. Modified Glicko

Rating calculations are based on the Glicko system, with some minor modifications. These modifications are as follows:
(a) Maximum RD is 100 (not 350);
(b) Ratings are updated after every game;
(c) At the end of each season, each RD is modified by increasing it 1/2 its distance to 100. For example, if a player's RD is 40, then the new RD is 40+((100-40)/2) or 40+(30) or 70. This eliminates step 1b in the Glicko process.

2. Initial Ratings Based on Board

Initial ratings for each player are based on which board they played the first time they played in a USCL match. Players who were Board 1 start with an initial rating of 2550; Board 2, 2450; Board 3, 2350; Board 4, 2250. All are set with an RD (ratings deviation) of 100.

3. Expected Score Depends on Color

Expected score from each game is modified by the color each player has. An initial investigation showed that, in the USCL, the performance rating of a player with the White pieces is about 72 points higher than the performance rating of a player with the Black pieces. Therefore, when calculating expected score of a games between two players, we temporarily add 36 points to White's rating, and subtract 36 points from Black's rating.

4. All Games Rated with Equal Weight

All games, including playoff games and blitz tiebreak games, were given equal weight in the ratings.

5. Scores for Draws Depend on Team Result

In team chess, where the goal of the team is to win the match (with 2.5/4 points or more, or 2/4 in a match with draw odds), individual results can be skewed by the circumstances of the match situation. For example, if it looks like your team is about to win two games and lose one game, winning the match comes down to what happens on your board. Suppose that in your game you have a moderate advantage, but the position is dynamic and unstable. In such a case, you might decide to trade all your pieces to reach a dead drawn endgame, so that you will draw the game and clinch the match for your team. This is fundamentally a good result for your team, as your will win the match.

Therefore, the result of the team match is considered when assigning scores for individual players' draws. If the final team match is drawn, each player who drew gets 1/2 point. If a team wins, all the players on the winning team who drew their games get 2/3 point each. Conversely, all the players on a losing team who drew their games get 1/3 point each. (Wins are still 1 and losses are still 0, regardless of the team match score.) Note that these point assignments are only for rating calculations only.

Final 2007 United States Chess League Rating List

  • Every game in every season of the USCL was rated, not just 2007.
  • Only players who were played at least one game in 2007 (i.e., active) are included in the list.
  • Players must have played at least three games in the history of the USCL to be included in the list.
  • While every effort has been made to ensure accuracy, re-rating might occur to correct any mistakes.
1 2591 Sergey Kudrin
2 2589 Vinay Bhat
3 2586 Patrick Wolff
4 2581 Larry Christiansen
5 2580 Hikaru Nakamura
6 2579 Jorge Sammour-Hasbun
7 2576 Joel Benjamin
8 2571 Drasko Boskovic
9 2569 Gregory Serper
10 2566 Julio Becerra
11 2543 Davorin Kuljasevic
12 2523 Jacek Stopa
13 2522 Eli Vovsha
14 2519 John Donaldson
15 2505 Pawel Blehm
16 2495 Josh Friedel
17 2494 Alex Stripunsky
18 2491 Lev Milman
19 2487 Georgi Orlov
20 2487 Dean Ippolito
21 2484 Pascal Charbonneau
22 2481 Eugene Perelshteyn
23 2477 Irina Krush
24 2466 Jay Bonin
25 2463 Vince McCambridge
26 2460 Bryan Smith
27 2454 Tegshsuren Enkhbat
28 2453 Dmitry Schneider
29 2449 Denis Shmelov
30 2441 Eric Tangborn
31 2440 Ron Burnett
32 2433 Marcel Milat
33 2432 Slava Mikhailuk
34 2428 Sam Shankland
35 2423 Oleg Zaikov
36 2415 John Bartholomew
37 2415 Robert Hess
38 2411 Richard Costigan
39 2407 Dmitry Zilberstein
40 2406 Jonathan Schroer
41 2405 Keaton Kiewra
42 2402 Irina Zenyuk
43 2399 William Kelleher
44 2397 Blas Lugo
45 2389 David Pruess
46 2389 Francisco Guadalupe II
47 2388 John Readey
48 2387 Andrei Zaremba
49 2386 Aviv Friedman
50 2382 Mikhail Zlotnikov
51 2381 Marcel Martinez
52 2380 Todd Andrews
53 2366 Parker Zhao
54 2365 Elvin Wilson
55 2361 Bruci Lopez
56 2360 Chris Williams
57 2358 Katerina Rohonyan
58 2354 Larry Kaufman
59 2354 Vadim Martirosov
60 2353 Daniel Yeager
61 2343 John Rouleau
62 2334 Mackenzie Molner
63 2333 Eric Rodriguez
64 2327 Bayaraa Zorigt
65 2322 Victor Shen
66 2319 Miguel Espino
67 2314 James Critelli
68 2311 Marc Arnold
69 2309 Michael Thaler
70 2307 Peter Bierkens
71 2304 Luis Barredo
72 2303 Loren Schmidt
73 2301 Evan Ju
74 2301 Gregory Young
75 2300 Matthew Herman
76 2297 Alejandro Moreno Roman
77 2295 Craig Jones
78 2287 Michael Lee
79 2279 John Timmel
80 2276 Ralph Zimmer
81 2265 Daniel Naroditsky
82 2264 Udayan Bapat
83 2263 Ilya Krasik
84 2261 Tsagaan Battsetseg
85 2258 Josh Sinanan
86 2239 Matthew Bengtson
87 2228 Peter Bereolos
88 2211 Jerry Wheeler
89 2209 Gerald Larson
90 2183 James Wu

NOTE: While the commissioner of the United States Chess League is aware of the existence of these ratings and the rating system, neither the ratings nor the rating system have been officially endorsed nor sanctioned by the Unites States Chess League, its affiliates, or sponsors.


Arun Sharma said...

Interesting system though I must say it wouldn't be near the top of my choices to use if we were going to modify the way we did it. While your system makes a lot of sense statistically, I really don't like any MVP type system which makes it completely impossible for the lower boards to be at the very top.

This is something I've argued with Greg about as well in our MVP system as it seems very unlikely for a Board 3 or Board 4 to ever win league MVP under our system either (which is the reason I made him change the Board 3 from 2 points to 2.5 points this year, but it's obviously still pretty unlikely for a Board 3 to win). To a point I think that's good, as the top superstars should generally be the ones to win an award of that sort, but certainly someone who goes like 9-1 on Board 4 probably is the most deserving person and yet they likely still wouldn't win (that score wouldn't have been enough to win this year anyway). Of course that's a matter of opinion as to whether they should, and my feeling that they should could have something to do with my personal prejudice since the Board 4 players are the league players closest to myself in strength (gotta stick up for the family!).

Bionic Lime said...

I'm assuming you're referring to the way of initial estimated ratings based on board played. I equated this with the initial assignment of rating by age in the USCF system. However, it will all work out in time as the players play one another. Note that this inital rating is done once, prior to the player's first USCL game.

That being said, I agree with you about the MVP scoring system being too biased toward to top boards, and I mentioned to Greg too -- I'm not sure what the solution is.

In any case, I'm not proposing the USCL rating system as a way of doing MVP voting or assessment, although I'm sure you could come up with ways of using it.