Are Washington State Players Really This Good?
Sup chess fans,
In my chess career, I've played chess in two main places: my hometown in the Bay Area California and Seattle, Washington where I moved up here to work for the last couple of years.
When I moved up here, I thought the chess scene would be weaker than that of the Bay Area. But I was sorely mistaken as now I firmly believe that Seattle chess players are significantly stronger than those of any other area I have ever played. In fact, I think that if we pitted a 2200 rated junior from Seattle against a 2200 junior from California, I would bet money that the 2200 rated junior from Seattle would win.
And I don't think this is another case of juniors being underrated. I recently discovered a new metric for measuring the strength of chess player called "Average Centipawn Loss" which I'll explain below. When I ran this metric against all of my 2000+ rated opponents in the last few months, I was shocked to find how accurately these Seattle juniors are playing against me. And it isn't just a one-off kind of fluke; these juniors are playing IM-GM level chess against me around half the time! No wonder I feel like I'm facing serious resistance every game here; these kids are actually playing phenomenal chess.
The New Evaluation Method: Average Centipawn Loss
Traditionally the main way we assess someone's strength in chess is based on elo ratings where our opponents are assigned ratings, and based on our performance against them, we can determine our own elo. For example, if you play 4 1800's, and you win 2 games and lose 2 games, that means you scored 50% against average opponent 1800, and therefore your rating should be 1800 too.
There is this new metric that is starting to gain traction that is not based on performance against other people, but rather on how your moves compare to a computer's. By calculating how off our moves are compared to the computer's, we can compute an average for how far our moves deviate from the best computer moves. And we calculate this value based on the "centipawn" value, which is the same as the computer's evaluation. This metric is called "Average Centipawn Loss" or ACPL for short.
So as an example, let's say a computer evaluates a position as +0.5 if I play its top choice. Instead though, I play the second best move, which results in a 0.4 computer evaluation. The difference in evaluation is 0.1, which is 10 centipawns, and assuming the value of a +1.00 is equal to 100 centipawns, 10 centipawns means I was off by a tenth of a pawn that move. So let's say over a 30 move game, I play a move that drops 100 centipawns every 10 moves. That means 300 centipawns/30 moves = 10 centipawns/move would be my ACPL for that game.
It turns out that these averages correlate almost perfectly with real elo ratings. The higher your elo, the lower your ACPL score is, as we would expect. If I had to categorize the average centipawn loss (ACPL from now on) to a certain level of play, I would categorize it like this:
ACPL Level Of Play
0-5 Elite GM
40-50 A player
50-60 B player
and so on.
You can see a graph of the direct correlation over 50000 games here: http://chess-db.com/public/research/qualityofplay.html
Note that even though on the graph for ACPL to ELO correlation says 10 ACPL is in the high 2400's level, I'm going to include any score between 6-10 as GM level because it is possible to play a game with 0 mistakes and still end up with some 10 ACPL games.
To back up my data, here is the ACPL of the top players in chess history:
And here is the ACPL for the top modern players from a recent tournament of theirs:
There are a bunch of caveats to this, for example they won't penalize you for playing a gambit if it is theory out of the opening. They also won't penalize you for playing a move if the evaluation of that move is greater than +/- 2.00, because at that point you might be playing suboptimal moves, but that's because practically it may be better to just trade pieces than to engage in perfect play. Also the max penalty you can get is a 300 centipawn loss, because the difference between a -3 position and a -4 position is nil; you are lost either way.
So how can we calculate our ACPL for a game? What I did was paste the PGN of a game here:
And click the "Request a computer Analysis" toggle at the bottom, and it will compute both white and black's ACPL for the game you submitted! It's really fun to do this for your games and for random GM games too. Just for fun I took a random game of Carlsen's against Caruana from the recent Olympiad to see just how good he is. These two guys played a 7 ACPL game with 0 inaccuracies from either side.
When I ran this ACPL analysis on all my recent games against 2000+ rated players here in Seattle, I made a pretty startling discovery. These Seattle juniors seems to be churning out extremely low centipawn games regularly against me!
The first set of games that I analyzed was the play of my opponents in the Kings vs. Princes Round Robin that I played a few months ago. This tournament was 9 rounds with average rating opposition of about 2200, so we would expect the players to play on average a 21-30 centipawn games. Not the case, these Seattle juniors produced some pretty phenomenal games, and I'm going to call out the four games where someone played extraordinary chess (low ACPL with few inaccuracies/mistakes).
The first person I'll mention who played phenomenally well was NM Sam He. He didn't do too well this tournament, but the one game I played against him, he decided to play like a COMPUTER and made 0(!!!!!) INACCURACIES OVER 50 moves.
9 ACPL game, with 0 inaccuracies, 0 mistakes, 0 blunders??? WOW!!! Sam basically played like a GM this game
As if this wasn't enough, I had to deal with another junior playing like a computer in Round 5 when Joshua Doknjas from Canada played a 9 centipawn game with 0(!!!) inaccuracies against me:
And then in Round 3 NM Kyle Haining decided to channel his inner IM against me with a not-as-perfect, but still 10 ACPL game against me:
And the last incredible performance was actually by me, when I managed to play a perfect 0(!!!) inaccuracy game as well against Anthony He, although I got dinged 13 ACPL because some of my opening moves might have been a second or third preferred move of the computer:
So there you have it. In a 9 round tournament against 2200 average rated opposition, 4 of my games featured at least one player playing near perfect games, almost 50% of my games! And in three of those games, one side played computer-like, perfect chess. *facepalm* I guess this is the state of Washington chess. You better start playing perfect games consistently if you want to be 2200+ cause apparently these Seattle kids can all do this with ease. This is also testament to the strength of the Kings vs. Princes tournament, because if so many of our games were low ACPL affairs, the quality of the games was clearly very high.
In my most recent tournament at the Washington Challenger's Cup, I played two experts, Vikram Ramasamy (draw) and a game against Naomi Bashkansky (loss). I actually didn't feel too bad about my results even though I lost rating points, because I don't think I played that poorly. I thought they both played extremely well, and the computer confirmed that they played IM/FM level games against me:
Since when was it normal for experts to average 14 and 17 ACPL games, at this kind of consistent rate?
After running these stats I felt pretty depressed about my own play. I took an average of my ACPL over the last two tournaments at the Master Series last weekend and at the Challenger's Cup, and over 7 games I averaged 26.3 ACPL per game. A 26.3 ACPL corresponds to a 2230.4 rating performance according to the ACPL to ELO chart I cited above, so you'd think I would gain rating points. But in fact in the last two tournaments, I managed to lose rating points, because my opponents here are playing such low ACPL games against me. I really think in the Seattle area if you want to reach 2200+ you need to average sub-20 ACPL games because of the disturbing frequency at which these Washington players play sub 15 ACPL games.
However, one thing that made me feel a little better is that these kids have played far more chess than most of us have. For reference, I've played 90 tournaments in my life with a peak rating of 2208. Roland Feng has played almost 180 games, double the number of tournaments I have to achieve a 2400+ rating. Kyle Haining has played 165 tournaments, also almost double. Naomi Bashkansky has played 150+ tournaments to get to 2100. The most amazing statistic I found was on Anthony He, a promising 11 year old junior who sports a hefty 2250 rating. He is less than half my age, but he has played almost double the number of tournaments, at 162(!!!) for his 11 years on this planet. Imagine if you played 162 tournaments by age 11, maybe you'd be a prodigy too! So if these kids have put in that much more time into chess than us, it is not surprising that they would play much better than us. Better get studying!
So what do you guys think? Is the Seattle chess scene really more difficult? Or are the margins of winning so razor thin at the master level that playing near perfect games half the time is expected? Also what are your guys' thoughts on the ACPL way of evaluating someone's chess strength? Let me know your thoughts in the comments below!
Until next time,