COQ Premier IPR's

[Kenneth Regan]

I blipped on my intent to post this a month ago. The two main takeaways are GM Wesley So's incredible accuracy (currently that's what my "Intrinsic Performance Ratings" are measuring most) and the 43-points-less performance by the natives. Jean Hébert and Haizhou from the Open Section are also included.



The two right-hand columns weight by the number of eligible moves (turns 1--8, turns in repetitions, and turns where one side is ahead over 3 pawns are excluded), and the figures for Canadians are one line lower. I forget what the "-726" was for.

What reminded me was finishing my run this week of the entire Canadian Zonal---my thanks to the game-compilers including Hugh Brodie for the file of all 150 games in perfect shape. There the news is better: the players out-performed their weighted FIDE average by about 70 points. Time control and schedule and other tournament conditions may account for such differences---as may variance within my error bars---but my work overall hints at motivation as a big factor.

Before I post the other chart, a query: My sense is to shy away from posting IPR's of whole Open sections or lower sections, as they're kind-of IQ-ish and folks out to have fun might not want their brains dissected. But if you play in a Premier section or place high then it's fair game... Any sense of objection, noting also that in the Zonal some minors were involved---? I have not had any so far when I've inquired privately in certain cases.

Currently I am running a selection of 450 games from the Open Olympiad, including all games by teams from federations I've corresponded with---this plus the two current super-tournaments will take about a month!

[Eric Hansen]

Re: COQ Premier IPR's

2208 performance :D:D:D

[Gary Ruben]

Re: COQ Premier IPR's

2208 performance :D:D:D

Then you went on to finish top 10 by performance rating on your Olympiad board scoring 70% and earning a GM title. Go figure. :p:D

[Kenneth Regan]

Re: COQ Premier IPR's

Indeed I'll be curious to see what your Olympiad IPR comes out to. I may also run the 2012 Canadian Open---I'd like to have more of the games by the lower finishers, to make a fuller comparison to my run of the whole 2011 event as in my published paper with Guy Haworth and GM Bartlomiej Macieja.

One thing shown in that paper is that the 95%-confidence error bars are pretty wide, and can easily be +- 200 or even 300 Elo for a 9-game event. The flip side of this is that the quality of your chess performance is widely variable: a 2400 player can perform like a 2700 or 2800---or like a 2100 or 2000---player fairly frequently. Thus turning in a 2200 and next a 2700 result is not so unusual. I hope that some general use can come of this, since chess is distinguished from things like concert music performing and sports in having Elo as a universal and mathematical measure of quality. And stable too---further not-yet-published results on all games between players near 2600 in round-robin or similar tournaments since 1971 show at most 30 points of inflation since the 1980's, a difference that is ascribable to faster time controls and no adjournments.

[Roger Patterson]

Re: COQ Premier IPR's

........

One thing shown in that paper is that the 95%-confidence error bars are pretty wide, and can easily be +- 200 or even 300 Elo for a 9-game event. .....

which is in fact pretty useful as an argument against those who conclude that because someone had a good tournament, they must have been cheating.

95% confidence intervals imply that pretty much every tournament will have someone who exceeds those levels [although the individual results are not independent of each other].

below is the distribution of performance ratings relative to pre tournament ratings for the CFC (established players only, 3 game minimum, 50 points wide bins are used).

[Kenneth Regan]

Re: COQ Premier IPR's

Thanks! I'd be interested to see the publication that comes from. Clearly the blue dots stand for intervals 50 Elo wide, centered on -25 for -50-to-0, +25 for 0--50, 75 for 50--100, and so on.

The main point raised by my model is that the lower set of error bars in the paper are projected. They are inherent in the model, not based on data after the fact as the chart is. Human variance can be due to people being under the weather, or playing after work, or booking up really well, or being on an emotional high. But the lower bars in my paper set a floor for variance that is unaffected by any human considerations---it is intrinsic to the limited knowledge we can have while facing the complexity of chess.

The upper bars in my paper, which multiply by 1.4 (i.e. 40% wider), were determined after matching to real data. They represent the above human variance and also the extent to which my modeling is deficient---e.g. I could use higher depth or a better mix of engines or have more factors in my equations (I aim to do all three). For cheating-testing my multiplier is only 1.15 (i.e. 15% wider), as I described in a general-science article here. The width of the lower bars was really surprising to me---I know, I made a "silly omission" in my original draft and had much narrower lower-bars that looked believable to me.

Again, the point is: because it's chess---not just because you're human (and indeed for computers too)---your mileage will vary.

[Roger Patterson]

Re: COQ Premier IPR's

Thanks! I'd be interested to see the publication that comes from. Clearly the blue dots stand for intervals 50 Elo wide, centered on -25 for -50-to-0, +25 for 0--50, 75 for 50--100, and so on.

.

yes. No publication though, I thought you would appreciate the data so I extracted it from a copy of the CFC ratings database.

...Human variance can be due to people...[bunch of stuff] ....

yes. also just plain statistical noise as well as mis-rated opponents

[Kenneth Regan]

Re: COQ Premier IPR's

I thought you would appreciate the data so I extracted it from a copy of the CFC ratings database.

Thanks even more. That is a beautiful bell curve!

[Gary Ruben]

Re: COQ Premier IPR's

What would a players performance rating under performing his rating by 250 to 400 points indicate? Controversy of suspicion of players "throwing" games some decades ago was reasonably wide spread.

[Kenneth Regan]

Re: COQ Premier IPR's

Gary, that is indeed one of the things to investigate. For example with The Hague 1948, my Single-PV run shows the performances of both Botvinnik and Keres to be sub-standard. But my full Multi-PV run produces the following IPRs:

Botvinnik 2583, opponents 2293, versus Keres: 2577
Euwe 2219, opponents 2678
Keres 2587, opponents 2422, versus Botvinnik 2421
Reshevsky 2503, opponents 2447
Smyslov 2624, opponents 2645

Given the large error bars, nothing convincing here.