Toronto International IPRs---and FIDE vs CFC Ratings

[Kenneth Regan]

I had previously posted my "Single-PV quick run" of the Toronto International; now here are the results of my Multi-PV run which evaluates alternative moves in full and allows the computation of "Intrinsic Performance ratings" (IPRs):



As with my full run of last year's Canadian Open (624 games---see table at the end of my published academic paper with Guy Haworth and GM Bartlomiej Macieja), the numbers support the contention that the FIDE ratings of the Canadian players are deflated---by over 50 points. Of course this is only 56 games, but I am currently running the COQ Premier and will run the Zonal to see if this result holds consistently.

My Excel spreadsheet image also breaks down by round---again it is hard to draw conclusions from only 8-9 games per round (I've omitted the error bars which are over 200 Elo points in both directions for every player or round), but it seems the effect of the two-round opening Sunday and evening thunderstorm were not felt until the next day!? The main reason the "Average" and "Aggregate" rows have different values for IPR's is that "Aggregate" lumps together all the moves and different players had widely different numbers of moves.

Incidentally I'll shy away from posting my entire IPR spreadsheet of a big Open, on grounds that the IPRs might be taken as "personal" like IQ figures, but I consider top finishers and Master-plus players and those playing in a premier section to be fair game...hope that's OK. As I offered before, people are anyway welcome to e-mail me for their IPR figures from last year's Canadian Open (top section).

[Gary Ruben]

Re: Toronto International IPRs---and FIDE vs CFC Ratings

As with my full run of last year's Canadian Open (624 games---see table at the end of my published academic paper with Guy Haworth and GM Bartlomiej Macieja), the numbers support the contention that the FIDE ratings of the Canadian players are deflated---by over 50 points. Of course this is only 56 games, but I am currently running the COQ Premier and will run the Zonal to see if this result holds consistently.

What's the margin of error in percentage? +/- ? %.

It's interesting you can do this with the Canadian Open events when some years they have been paired as regular Swiss, accelerated Swiss or Double accelerated Swiss. Not to mention the so called "Swiss Gambit".

[Jordan S. Berson]

Re: Toronto International IPRs---and FIDE vs CFC Ratings

Kenneth,

You post a whole lot of this stuff, however I can't make sense of any of the jargon.

Do you have a site that explains everything you're posting?

Thanks, Jordan

[Kenneth Regan]

Re: Toronto International IPRs---and FIDE vs CFC Ratings

Error bars are given in my paper---and they don't simply depend on the # of moves but also on how tricky those moves are (e.g., obvious recaptures basically count as nothing). Very roughly the two-sigma confidence interval is about +- 200--250 Elo for a player-performance of 7--9 games, and +- 80--100 for a tournament of 45 games or so. Expressed as a percentage the latter case is about +- 7%. For instance, Toronto 2012 had 56 games, and my program outputs for the total AE of all games:

259.61, 2-sigma range: 242.39--276.84

I currently have 40% modeling error on top of that, so you can call it 10% error for tournaments. Very rough---and I meant to say that the last digits of my IPRs are not significant and they should be rounded to the nearest 10. The y-intercept on the Elo scale is about 3600, so if I say someone played 9 games at 2400 +- 200, that's about 200/(3600-2400) = 17% error for a player. It sounds a little better if I say 8.5% is the standard deviation.

Jordan---please let me know if my above-linked paper with Haworth and Macieja is helpful. My ongoing "Intrinsic Ratings Compendium" draft is trying to be simpler.