Author: Stephen A. Boak
Date: 21:20:12 01/18/02
Go up one level in this thread
On January 18, 2002 at 18:53:09, Mike S. wrote: >Somebody pointed me to this ratings conversion applet (at the website of a swiss >chess club): > >http://sg.biel-bienne.com/vorstand.htm#USCF > >I don't know the formula though; they do not just subtract a fixed value (the >difference is smaller for higher rating numbers). > >Can somebody acknowledge that conversion? I'd like to use, and maybe recommend >that webpage to convert USCF rating estimations of chess computers and programs >to FIDE elo, and vice versa. But I thought, better get an 2nd opinion or >information from an ratings expert before doing so. > >Thanks, >M.Scheidl There is no true (exact) conversion formula (for math reasons--different pools, different 'spread', etc), although some rough approximations may be usable in some instances--for some purposes. I had often heard (noted) that USCF - 100 = FIDE was about right. Directors may, perhaps, add 100 pts to a FIDE rated player with no prior USCF rating--for pairing purposes in a USCF tournament. Maybe this is a common assumption that has been used. I've seen this done (and at first, I wondered why the differences showed exactly 100 pts on the table name displays! heh heh! [ example--fictitious: S. Volkalov IM FIDE 2475 USCF 2575 ] A year or two ago, while at a medium to large size USCF tournament, I noted that the Directors had also listed FIDE ratings on the cross-table sheets for players with a FIDE rating. For fun, I logged both the USCF & FIDE ratings on a piece of paper for those approx. 17 players. I hand graphed the results on the scratch paper. The USCF/FIDE ratings ranged from approx 1900 to approx 2500+. Per my memory (not perfect I'm sure) I think the average delta was the better part of 100, perhaps 85 pts., for the limited sample size. And USCF was usually higher than FIDE, as I had heard & previously noted (casual observation). This sampling roughly confirmed/supported the old adage (FIDE + 100 = USCF). I noted from my particular graph that players above 2300-2350 or so tended to have variances that were regularly different (more or less I don't recollect) than those below that level. I can't remember exactly. This does not say much, since there may have been only about 8 players on either side of that line--not many from which to draw conclusions. In one, maybe two instances, players with approx 1900 or 2000 USCF rating were *lower* than their FIDE rating by some amount (50-150 pts). All in all, I myself use a 100 pt. approximation, in contemplating such questions, but it isn't based on a large math sampling, but only on observed examples and the above small investigation. Applied to individual players, who have only a single type of rating--FIDE or USCF--the adjustment may be totally incorrect, depending on many factors. I suspect, in my 'test', that many of the lower rated players didn't have much opportunity to obtain a FIDE rating, and may therefore have a FIDE rating with only a few dozen games behind it, from who knows how long ago, whereas their USCF ratings would likely be based on regular (a half dozen larger tournaments per year, plus other weekly or monthly club & local tournament play in some cases). The next time I am at a big tournament, where FIDE ratings are also shown, I may try to repeat this experiment--and will post the results if I do it. Also, if I figure out a way to randomly sample a few dozen USCF players of 'higher' ratings (who may likely have played in several FIDE tournaments), I might look them up on the FIDE website to see what the data showed. I didn't look at your link, to see what formula or methodology is being used, but if it isn't based on several hundred (minimum) samples of active FIDE players who are also active USCF players, I wouldn't put much trust in it. If it somehow averaged at USCF - 100 = FIDE, I would probably tend to think it supported the old adage, but I'd like to see how it worked in both the top and bottom tails of a 'normal' (bell-shaped) rating distribution before jumping to any strong conclusions. Here are my guesses on this issue: As I conceptually envision it, a list of *all* USCF players with FIDE ratings would look like this, upon examination: 1. The USCF ratings plotted by quantity across all rating levels would look approx. and closely bell-shaped. 2. The FIDE ratings plotted by quantity across all rating levels would look approx. and closely bell-shaped. The reason for the bell-shape guess is twofold--standard probability theory, plus the notion that normally only the more experienced (mature), higher rated players would have FIDE ratings, and there would be very little influx to FIDE of low rated players that would heavily skew the curve at the bottom rating side (which might be the case because of a lot of new/young players with initially very low ratings since they haven't developed much yet as players, and haven't yet been sorted or differentiated across the overall bell-shaped continuum as would more mature players with a lot of experience). 3. The Mean of the USCF and FIDE bell-shaped curves would be different, perhaps 100 points apart, with USCF - 100 = FIDE (at the Mean). 4. The Standard Deviation for each curve would be different, perhaps +/- 100 points for the USCF ratings, perhaps something less (+/- 80 points) for the FIDE ratings. [NOTE--I suspect there are less FIDE players in the world pool, than there are USCF rated players, so I would guess based on some probability theory math concepts that the spread (SD) would be larger for USCF than for FIDE.] BOTTOM LINE--looking to validate the link's USCF / FIDE conversion formula would be hard, without a lot of data. Finding out what the link's formula really is, and why it was constructed as it was, would be interesting. I'd like to hear the theoretical underpinnings for such a formula (how created & why) before jumping to conclusions about how good/bad it is. I would not use it to predict for individuals, unless, based on Point 3, above, it was based on a calculated figure for the difference between the two Means (or something very similar in 'accuracy'). I simply mean that without more information, I would not leap to trust it for any reason where high confidence was needed, or high confidence was to be placed in the results of the conversion formula. If the formula was used by Directors, in some tournaments in some countries, to allow reasonably accurate Swiss pairings, then it might be fine (since all ratings fluctuate, and have theoretical swings about the 'true strength' of the player--according to probability theory). It would be 'good enough', although never perfect. Just some thoughts, --Steve
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.