Author: Stephen A. Boak
Date: 11:05:10 12/24/99
Go up one level in this thread
>On December 24, 1999 at 07:12:24, Albert Silver wrote: <snip> >I have played in tournaments where my TPR was considerably >higher than my rating, and even so, lost rating points. > > Albert Silver Hi Albert, I am skeptical that a player could have a TPR 'considerably higher' than his starting rating, and still lose rating points. Can you provide an example--I'd like the starting rating, the opponent ratings, and the results. Were you or any of the opponents provisionally rated? Also, which ELO system are you referring to--FIDE, Brazilian, USCF? There may be some unusual circumstances in a particular implementation of the ELO system, I don't know. In addition, if several tournaments are rated in the time period between rating publications, then poor results in another tournament could cause a net loss of rating points (between rating publications) despite the excellent result in one specific tournment. Example (based on the USCF ELO system)-- (I made these calculations in my head, they may not be exact, but they should be very close--I have programmed TPR and USCF ELO calculations in many computers over the years, keeping track of my own chess career). A player is rated 1900 and plays three 2000 rated opponents--none are provisionally rated. By the ELO formulas, the player's expectancy is approximately 1.2 points (about 40%) out of 3 games. If the player wins 1 game and loses 2 games, his TPR rating is 1867 (approximately his actual rating but a little lower) and the player's rating will actually drop a little (about 8 points) when updated after this tournament. If the player happens to win 2 games out of the 3 (67%, better than the ELO-calculated 40% expectancy, but not unlikely at times), his TPR rating for those few games will be 2133 (probably not his true playing strength since he is currently rated only 1900) and his ELO rating will rise about 28 points to 1928. In both scenarios, the gain or loss of ELO points is in accord with whether the player performs better or worse than ELO-calculated expectancy, and the TPR measure also corresponds with the gain or loss. I am unable to envision a scenario where a player could have a TPR considerably above his starting rating but still lose rating points. __________ The formula for TPR (Tournament Performance Rating) is *not* the same formula as the normal ELO formula. It is used to approximate or provide an 'ELO-based' rating under limited circumstances, using the ELO ratings of played opponents. The limited circumstances are typically-- --to determine a measure of performance of a player in a single tournament (this is easily extended to several tournaments, if one desires to calculate TPR for more than one tournament); --to establish a provisional rating for a new player, until enough games have been played that the players rating begins to be calculated using regular ELO formulas, and no longer by using TPR method. As the number of consecutive games increases in the TPR calculation, the TPR more closely approaches a player's average ELO rating (over the same series of games), but TPR is never guaranteed to equal a current ELO rating (although it might accidentally equal a current ELO rating on occasion). The main reason for this, as pointed out in another post, is that the ELO formulas give more mathematical weight to the recent games of a player and less weight to older games. The best one can say is that the TPR for a player for many consecutive games (perhaps in the many hundreds if not thousands of rated games) will be 'close to' the average ELO rating maintained by the player over that same series of games. However, the current ELO rating of the player may easily be higher or lower (depending on the player's general trend to improve or decline in playing strength) than the overall 'average' TPR earned for the same period. For many consecutive games, the current ELO will more closely track the player's current strength than a TPR calculation for those many games. This is why TPR figures, while interesting to chess fans and players, must be take with a grain of salt. When a great player, say A. Morozevich, earns a TPR of over 2800 or 2900 in a single tournament or two, we know by the TPR that he played extremely well versus the field he faced, perhaps even brilliantly (depending on the actual quality of the games and whether his opponents made a lot of silly blunders that gave him several games) and had an exceptional result or two. The TPR for a few games only, even for a few consecutive tournaments, doesn't mean his true ELO strength is at least equal to Kasparov's 2800+ rating (it could be Morozevich has improved that much, but only time, i.e. more games, will show). After such results, Morozevich's FIDE ELO rating might rise a few dozen points, but it wouldn't vault him to 2800 after only 2 or 3 such brilliant results. ELO formulas do give *some* weight to older games, after all, not just to the last 10 or 20 games only. --Steve Boak
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.