What is the largest USCF rating gain that you’ve seen from one tournament?
Here’s an eye popper from the Bay Area Championship #27 last Saturday: 1372 (P5) => 2211 (P9). That’s +839 points. Yes, we’re still talking a highly provisional player. Even with that caveat, how often do you see D players become masters overnight? In fact, the USCF ratings calculator only predicted a 1750ish rating for this tournament.
So how did this player gain 839 points? Well, it turns out that he won all 9 games and the highest person he beat was 1814, meaning the ratings software assigned him a rating approximately 400 points above that.
In case you’re wondering, we are not talking about a Russian immigrant but rather a 10 year old kid (Paul Richter, USCF 13847644) with no prior chess experience or coaching! I am sure that he’s got some skill, but also a lot of luck. Unless this kid is the next Bobby Fischer, then I am pretty sure that he is not 2200 strength or even close (note: the nation’s top 10 year old was 2129 on the December list). Probably that 1750 would have been a more accurate rating for him. Oh well, I just hope he doesn’t get discouraged by the inevitable drop in his rating.
I played in that tournament as well, and I’d like to add that Paul also
defeated players rated 1626 and 1687. The tournament was a four
round G/45 tournament, and Zachi (the 1814 player) and Paul faced off
in the final round. I might have got the chance to play Paul, except that
Zachi had defeated me the previous round.
I looked at the final game once, and it seemed to me that Zachi had a
slight edge. He was up a pawn in a complicated position, but he had used
up a lot of time on the clock, while Paul still had over 30 minutes remaining.
I didn’t see how the game went from there.
In any case, Paul was playing extremely well. While I agree with Michael
that the 2211 provisional rating is likely overstated, that is the way that
the provisional ratings work. This was Paul’s second tournament, and
he won all the games in his first tournament as well.
I also wanted to add, in case anyone got the wrong idea from Michael’s
post, that there was nothing suspicious going on. The only fault here,
if there is one, is in the provisional rating formula.
If he has won all his games in both of his events, then his rating will be 400 points above the rating of the highest rated player he has faced. That’s because about all we know about that person’s rating is that he is apparently better than the highest rated player he’s faced so far.
Provisional ratings are called provisional for a reason, they’re based on a small set of games and as a result are not as reliable as established ratings.
Incidentally, a player is not entitled to be called a master until he or she has an ESTABLISHED rating of 2200 or higher.
I just took the time to see why his rating was in the 1300’s after his first tournament.
And it’s readily apparent … he played no one over 1000 … (a couple of those players did break 1000 after that tourney by a few points). He STILL came in with his first provisional rating nearly 400 pts higher than anyone he played. And yes, he made a clean sweep 5-0!
This kid should be placed on the prodigy watch list!
If his rating is still quite high when he gets 26 rated games, even if he drops below 2200 or 2000, he should be very near the top of the Top 100 list for his age group and should qualify for the All-American Team too.
That was exactly my point. While it is always difficult to predict provisional ratings, this example is a bit too extreme, IMHO.
Now that he is 2211, he will have to play Master/Open sections. If he plays against two 2200s, two 2100s and two 2000s in a hypothetical 6 round open and scores 0-6, then the ratings calculator still puts him at 2003. He’ll be overrated for some time. Hopefully this won’t discourage him too much.
Of course, he might be the next Bobby Fischer, in which case I will gladly eat humble pie.
I’m not sure the ratings estimator correctly handles the transition between a provisionally rated player who has won all his games and a provisionally rated player who has both wins and losses, because the ratings estimator DOESN’T KNOW that his 2211/9 rating is based on all wins.
The ratings programming does know that, though I’m not entirely sure what going 0-6 against the field you suggest would do to his rating. I think it would be lower than 2000, but I’m not sure how much lower. (And I doubt if I could find a recent comparable situation.)
According to the ratings formula, a player with 9 or more completed games (prior to the current event) is usually rated using the ‘regular’ formula rather than the ‘special’ formula, EXCEPT when that player has won or lost all his games, as is the case here.
If it completely reverted to a form of the old W+400/L-400 style provisional ratings formula, then those wins against the lower rated players would tend to pull his rating after 15 games down considerably.
BTW, to answer your lead question (even though it wasn’t really the question you were asking), it appears that since 2004 the largest regular ratings gain a player with an established regular rating has gotten from a single event is 508 points by a player who went 8 1/2 out of 9 in the 2006 World Open U1200 section.
Since 2004 there have also been 18 other instances of an established regular rating going up 400 or more points.
To be honest, I’m not 100% sure. As long as he has all wins, his rating would continue to be computed using the ‘special’ formula. The first time he has a draw or a loss, it would change how he’s coded, but only after computing his new rating.
Some months ago I asked the Ratings Committee chair if someone who has more than 25 games but all wins (or all losses) is still considered provisionally rated, I don’t think I got a definitive answer. The Ratings Formula does not appear to specifically address this, either.
That’s not going to happen. In his first tournament he was placed in a
low rated section, because he was unrated. After that his rating was over
1300, so he was put in the “Elite” section, which is the open section of
these tournaments. So if he continues to play in these tournaments, then
he will be in this section from now on.
The Elite section had an Expert and a class A player, as well as several
class B players in it. Frisco, the Expert, had to leave early, or else Paul
would likely have faced him.
Of course, it is inherently impossible to satisfactorily compute a new player’s rating if he wins all (or loses all) his games.
I’ve often wondered, however, why the ratings committee didn’t adopt the following idea for new players, at least for those who do not win or lose all their games: A player’s first rating would be that rating which, if used as the pre-tournament rating, also comes out (using the established formula) as the post-tournament rating. (If, due to rounding, there is a range of such ratings, just use the rating in the middle of that range.) Any K-factor could be used, since we are interested only in whether, rather than how much, the player gains or loses points.
Unfortunately, for a new player who wins all or loses all his games, this idea wouldn’t work, because there would be no rating which would remain the same when run through the established formula. (A player who wins all would always rise, a player who loses all would always drop.) But you could still use that method if you assume a tournament score (for calculation purposes) of, say, 0.25 points below a perfect score (for an all-winner) or 0.25 points above a goose egg score (for an all-loser). For example, a player scoring 9-0 would be figured at 8.75-0.25, or one scoring 0-9 would be figured at 0.25-8.75.
Hmm. Mike Nolan, who a few years ago implemented the rating formulas into the current ratings software (and who was justifiably proud of this accomplishment, given the complexity of the formulas), now isn’t sure how it works. Interesting.
Bill, the formula doesn’t specifically address the issue of whether someone with more than 25 games but all wins or all losses has an established rating, because it really doesn’t define what an established rating is. I think that’s been more of a political issue than a mathematical one.
I’m not a theoretician, I’m an empiricist. I tried to write the program to follow the ratings formula as closely as possible (which involved a lot of email with Mark Glickman and Tom Doan, not to mention hours of banging my head against a wall), but I don’t claim to fully understand the theory of what will happen when he finally loses or draws a game.
However, here are some recent examples of players who had 9 or more wins before their first loss or draw. (A type of ‘W’ means all their prior games have been wins, a type of ‘P’ means that they now have an ‘ordinary’ provisional rating, which means they must have drawn or lost a game in that event.)
[code] ID Event Prertg Games Type Postrtg Games Type
13402593 200702096271 1228 9 W 1203 13 P
13532632 200702044971 1317 13 W 1326 17 P
13552222 200704280901 1238 10 W 1249 15 P
13591947 200701244891 1492 10 W 1416 14 P
13735162 200712083171 1865 9 W 1795 12 P
13743031 200709021861 1633 11 W 1588 14 P
13744411 200712098971 889 13 W 836 20 P
13778628 200801129001 1192 10 W 1095 14 P[/code]
I will leave it up to the theoreticians to argue whether the ratings formula inadequately addresses situations such as this.
I took the highest rated person from Mike Nolan’s dataset (13735162, rated 1865 after starting out 9-0) as an example. He scored 1.0 against a 1400, 2060 and 1700 for a 1530 performance. The ratings calculator predicts a new rating of 1804. The actual calculation was 1865 ->1795P12. That means the ratings calculator is reasonably precise for this example.
If also find it interesting to note that none of the players in the dataset dropped a large amount (more than 100 points) when they finally proved that they were human. That means the +400 approximation for the provisional rating was an accurate estimate for these players’ strength.
On the other hand, if Paul Richter’s 2211 rating is even close to accurate, then I am going to make sure to get his autograph before he charges a fee for it.
I wasn’t predicting it would happen. But I can’t help but think it would be possible for a kid, because he could probably find enough rated scholastic events where he wouldn’t face anyone capable of posing a threat. I certainly would expect that a player of his apparent talent would be unlikely to have parents willing to put him up against such lame opposition. So my question was theoretical.
A week ago, young Alex Katz of New Jersey, age 11, USCF rating 1802, turned in a 2500+ performance rating.
Even though he only played two rounds, I’d say it was quite a super bowl Sunday for the kid. Still, I’ve got this nagging suspicion that he’s not quite that strong yet.
This illustrates the inherent flaw in the ‘performance rating’ concept.
Defeating two 2100 players (which is one way to get a 2500 performance rating) is not really indicative of 2500 strength over a longer string of games.
Now, to continue do something like that over 26 games (which is what it takes to get an established rating), preferably against more than one or two opponents, well that does say something about the strength of that player.
