What was your highest rating in the middle of a tournament? (not final, but according to the rating estimator) Mine was in the middle of Vegas as 2055!
The curse of the last round
Sort:
JMB2010 wrote:
Hmmmm, the tournament was rated and the Rating Estimator on the USCF site is off by a lot more than it usually is. Anyone else experiencing this?
Have you heard of the rating system change? It makes everybody's ratings change more than it did before: most people go up/down 10-20% more, but 1800-2200's change 40% more than before. The rating estimator on uschess.com isn't the current one, you have to click a link at the top of the page to get to the current one. Yeah, I know it's stupid.
I'm 2073! MUAHAHAHA... even though I really screwed up that tournament. I was 4.5/6, and blundered away 2 draws in a row (including one against JMB), and ended up with 5.5/9. Money was there. It was, but I missed it.
Yeah, Abhishek told me something about a more sensitive K factor. USCF ratings already moved up and down somewhat more drastically than FIDE ratings did as far as I know, so for the k factor to be even higher seems pretty crazy. Although in my case I don't mind that because if I improve -- and I think I'm an improving player -- the ratings will reflect that without me having to play as many rated games.
"most people go up/down 10-20% more, but 1800-2200's change 40% more than before."
Wait, really? So how much your rating changes depends on, everything else being equal (such as how much higher or lower rated you are than a given opponent), your rating? Strange, strange stuff.
Yeah, I told Elubas about a more sensitive K factor. Actually, after I played USCF president Ruth Inez Haring, we were analyzing the game afterwards and she told me a lot about her job, and why the K factor was changed for a certain rating range. Ha! I got to meet the USCF president.
Well, since there are SO many old people in the USCF who have the same rating for a long time, they had a meeting and decided that more people would want to pay for membership if their ratings went up, so if you win or lose you would gain or drop many more points than with the original K factor. So, JMB's rating should have only gained <10 points, but with the increased K factor, he got a higher rating. Since Elubas, me, JMB, and gimmewuchagot are all in the rating range this affects all of us.
"More people would want to pay for membership if their ratings went up"
But I'd guess all an increased K factor would do is increase the gaps between players. Yeah that will push some people up but it will push others back as well. If you're rated under the average rating and play regularly at the same strength you have been for years then your rating should drop.
And yes, those furthest away from the average should experiance the most change.
"If you're rated under the average rating and play regularly at the same strength you have been for years then your rating should drop."
I'm not sure how that works. I would imagine if you have an established, "accurate" rating of say 1600, and you don't improve or get worse, you will still average out at 1600. True, you might dip a little more than you normally would, maybe get down to 1550 or less, other times 1650 or more, but you'll still average at the same spot.
And of course someone who is 1600 but improving will eventually move to his "real strength" regardless -- all the k factor does is increase the rate at which that will happen.
Of course, not everyone's rating will go up, but I think the economic reasoning is that people will be excited at the chance to improve their rating quickly, (just as those partaking in the lottery pay for the chance to win big) even though in reality not everyone will achieve this.
Well, I am improving... and thanks to the new (overly) high K-factor, my rating has gone up 120 points in two somewhat disappointing tournaments, which could have been way better.
NY International, 3.5/6, 1956 => 1990
World Open, 5.5/9, 1990 => 2074
I thought it was the new rating system too, but I used the new rating estimator and it was off. Maybe the tournament used post tournament ratings, and 2 of my 1900 opponents went to high 2000s and 2 of my 2100 opponents went to 2200
Well... maybe they need to fix the rating estimator :P
Lol, maybe Abhishek2 should tell Ruth Haring this the next time he plays her :)
Huh maybe I should play a rated tourney! Just kidding. Great games JMB 2010, unfortunately I have no useful comments regarding them as you are out of my league. Good luck going for a title!
Elubas wrote:
I'm not sure how that works. I would imagine if you have an established, "accurate" rating of say 1600, and you don't improve or get worse, you will still average out at 1600. True, you might dip a little more than you normally would, maybe get down to 1550 or less, other times 1650 or more, but you'll still average at the same spot.
Ratings are a relative measure only, it's not like a five minute mile. A 2000 rating here isn't a 2000 rating there (I'm sure you know this already).
So short answer is, when you change the formula, people's ratings will change even if their performance doesn't. As you said not everyone's rating can go up. Some will go down.
If you understand that it will increase the gap between all players then you understand that some ratings will have to go down. Their overall placing (e.g. 60th percentile) will not change, but the representative number will.
Elubas wrote:
Of course, not everyone's rating will go up, but I think the economic reasoning is that people will be excited at the chance to improve their rating quickly, (just as those partaking in the lottery pay for the chance to win big) even though in reality not everyone will achieve this.
If lower rated players misunderstand how it works, then yes, they'll be excited :p
Yes, and the k factor does not change relative skill level.
I don't see how you have refuted my point about a person (who doesn't improve or get worse at chess) averaging at 1600 regardless of the k factor. Yes, for every time someone scores a point off of the 1600, the 1600 will lose more points than he would have with the old k factor (I'm assuming this is why you are arguing the 1600 who plays the same is doomed to a lower rating). But if the 1600 really plays like a 1600, he can just as easily get all of those many points that he lost back the next tournament. As I said, the ratings will move back and forth a bit more, but the 1600 will average at the same spot. Instead of the 1600 losing, for example, 10 points for a bad tournament, he may lose 20. But he'll also gain more when he has a good tournament, evening things out.
In the long run (I can clarify what I mean by this if you want), nobody's rating, actually, goes up more than it otherwise would; all the k factor does is change the speed at which improving players hit their "true" rating, as well as making back and forth rating swings more dramatic for those who don't improve or get worse (as I talked about in the first paragraph).
Maybe an easier way to put this: 1599, 1599, 1600, 1601, 1601.
Average is 1600.
1570, 1570, 1600, 1630, 1630
Average is still 1600. Standard deviation is indeed higher than in the other set, accounting for some larger temporary "gaps," but the average remains the same.
Yes, players right at top of the bell curve (the average rating) should not experience any shift.
For players who are not at the average however, it's not that the person's performance changes, but the scale itself shifts under them so to speak. As all the ratings bounce up and down at once, it has a chain reaction and everyone tends to spread out. Or to put it another way, it's harder for two players to maintain very close ratings.
I'm not sure why you call the gaps temporary. It's not useful to talk about a players in this way: "A 1600 who plays like 1600" because after all 1600 is relative. What I'm saying is the 64th percentile player who continues to play as a 64th percentile player will be assigned a rating which is the new representation of the 64th percentile.
It's also easier to imagine it from ends and work towards the middle. Imagine top USCF players like Nakamura. Their ratings will tend to increase simply because their high win percentage. 2nd tier top players (so to speak hah) will increase also, but a little less. And so on toward the middle. The rating range that represents each percentile range has been widened.
Take for example BCF ratings. Our masters are 2200 while theirs are 200. Are ours significantly stronger? No. It's just in that system 200 represents the 2 percentile (whatever masters are, I'm not sure). Because they gain and lose so few points per game, the numbers required to represent their distribution of players needs only 300 points or so while USCF needs about 3000. Our classes are 200 points while theirs are ~15.
"For players who are not at the average however, it's not that the person's performance changes, but the scale itself shifts under them so to speak. As all the ratings bounce up and down at once, it has a chain reaction and everyone tends to spread out. Or to put it another way, it's harder for two players to maintain very close ratings."
I would apply my same reasoning to an 1800, 2100, doesn't matter. To be honest this is very vague, and I don't really know what you're talking about. Can you give me some sort of scenario? Chain reaction... perhaps you mean if one guy plays unusually well, his rating will go up a lot, which will carry over to the other people he plays? But this can just as easily happen in the opposite direction -- if I play unusually well for a while, then you could say I'm "taking away" points from people in my rating group. But unless I sustain that level of play forever, I will eventually start having more "normal" and sub par performances, losing points, and thus "giving back" points to my opponents.
"It's not useful to talk about a players in this way: "A 1600 who plays like 1600" because after all 1600 is relative. "
Well, let's see what we mean when we say this. If you're 1600, that means you beat 1400s x amount of the time, or if you don't, you make up for it by beating 1800s x amount of the time (or something similar). All of such players are of course in the USCF.
Now, let's take this 1600 and 1400. Let's change the k factor. Does that have any effect on the odds that the 1600 will beat the 1400? Nope. In fact if they played a 20 game match or something, and the result was what was statistically expected, they would remain at about the same rating.
In fact, even if the k factor was crazy, like maybe the 1600 gets 40 points everytime he beats the 1400, if the result is as statistically expected these will be balanced out when the 1600 loses once or twice -- he might lose 150 points or something. If they played a lot, both players would still hover around 1400 and 1600, although, as I said, standard deviation is high.
In tournaments this is of course more complicated because some people are overrated or underrated or have unusually good or bad performances. But, as I have said many times, everyone plays underrated and overrated people sometimes. You can expect things to even out.
"Take for example BCF ratings. Our masters are 2200 while theirs are 200. Are ours significantly stronger? No. It's just in that system 200 represents the 2 percentile (whatever masters are, I'm not sure). Because they gain and lose so few points per game, the numbers required to represent their distribution of players needs only 300 points or so while USCF needs about 3000. Our classes are 200 points while theirs are ~15."
Ok, but I don't think this has anything to do with the k factor. It's just that it's arbitrarily set up so that a 100 rating point gap results from scoring x against some weaker player. In the BCF, perhaps if the same two players played there would only be an 8 point gap or something. That's simply how it's set up. I think you are getting this confused with the k factor. It's not the k factor that causes these number differences, it's simply the definition of what a 100 rating gap means that does.
"It's also easier to imagine it from ends and work towards the middle. Imagine top USCF players like Nakamura. Their ratings will tend to increase simply because their high win percentage."
If Nakamura were to regularly play USCF games, he would of course win most of his games. But for every game he lost, he would also lose points massively. So, sure, with a different k factor at some points he might be rated 2950 (his uscf rating is 2880 but let's assume he's having a good tournament). But if he has a bad tournament, he might, with a high k factor, drop below 2800! So it can work both ways. Again he will average out in between these two extremes, at his current rating.
I could clarify what I was thinking, but I don't think it would be very interesting because I was thinking about this incorrectly. I think you're right. Thanks for continuting to explain yourself.
How much? It can vary, although it's usually pretty decent. I think the way ratings are actually done is one at a time. So for example if you played 4 tournament games, they would take your first game, the result, your opponent's rating, and then your new rating. Next, for game #2, they would take the new rating from game #1, (and the opponent's new rating from game #1 as his rating changed too going into game #2), and create a new rating for you and your opponent from that. The same is done for game #3, and so on. This is a little different from just putting all 4 games into the estimator. For example if your rating starts out as 2100, you're not simply a 2100 for all four of the games; in round 2 you might have increased to 2108, in round 3 you were maybe 2122, in round 4 maybe 2110. Same goes for any opponent you face. Those things affect how many points you will gain or lose in later rounds.
In fact I just messed around a bit with the estimator. In one scenario I made "my" rating 1800, put in four 1800s, gave a score of 3 out of 4, and it said my new rating was 1830. Then I tried it individually. I took my 1800 rating and saw how it changed after beating just one 1800, it went up to 1813. Then I took my 1813 rating and saw how it changed after beating another 1800, and same for the third win. But when my rating was 1837, a loss against an 1800 cost a lot of points, and ultimately I ended up with 1823 in this case. 7 points less. I think it's because if you are going to lose, you are better off losing with a rating of 1800, rather than with a rating of 1837. If you lose early on, the next three wins will all be amplified slightly. I tried the same experiment with losing to an 1800 first, then beating the next three 1800s. The final rating came out to be 1826.