inflated ratings

[epeemike81]

Originally posted by BugabooX
I don't think fencing includes very many valid "expected values" at all. So the "edge," and resulting fluke rating, simply doesn't exist where most people assume.

It's the same sort of paradox inherent in the statement, "the only constant is change." If you keep seeing unexpected results, then the results really shouldn't be all that unexpected, neh? Regardless of whether our goofy protagonist stumbled into a good rating because everyone else screwed up or whether he had a brief flash of skillful brilliance, the rating was fairly earned and it exists within the bounds of likely fencing outcomes. No flukes.

Actually, it was fairly earned and it exists within the bouts of fencing outcomes, not LIKELY fencing outcomes. whether or not it is likely can only be judged by how often it is re-earned. if an examination of re-earning shows it to be UNlikely, then it is a fluke. As for the "edge" not being where most people assume, I addressed this farther up. If you recall, I noted that I had to shift my mental barometer this year, since I had seen some that at the time I considered flukes so often that they could no longer be considered flukes. however, the fact that most people don't judge correctly which are flukes does NOT mean flukes don't exist. you yourself said there is a range of A's, correct? well, if you could judge the number of A's of each strength, that would give you your bell curve. Below a certain level (10% of the frequency of the peak) can be considered flukes. As for a way to judge that strength, see above.

-m

[edew]

Mike, it seems like you see a result and claim that the result couldn't have happened because it's unlikely. But it happened. So it's not unlikely. YOU must readjust your perception and reassess the probability that the person could or could not earn a rating.

While at the DITD, I went over to the poker tables after the tournament. At the table I played at, we had one person win with four 3's, all showing. Winning hands almost required a straight or higher (this is 7-stud) to win. It used to be trips or two pairs would duke it out for a win. But suddenly at this table, people were competing with boats and flushes. Practically every other hand had something higher than a trip. Unlikely, yes. Do I make adjustments to my betting because of this fact? Well, yes and no. No, because each round is as random as the previous. I'm not going to bet differently because of streaks. But yes, when I see someone with three fours showing, I'm not going to bet along if all I got is four to an inside straight.

What you're doing is seeing a streak of mediocrity from one fencer, and not willing to accept the possibility that he or she can have a break-out day. That's no different than if I were to see a streak of high hands, and then bet high assuming that I'll have a high hand soon, too.

Probability is the measure of your full knowledge of the uncertain event. It is not calculable. The more information you get about the uncertainty, the more "accurate" your probability assessment will be. Accurate means that you can set a probability to the possibility (mapped to some fungible yardstick like dollars), and experts would not be able to choose one way or another.

A probability alone has very little value if you don't attribute it to the corresponding possibility (i.e., outcome).

[edew]

Mike, here's a question for you:

Suppose you're at the roulette table and ten consecutive spins resulted in an odd red number. Will you put more money on an odd red number? Suppose 100 consecutive spins resulted in odd red numbers. Will you then? Suppose 10,000 consecutive spins did that? At what time will you decide to put more money on odd, red numbers?

Bayesian probability will tell you that you should, immediately. But the additional amount that you would put chips on odd, red numbers would initially be very small. After 100 trials, the amount you would put on odd, red numbers should increase, if there were 100 consecutive odd red results. It's a standard Bernoulli trial formula.

If you still stick to saying, "no, this is a fair roulette wheel, I'm not going to assume anything." Then you're ignoring all the information that you've just picked up.

[BugabooX]

It keeps coming back to the admission that we need to collect data on how many fencers earn a higher rating and then fail to do so again. ... AND somehow project prejudice onto that, believing that those people didn't deserve that rating to begin with.

Well, we don't have quantifiable data, only a bunch of biased assumptions that there are undeserving fencers earning grades against all expectations.

[epeemike81]

Originally posted by edew
Mike, it seems like you see a result and claim that the result couldn't have happened because it's unlikely. But it happened. So it's not unlikely. YOU must readjust your perception and reassess the probability that the person could or could not earn a rating.

At what point do I say it could not have happened? that would be downright delusional. What I say is that I believe (a qualitative assessment) that that is an unlikely result (fluke) and that it will probably not be repeated. In multiple instances, I have been right about this (so far).

While at the DITD, I went over to the poker tables after the tournament. At the table I played at, we had one person win with four 3's, all showing. Winning hands almost required a straight or higher (this is 7-stud) to win. It used to be trips or two pairs would duke it out for a win. But suddenly at this table, people were competing with boats and flushes. Practically every other hand had something higher than a trip. Unlikely, yes. Do I make adjustments to my betting because of this fact? Well, yes and no. No, because each round is as random as the previous. I'm not going to bet differently because of streaks. But yes, when I see someone with three fours showing, I'm not going to bet along if all I got is four to an inside straight.

A very sound philosophy. kind of irrelevant to this conversation, though.

What you're doing is seeing a streak of mediocrity from one fencer, and not willing to accept the possibility that he or she can have a break-out day. That's no different than if I were to see a streak of high hands, and then bet high assuming that I'll have a high hand soon, too.

I'm perfectly willing to accept the POSSIBILITY that he/she CAN have a break out day. you, however, seem unwilling to accept the possibility that such an uncharacteristic is NOT a break out day, but rather a one time occurance not to be repeated. Past pattern doesn't dictate future performance, true. However, in activities based on skill, not luck, like fencing, it is a pretty good indicator. There is a fundamental difference between random games of chance and a sport in which skill largely dictates performance. If somebody is a D and has re-earned it many times and suddenly earns a B, though it may be a "break out day", it is more than likely a fluke performance which won't be repeated anytime soon. As I said, it IS my OPINION that some of these ratings are flukes. however, in hindsight, I was right about many of them (including oiuyt's friend) and wrong about some (some who are in my area, and have re-earned multiple times).

-m

[oiuyt]

Do I make adjustments to my betting because of this fact? Well, yes and no. No, because each round is as random as the previous. I'm not going to bet differently because of streaks. But yes, when I see someone with three fours showing, I'm not going to bet along if all I got is four to an inside straight.

Quick question- If all the previous hands had been won by pairs (or even low pairs, for whatever reason everyone had been getting just X-high hands) and you saw the above (someone else with three 4's showing and you have four to an inside straight) would you bet along?

Probability is the measure of your full knowledge of the uncertain event. It is not calculable. The more information you get about the uncertainty, the more "accurate" your probability assessment will be. Accurate means that you can set a probability to the possibility (mapped to some fungible yardstick like dollars), and experts would not be able to choose one way or another.

A probability alone has very little value if you don't attribute it to the corresponding possibility (i.e., outcome).

Not calculable? How so? MANY probabilities are calculable. How long will the average American live? That's calculable, the actuaries do it all the time. How much rain will San Francisco get in a year? Likewise. The number of heads I'll get if I flip a coin 100 times is calculable. Will it always be correct? No. It's a probability distribution that estimates the outcomes of a series of random events. Doesn't mean it's not calculable.

In this case we CAN attribute a probability to a specific outcome. What is the probability that fencer X will re-earn rating Y within timespan Z.

5000 to 1 is just as testable as 6 to 1. I just threw a number out there just as you did.

No, I threw a number (with some justification) out there just as YOU did. I was happy with using the common definition of the word, it was you who rejected the dictionary definition and started with the making up of numbers defining.

You need to understand probability better.

Funny, I was thinking something very similar....

(I'd consider myself a genuine "A" fencer, but my odds of getting an A in San Diego, say, is far lower than 6 to 1. Maybe 300 to 1.)

Okay, this is a good point. Because of the nature of our system (that fencers aren't expected to earn their rating every event possible) my 6 to 1 odds does break down somewhat. Clearly MOST A's at a div I NAC aren't going to earn an A (after all there are considerably more than 8 of them). My numbers hold a bit better when looking at local events, especially smaller/mid-sized ones where the numbers of ratings going out closely mirrors the numbers and quality of ratings already held. In point of fact, my friend (he of the B and the "How'd THAT happen" reactions) told me about his rating the day of a div III NAC (he'd earned the B more recently than the registration deadline, at which point he'd had a D). He was clearly the #1 seed (entering) that day. He finished 24th of 131, about where I would have expected if I hadn't heard about the B. Note that I mentioned div III NAC. This was against D/E/U fencers that he placed 24th.

Looking at my personal history I have never skipped a rating level (of the 12 levels earned in the three weapons) and I have never had a rating year go by without reearning a rating. The only rating that I've ever held that I didn't earn multiple times was a sabre B (I first earned the A 6 months after I earned the B, I've since reearned the A). Maybe my perceptions of how hard it should be to earn a rating multiple times is biased by my history, I don't know.

An impossible feat is more than a fluke, it's literally (and by definition) impossible. Flukes are improbable events (or highly improbable events). They can and do happen whenever there are random processes involved. They happen infrequently.

For those claiming that epeemike needs to reevaluate the level that he thinks people are, you're cycling back to our starting point. The whole "what is a fluke and do they actually exist" bit started as an off-shoot of epeemike saying that there were so many flukes that he was changing his expectations to be lower (and thereby redefining many of the flukes as expected).

Okay one last thing (at least for this post, I promise). Eric- would you agree with Achilleus that you making top 8 in San Diego would be a fluke? If so why? If not, why not? (and for what it's worth I also consider you an genuine A)

-B :)

[DHCJr]

Jumping two ratings does not always mean that it is a fluke. I will give an actual example. This is about Jens Stephan. On the 8th of September 2002, he came in 7th at his first USFA tournament, which happened to be an A2/B2 tournament earning his C02. The next month at the LBI he dropped down to 8th, but this was an A4 tournament with over 100 fencers, over 20 A’s and over 20 B's. He earned his A02 that day. In one month he went from unclassified to A. Dylan Bethel also went from a C to an A02 at the LBI coming in 7th. Do we call them both flukes or do we have to wait to find out if they were flukes. By the way, Jens just arrived from Germany.

I feel we are getting more A’s because we have more fencers, which is giving us more opportunity. Twenty years ago, I wouldn’t have expected an A2 level ever at my local division tournaments. Now it is rare, that it is not at least an A1 for Epee and this season all have been A2/B2 or higher. It also seems to be self-perpetuating, when you start having higher rated tournaments, more people want to come which equates to even higher rated tournaments. Why should an A go to a tournament, where they will be the only A and no chance to re-earn the A? We have more fencers and more opportunities. See my previous post for a further analysis.

[epeemike81]

Originally posted by BugabooX
It keeps coming back to the admission that we need to collect data on how many fencers earn a higher rating and then fail to do so again. ... AND somehow project prejudice onto that, believing that those people didn't deserve that rating to begin with.

not that they didn't deserve it, since they DID earn it, just that it CAN be considered a fluke and a weak rating. and that is only if they have had ample opportunity to re-earn and haven't. What I am NOT implying, which you seem to think I am, is that fluke ratings are undeserved ratings. If you earn it, you deserve it. the point is that our system rewards flukes WAY too much. the solution is to change the system.

-m

[epeemike81]

Originally posted by DHCJr
Jumping two ratings does not always mean that it is a fluke. I will give an actual example. This is about Jens Stephan. On the 8th of September 2002, he came in 7th at his first USFA tournament, which happened to be an A2/B2 tournament earning his C02. The next month at the LBI he dropped down to 8th, but this was an A4 tournament with over 100 fencers, over 20 A’s and over 20 B's. He earned his A02 that day. In one month he went from unclassified to A. Dylan Bethel also went from a C to an A02 at the LBI coming in 7th. Do we call them both flukes or do we have to wait to find out if they were flukes. By the way, Jens just arrived from Germany.

Why do people have such a hard time thinking in shades of grey? As I said, it certainly IS possible for such jumps to occur, they are just unlikely. BTW, the example you note is NOT a counter to what I was saying. In my example, this was somebody who had consistently earned D's for a long period of time, and then earned a B. does this mean that nobody fitting my description had made that jump? of course not. I KNOW somebody who has, in fact. most of the time, though, in my example it IS a fluke. A one time, unlikely result that won't be repeated until the fencer improves GREATLY.

As for whether to call them flukes, I have no clue, as I don't know them. you can come to your own opinion as to whether you think it was a fluke, but we can't tell for sure either way for probably another year.

-m

[a517dogg]

change of plans. i've decided that i'm going to make top 8 in both division one and junior, and reearn my A twice. no more fluke. so there.

just, ya know, to let you know.

[Inquartata]

Ah, if it were only as easy as deciding to do it...

Man, I feel all at sea. A raging argument upon which I have no strong opinion one way or the other! It must be chilly in Hell tonight...

[Peach]

Originally posted by Inquartata
Ah, if it were only as easy as deciding to do it...

Man, I feel all at sea. A raging argument upon which I have no strong opinion one way or the other! It must be chilly in Hell tonight...

Aaaaagh! Say it isn't so! I won't be able to sleep tonight.

For the record, I have never had an E or a C, and never re-earned my D or my B before earning a higher rating, though it took me something like 7 years. I prefer to think of it as "judicious training peaks" rather than "undeserved ratings."

[edew]

Originally posted by oiuyt
[...]
Not calculable? How so? MANY probabilities are calculable. How long will the average American live? That's calculable, the actuaries do it all the time. How much rain will San Francisco get in a year? Likewise. The number of heads I'll get if I flip a coin 100 times is calculable. Will it always be correct? No. It's a probability distribution that estimates the outcomes of a series of random events. Doesn't mean it's not calculable.
[...]
Okay one last thing (at least for this post, I promise). Eric- would you agree with Achilleus that you making top 8 in San Diego would be a fluke? If so why? If not, why not? (and for what it's worth I also consider you an genuine A)

-B

Probabilities are never calculable. In the case of no other information, you can rely on the statistical probability, which IS calculable. The "true" probability (whatever that means) of some uncertainty is never calculable. I took a course in decision analysis which discussed this. The explanation by this professor finally enlightened me from all those years of aggravation in trying to understand why probabilities don't seem to work as they (supposed to) do.

As a good example, take a plastic spoon. Toss it up. If it lands concave up (like a spoon set down by a waiter), call it heads. If it lands convex up, call it tails. What is the probability of it landing heads? Try throwing ten or fifty or five hundred trials and tell me if you're willing to stick with your initial "calculated" probability.

As for me making top-8, I would not call that a fluke. I know skills wise, I can do it. There are other rationalizations that can justify why I won't, but there are just as many rationalizations that can argue that I could. I know how good I am, relative to some fencers. I know how many of the top-24 I can and cannot beat, on a good day. If I draw the right bracket, I-COULD-GO-ALL-THE-WAY.

[epeemike81]

btw, I would like to point one thing out. I just conducted a search on this thread for the word "undeserved". Indeed, the first use was from me.

Originally posted by epeemike81:
No, there are not undeserved ratings.

I have repeatedly reitterated the point that I do NOT believe ANY ratings are undeserved, even flukey ones, yet people continue to accuse me of saying that....

Is it so hard to actually READ what is written?

-m

[epeemike81]

Originally posted by edew
As a good example, take a plastic spoon. Toss it up. If it lands concave up (like a spoon set down by a waiter), call it heads. If it lands convex up, call it tails. What is the probability of it landing heads? Try throwing ten or fifty or five hundred trials and tell me if you're willing to stick with your initial "calculated" probability.

what is the probability of it landing heads? not sure, but certainly NOT 50/50. it is weighted unevenly, which is calcuable. if that is taken into account, then yes, I would stick with calculated probability.

-m

[edew]

Why not fifty-fifty? A regular coin isn't symmetric either. One side has a silhouette of George Washington, the other has some state capitol, or a lighthouse or leafless trees...

How about a shallow spoon? If not fifty-fifty, which outcome should have the higher probability? Why?

Try this exercise: take a penny. Stand it up and spin it like a top (I'll let you decide on how to make that happen). Eventually, it'll fall down (unless you're really careful). Sometimes, it falls down with Lincoln on top, and sometimes it falls down with the Lincoln memorial on top. What's the probability now?

About fifteen years ago, I sat at a pizza parlor with two other mathematicians, one a professor and one a fellow graduate student. The professor was casually spinning the penny, and it landed twice memorial side up. I offered, "I bet you can't get it tails 20 times in a row." 15...16...17...18, uh-oh 18 times tails in a row. Now, we're scrambling around wondering what forces are causing the penny to fall tail side up. 19 tails, then one head. Then another string of 10 tails.

So, what do you think? Is coin spinning a 50-50 chance? Is there something nefarious about the table (usual rutted out pizza table) or the edge of the penny? Looked pretty square at the bottom, to all of us.

I went home, tried about fifty more trials and it didn't show any strings of tails.

You were taught that the probability of a fair coin toss is 50-50. But it was never honestly explained to you. Statistically, without any additional information, that is the default assumption for the probability: two possible outcomes, no advantages that you know of, statistics says 50-50. But when you see some outcomes, it changes your judgement.

(Actually, the Bayesian probability lecture is an argument FOR your fluke argument: you've seen Joe Fencer not do well time and time again. Your expectation is that he shouldn't do well the next time. However, there's lots of other variables in play here: Maybe Joe Fencer changed coaches. Maybe Joe Fencer decided to be serious about the sport. Maybe Joe Fencer got a lucky break and got into a nice bracket. Joe Fencer's fencing ability isn't independent of prior outcomes, unlike [we assume] coin tosses, coin spins, or spoon tosses.)

[epeemike81]

Originally posted by edew
Why not fifty-fifty? A regular coin isn't symmetric either. One side has a silhouette of George Washington, the other has some state capitol, or a lighthouse or leafless trees...

How about a shallow spoon? If not fifty-fifty, which outcome should have the higher probability? Why?

Try this exercise: take a penny. Stand it up and spin it like a top (I'll let you decide on how to make that happen). Eventually, it'll fall down (unless you're really careful). Sometimes, it falls down with Lincoln on top, and sometimes it falls down with the Lincoln memorial on top. What's the probability now?

About fifteen years ago, I sat at a pizza parlor with two other mathematicians, one a professor and one a fellow graduate student. The professor was casually spinning the penny, and it landed twice memorial side up. I offered, "I bet you can't get it tails 20 times in a row." 15...16...17...18, uh-oh 18 times tails in a row. Now, we're scrambling around wondering what forces are causing the penny to fall tail side up. 19 tails, then one head. Then another string of 10 tails.

So, what do you think? Is coin spinning a 50-50 chance? Is there something nefarious about the table (usual rutted out pizza table) or the edge of the penny? Looked pretty square at the bottom, to all of us.

I went home, tried about fifty more trials and it didn't show any strings of tails.

You were taught that the probability of a fair coin toss is 50-50. But it was never honestly explained to you. Statistically, without any additional information, that is the default assumption for the probability: two possible outcomes, no advantages that you know of, statistics says 50-50. But when you see some outcomes, it changes your judgement.

(Actually, the Bayesian probability lecture is an argument FOR your fluke argument: you've seen Joe Fencer not do well time and time again. Your expectation is that he shouldn't do well the next time. However, there's lots of other variables in play here: Maybe Joe Fencer changed coaches. Maybe Joe Fencer decided to be serious about the sport. Maybe Joe Fencer got a lucky break and got into a nice bracket. Joe Fencer's fencing ability isn't independent of prior outcomes, unlike [we assume] coin tosses, coin spins, or spoon tosses.)

all of this, though very interesting, is irrelevant to my statistical argument (re-earn quotient and all that). Would you care to comment on THAT argument? Also, a coin is MUCH closer to balanced than a spoon, though that, again, is neither here nor there regarding the real discussion here.

-m