inflated ratings

[Inquartata]

So maybe if Eric were to win at San Diego, going through all of his DEs 15-0, we could call that a fluke? At least until he did it again?

On the subject of mathematicians, I am contrite, I hereby retract my criticisms and boldly assert that CD's of Wiles, Wolfram, Conway and Thurston reciting equations ought by rights to be at the top of the music charts, instead of those darned rappers, and they should have their own talk shows as well, since theirs is the most fascinating topic for conversation throughout all levels of society!

[edew]

Originally posted by Inquartata
So maybe if Eric were to win at San Diego, going through all of his DEs 15-0, we could call that a fluke? At least until he did it again?

On the subject of mathematicians, I am contrite, I hereby retract my criticisms and boldly assert that CD's of Wiles, Wolfram, Conway and Thurston reciting equations ought by rights to be at the top of the music charts, instead of those darned rappers, and they should have their own talk shows as well, since theirs is the most fascinating topic for conversation throughout all levels of society!

That would indeed be a fluke. (Having Wiles', Wolfram, Conway, et al's CD on the top-10 charts, that is. The other is the natural burden of expectations. Sigh.)

[edew]

Originally posted by oiuyt
DHC- You've mis-read epeemike's method. The use of 5 opportunities was the minimum opportunities floor before he would consider having enough data to look at someone's results. At that, the floor is probably too low. With a higher floor we'll have problems gather enough data (even if we were trying to gather data which we're not) for A's, but his method should hold for any of the other ratings as well.

I'll try to re-word what epeemike was saying. Hopefully I won't maul it, and maybe my re-statement will make sense to someone to whom the original phrasing didn't.

Take all of the people with rating X.

Determine for each of them the frequency with with they re-earn that rating (# of re-earns/# of re-earn opportunities) after discarding those with too few re-earn opportunities.

Graph the frequencies.

Look for the peak frequency (ie people with rating X re-earn rating X most often at a rate of Y). Anyone who re-earns less frequently than .1 * Y is flukily over-rated. Anyone who re-earns more frequently than 10 * Y is flukily under-rated.

Personally I think that epeemike was too extreme in what he calls flukes, but he got the numbers from similar methods in fields other than fencing-flukiness evaluation (he may have mentioned WHERE it came from, I don't recall. He mentioned that he was not the source of the 10% figure). Personally I think it's much more likely that anyone in the bottom 10% in frequency (or top 10%) is probably a fluke (obviously top 10% of the top range (A) is not a fluke, we just don't have the higher category to which they should belong). This wouldo be a noticibly larger number of people than epeemike's method would select.

Mike- check what I said, it wouldn't surprise me if I've missed some bit of what you said or worded it incorrectly.

Mike does not propose anywhere what the threshold frequency for flukiness is for any rating (it would be different for A's than it is for C's). It would have to be empirically determined and, AFAIK, there isn't a large data source which would provide the needed information. Perhaps if AskFred.net gained more widespread use then after some time we would be able to use the results contained therein to make such measurements.

-B

You raise a very interesting observation here that no one has addressed. There is a big difference between an A and a B or C or other classifications. One difference is obvious: an A fencer is more likely to beat a B fencer, blah, blah, blah. But the big difference is that there is no upper bound to an A classification.

So, obtaining A's at a lower frequency is not necessarily a fluke, as opposed to re-earning B's or C's at a lower frequency (would indicate flukiness in getting them in the first place).

I am an A-rated fencer. So is Jed Dupree, but we can all agree that he's a much better fencer than I am. And there's Ralf Bissdorf, who would be an A rated fencer in the US after one (big enough) tournament. And Ralf is probably so much better than Jed than Jed is to me (I'm not exactly sure on this, but the gist is correct). And potentially, one can be infinitely better than Bissdorf and still be an A.

So, there's no upper bound to an A-level fencer. Thus, one needs to make adjustments to what qualifies as an A-rated fencer on the flukosity meter that one need not make for B's and C's. While the best B probably can beat the worst A most of the time, the average B should lose to the average A. And the claim is that the best B in the tournament is supposed to lose to the worst A in the tournament (otherwise, why else do we seed tournaments as we do?). So there are upper bounds to B's and C's and below, which we don't have for A's.

So if you give B's and C's 1 in 10 tries to re-earn their classification before their original rating is considered a fluke, you might have to turn down the level for A's. Maybe one in 20 tries because the number and quality of A's may be infinitely high.

A B fencer (or C or lower) may attend B-specific or C-specific (i.e., Div II) events. There, there is an upper bound on the quality of the fencers.

To be fair, most local competitions aren't going to be a Group B tournament by fiat. They're generally opens. But you're also not going to have Jed Duprees showing up to those events either. Thus, local events will have a natural lower level overall, and so your frequency for fencers to re-earn should be higher as a consequence.

[Purple Fencer]

Originally posted by fencerX
Why pick on me? Now that's not very nice . . .

Heh...take the director's written exam sme time...you ALWAYS get scfewed on rules interpretations!

Welcome to the board!

[epeemike81]

Originally posted by DHCJr
I believe we have reached an impasse. We cannot test, if there are flukes, since we cannot agree what a fluke is. For example and I hope I get this right, Epee Mike believes a fluke is someone who has not been able to re-earn an A after 5 opportunities.

Sorry, wrong again.

Once again, the point is that you need a large amount of data to determine what is and isn't a fluke (statistically speaking). five opportunities is just the minimum to even be useful information if a probability curve was drawn up.

btw, EDew, It doesn't really matter if it is or isn't a bell curve. the point remains the same about the fringes no matter what type of curve it is (within reasonable statistical expectations).

-m

[epeemike81]

Originally posted by DHCJr
Also, the USFA already takes care of (your definition) of a fluke. If you do not re-earn your rank after a certain amount of time, you loose the rating.

the four year thing is too long to be considered a fluke fixer (note: I am NOT suggesting ratings dropping sooner, just that it doesn't help with this problem). The major problem with the USFA's system is that it takes high water mark, so when a fluke does occur, the person is overrated for the next four years. In a numerical spectrum rating system (like chess) flukes have very limited impact on your rating, long term. The fact that flukes exist is clearly not the USFA's fault, it is a statistical fact of life. however, how much impact the USFA lets them have on ratings certainly IS the USFA's fault.

-,m

[three_hundred_fifty_five]

This thread has two points of views.
One is that people have inflated ratings because the tournament a person received a new classification did not have a strong pool of fencers.
The second one is that people have earned a new classification through luck or it was a fluke.

The classification system is a way to gauge a person's level but it's not perfect. However, it does give an approximation of someone's skill level.

This is all moot if your goal is beyond the confines of the USFA classification system and is targeted at bigger things like the Olympics or World Cups/Championships.