Quote:

>I heard from my division secretary in CT two week ago that the USFA is

>thinking of adopting a new rating system in '96. This would be a point

>system similar to the one used in Chess competitions. This system would

>take into account the number of tournaments you enter, the number of

>participants in those tournaments and where you placed. It is my

>understanding that this method would require each division to appoint

>someone to run this ystem on computer as the ratings are tabulated

>nationally thoughout the season. I think it might create a more acurate

>picture of how you compare to other fencers vs. the system in place now.

>Any Comments?

Well, chess ratings (really Elo ratings, named for the person who

did the serious mathematical work on chess rating systems) should work fairly

well. Simplified, the idea is that everyone has a rating. If you beat

someone, you gain points and they lose points. The number gained and lost is

equal and is proportional to the result "expected" from the rating difference.

Once the ratings have settled down, comparing two people's ratings

should also tell you the approximate chances of each fencer winning. It's

great for seeding tourneys as well, and makes it much easier to compare

yourself against other fencers - certainly better than the current A, B, etc.

Example: say you have two fencers rated 1400 (which we'll define as

the rating of an "average" fencer. The actual values are fairly arbitrary;

differences between ratings are all that have an real meaning.) If they

fence, the expected outcome is 50-50, and the winner will gain (say) 16 points,

and the loser lose 16 points. If they fenced again, the difference in rating

would be 32 points (1416 vs 1384), and the expected result might be 55-45.

If the higher-ranked fencer won again, he'd gain (say) 14 points and

the lower-ranked would lose 14. If the lower-ranked upset the higher, he'd

gain (say) 18 points, and the higher-ranked would lose 18.

If the fencers were rated 1500 and 1300, you'd expect the higher-

ranked fencer to win about 80% of the time or so, and so the higher-ranked

fencer would get only a few points for a win (say 6), while the lower-ranked

would get a lot (say 26).

All of this is pretty trivial to handle on computers, and isn't all

that hard by hand (it's been around for >30 years in chess).

Note that the number of points at risk (32 in the above example) can

be adjusted to deal with the amount of variablility in performance seen in

the sport (i.e. how much the performance of an individual is likely to vary

on a given day). This can be figured out fairly easily by seeing if the

predictions of performance match the expected results over a large number of

fencers and other mathematical tools. I'd assume they've done this.

There are various ways of dealing with newcomers to give them an

initial rating once the system is running.

--

Randell Jesup, Scala US R&D, Ex-Commodore-Amiga Engineer class of '94

#include <std/disclaimer>

Exon food: <offensive words censored by order of the Senate>