On Wed, 26 Sep 2007, David Artman wrote:
A LITTLE over my head, but one thing stands out:
From: Andrew Plotkin <erkyrath@xxxxxxxxxx>
I don't have the IGDC#5 ballots in front of me, but I'm positive that
changing them to "unlisted entries tie for last" would change the outcome.
What about my "Guestimate" summation method (see the Google spreadsheet,
sheet two):
http://spreadsheets.google.com/ccc?key=pqnM2cZcPLm0qxv1F1TILMQ&hl=en
In that page, I applied the next-lowest rank to unranked games, NOT last
rank. The presumption is that--though equally bad--they are still
"tieing" for the next-lowest position in the ranks.
That's exactly the same as what I said: tied for last place. Remember,
only relative positions matter.
(If you play exactly two games, and mark A in first place and B in "tenth
place", that's just second place. Either way, it boils down to "I think A
beats B.")
The solution I'd actually recommend is to say: any game which is listed on
fewer than N ballots is eliminated.
(In case you were wondering, I'd hand-wave N as around 5% of the number of
voters.)
An interesting compliment to my (proposed earlier) idea of disregarding
any ballots with fewer than, say, half the competitors ranked.
That doesn't solve any real problems for the algorithm. (Although it gives
you a more educated voting population, if that's what you're aiming at.)
In general, the Ranked Pairs method is happier when you give it more
information. If somebody plays just two games and votes one over the
other, that gives the algorithm a fact to make use of. You might as well
let it.
--Z
--
"And Aholibamah bare Jeush, and Jaalam, and Korah: these were the borogoves..."
*
You don't become a tyranny by committing torture. If you plan for torture,
argue in favor of torture, set up legal justifications for torturing
someday, then the moral rot has *already* set in.