On Thu, Nov 23, 2000 at 09:33:58AM -0500, Raul Miller wrote:
> On Thu, Nov 23, 2000 at 11:31:44PM +1000, Anthony Towns wrote:
> > Suppose you have the following ballot, and the following votes:
> > A: Remove non-free
> > B: Support non-free
> > F: Further discussion
> > 60 votes ABF (Would prefer to remove non-free, but either is okay)
> > 50 votes BFA (Insist that non-free not be removed)
> > then A dominates B and F, 60 to 50, and B dominates F 110 to 0. The
> > clear condorcet winner is A since it dominates all the other options
> > (ie, it's what the majority of voters prefer) yet it doesn't have enough
> > votes to claim a 3:1 supermajority.
> Nope.
>
> With a 3:1 supermajority, the 60 raw votes for A are equivalent to 20
> real votes, so A does not dominate B.
I'm no longer at all sure what you're talking about then. Note that I'm
not referring to the system described in the constitution here, but to
general Condorcet voting schemes.
Consider another possible vote outcome:
60 votes A
21 votes BAF
in which case:
A dominates B : 60 to 21
A dominates F : 81 to 0
B dominates F : 21 to 0
If you go ahead and divide A's comparisons by 3 (assuming you need a 3:1
supermajority for A to succeed), you end up with:
B dominates A : 21 to 20
A dominates F : 27 to 0
B dominates F : 21 to 0
and B wins, in spite of everyone being happy with A, and almost everyone
prefering it.
> Furthermore, you've not specified quorum -- F gets quorum votes,
> automatically.
I'm not sure where you're getting this from at all. The constitution handles
quorums as follows:
8. If a quorum is required, there must be at least that many votes
which prefer the winning option to the default option. If there
are not then the default option wins after all. For votes
requiring a supermajority, the actual number of Yes votes is used
when checking whether the quorum has been reached.
You don't add pretend votes to the default option. Indeed "adding" votes
isn't entirely meaningful, in Condorcet systems votes indicate a preference
amongst options, they don't get assigned to an option.
> > > > b) A tie for first place (ie, the schwartz set has two or more
> > > > options in it), where "further discussion" is one of the
> > > > equal winners, and it pairwise beats whatever is chosen as
> > > > the real winner.
> > > If this is a true tie, we need a tie-breaking vote (casting vote). That
> > > would mean it's up to the leader for the stuff we're talking about here.
> > A tie for first place in condorcet terms is more complicated than that. For
> > example:
> > 30 ABC
> > 25 BCA
> > 35 CAB
> > A dominates B (65 to 25), B dominates C (55 to 35) and C dominates A
> > (60 to 30) to complete the cycle. A single casting vote isn't enough to
> > reverse any of them: you'd need to change at least 10 "casting" votes,
> > which could reverse the B dominates C pair and let C win by dominating
> > all others. This is roughly how Tideman works, I think (ie, by choosing
> > the option that'd require the fewest swaps to make it the clear winner).
> According to the debian constitution, in this circumstance you eliminate
> B and retally:
Well, under A.6(3) your first action is to discard all options (since
they're all dominated by at least one other) and ignore references
to them in ballots. A.6(4) doesn't apply since nothing dominates all
others, A.6(5) doesn't apply since there aren't any options remaining.
A.6(6) could apply, I suppose, in which case any of A, B or C would win
depending on what the DPL decided.
And I'm not trying to describe what the constitution says happens because
I think it's broken. I'm trying to understand what *should* happen.
Cheers,
aj
--
Anthony Towns <aj@humbug.org.au> <http://azure.humbug.org.au/~aj/>
I don't speak for anyone save myself. GPG signed mail preferred.
``Thanks to all avid pokers out there''
-- linux.conf.au, 17-20 January 2001
Attachment:
pgpe0oHj_n569.pgp
Description: PGP signature