[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Better quorum change proposal (with justification)

On Sat, May 24, 2003 at 09:24:34PM -0400, Nathanael Nerode wrote:
> Consider if all the people who, in my original example, ranked A=D,
> instead ranked A *below* D.
> 19x ADB
> 19x ABD
> 1x BDA
> A vs. B
> 38 to 1
> D vs. A
> 20 to 19

Huh? YM A vs D: 38:1.

> B vs. D
> 20 to 19
> There's no Condorcet winner.  

And A meets quorum and wins.

If you meant:
	19x DAB
	19x ABD
	 1x BDA
You have:
	A beats B, 38:1
	B beats D, 20:19
	D beats A, 20:19

and under the proposed system, A is dropped for lack of support, and B wins.
What are the possible outcomes?

	A wins, in spite of not having a quorum of support
	B wins, in spite of A being preferred by a vast majority.
	D wins, in spite of B being preferred by a clear majority.

A is unacceptable, but could easily have been made acceptable by the people
who support A getting a single additional vote. This has never been a problem
in the past.

As A is unacceptable, the qualification against choosing B as the winner
is fairly weak. Further choosing D is unacceptable, since that causes
process problems in the future: what do we do instead? So choosing B is
the least objectionable, which is to say best, course of action.

Note that without quorum, A is dropped any way, since it doesn't defeat
the default option by its majority requirement, and B wins then too. The
way the proposal is constructed, the elector with a casting vote can
never choose to void the election if any options are acceptable.

> With Manoj's quorum, A is eliminated and B wins by a hair, same as with 
> the equal rankings.  But 38 people prefer A to B and only 1 person 
> prefers B to A.  Now B is implemented.  Some people may think that this 
> is desirable, of course.  I certainly don't.

A majority of 20:19 think it's more desirable than doing nothing,
by assumption.

> Consider my original example:
> 19x A=DB
> 19x ABD
> 1x BA=D
> Under Manoj's system, B wins!  B is implemented!

> A new ballot ("Fix the results of the stupid ballot we just had!") 
> proposes A vs. "Keep the present system" (B) vs. "more discussion" (D).  
> The first set of voters hated B and want to get rid of it before 
> discussing things further, so they change their votes:
> 19x ADB
> 19x ABD
> 1x BA=D

So, if the first set of voters find A acceptable, why didn't they vote
that way originally? They could have, quite easily.

> ---
> The problems are not really present for the Technical Committee, since 
> the quorum is 2.  For technical reasons, this cannot cause any odd 
> behavior.

Not due to quorum perhaps, but you can still get:

	3 ABD
	3 DAB
	2 BDA

A beats B, 6:2; B beats D 5:3; D beats A 5:3. A doesn't make its majority
requirement, and is eliminated, so B wins. Cloneproof SSD alone would
have eliminated B beats D and D beats A, leaving A or D as the winner.

> --
> It also seems unlikely for the aforementioned problems to happen in a GR 
> when the number of voters is >=2*R, where R is the quorum requirement.
> Unfortunately, it *can* happen with an arbitrarily large number of 
> voters.  

Yes, and it's possible for a meteorite to hit master.debian.org just as
the vote's closing and for D to thus win by default, even if _everyone_
voted for A!

It's not a likely thing to happen both historically, and because in
the balance, it's beneficial for you to express full preferences,
wherever possible.

I don't think you can actually get any problems with quorum with more
than 12R voters, actually. That's 570 of 1000 developers voting, so isn't
particularly unbelievable. You also need a linearly increasing number of
options to manage this, afaict; I think you need at least 7 options to
have a vote where an option beats the winner, but doesn't make quorum,
with 12R voters, as well as an utterly unbelievable distribution of

> How?  When there are lots of options and lots of rankings equal to 
> the default option, as well as a highly splintered electorate.

You'll note that we try to avoid having a splintered electorate --
you know, decision by consensus and all that; and have a voting system
that's designed to find a compromise that's as acceptable as possible.


Anthony Towns <aj@humbug.org.au> <http://azure.humbug.org.au/~aj/>
I don't speak for anyone save myself. GPG signed mail preferred.

  ``Dear Anthony Towns: [...] Congratulations -- 
        you are now certified as a Red Hat Certified Engineer!''

Attachment: pgpGFwu3pMvul.pgp
Description: PGP signature

Reply to: