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

Re: Call for votes for the Condorcet/Clone proof SSD voting methods GR

Hamish Moffatt wrote:

For the benefit of the average non-voting-geek Debian developer,

could the proponents of this amendment please explain what problem
it attempts to solve, with real life examples?

The main problem is that the existing voting system as described in the
Debian Constitution is poorly defined, with some inherent
contradictions.  It wasn't looked at too closely when the Constitution
was first developed.  Among other problems:  It tries to implement a
Condorcet-based voting system, but calls it "Concorde" instead; it has
the property that if there is no clear winner (what Condorcet proponents
would call an "ideal democratic winner" or a "Condorcet winner", then
all the options are rejected before a winner is chosen from the
"remaining" options, etc.

There were also severe questions as to how the supermajority
requirements should be implemented.

These issues came to a head when a GR was proposed a modification to the
Social Contract that would eliminate the commitment by Debian to support
the "non-free" section.  It became clear that any ballot would contain
an amendment to the SC (viewed as requiring a supermajority to change),
a proposed policy statement (viewed as requiring only a simple
majority), and the "default" option.

A review of the constitution to figure out how to conduct such a vote
provided more confusion than answers.  So the secretary decided to
shelve the non-free GR until the voting issues were cleared up.  It's
been a while, however.

An explanation of why we need such a complicated system at all would be
interesting too.

Simple reason:  When you get more than two choices to vote for, all
election methods suck (lookup "Arrow's Theorem" for a proof of this),
but some suck more than others.   We want to find an election method
that has a minimal amount of suckiness for our goals.

More complicated reason:  While normal voting methods and procedures are
sufficient for deliberation in person, where everything can be settled
by multiple procedural votes, online deliberations have more complicated
requirements, requiring more complicated voting procedures.

We have the following complications:

1) We want to have as few votes as possible to settle an issue, since
each vote requires two weeks to run in order to get the most input.
This means that we can't follow a traditional procedural amendment
process -- each vote has to have all proposed (and seconded) amendments
on it, and the procedure has to select from among them.

2) We want to be conservative, and err towards continuing discussion, if
we can't achieve some sense of "consensus" on an issue.  As such, we
want to require continuing discussion to always be a ballot option

3) Very often more than a yes/no or either/or choice -- 3+-way decisions
are the norm and possibly required.  This is because of the multiple
variant issue mentioned in (1) above, and because of the "continue
discussion" option mentioned in (2).

4) Some options must be approved by a supermajority.  Other options
don't.  So some votes will be a mixture of supermajority requirements.

5) And we want some justifiable semblance of majoritarian rule.

The most common method, where everyone gets to vote for one (and only
one) option, and the option with the most votes wins suck big-time,
because in the presence of lots of options, the vote tends to get split
to the point where it's hard to justify saying any particular option has
a majority.

The methods chosen by Debian, both in the current constitution and in
the proposed constitutional amendment, are variants on a method
developed by a scientist named Condorcet over 200 years ago in France.

Condorcet was trying to answer the question "what does majoritarian rule
mean when there are three or more candidates in an election?".   In a
two-way election, it's easy to see what "majority rule" was -- whoever
got the most votes would win.  But in a three-way race, it's possible
for no candidate to get the majority of votes.  He proposed the
principle that if there was a single candidate who, when pitted against
all others in two-candidate, one-on-one races, would win all those other
races, then that single candidate should be the winner.  Both the
current and proposed election methods are at least supposed to have that

But Condorcet's principle doesn't say anything about what to do when
there is no single candidate that would be undefeated against all other
candidates one-on-one.  Some evidence (via simulated elections) seems to
show that in 95% of elections there will be a "Condorcet winner", but in
5% there won't.  Hence any election method has to be enhanced to deal
with that issue.  The particular technique proposed here is called
"Cloneproof Schwartz Sequential Dropping" (cSSD), which refers to some
specific properties of the technique -- it is impervious to certain
types of election strategy.

Standard cSSD solves many problems; it allows for as many ballot options
as we choose, including a default "keep talking" option, it resists many
forms of ballot strategy (i.e., ways to manipulate your vote to enhance
favorable results, other than simply voting the way you really feel),
etc.  It's a good method.

However, supermajorities were not considered by the folks who developed
and evaluated cSSD.  To support supermajorities more modifications were
needed.   The constitution as written does a poor job at supporting
supermajorities, so I will only discuss the proposed change.

The major modification was to adopt, on top of cSSD a version of
"Approval" voting.  A standard approval ballot has one checkbox next to
each option, and the voter is allowed to check as many boxes as they
choose.  Approval is known to work well for supermajority requirements.
As an example, the annual balloting for the Baseball Hall of Fame uses
approval with supermajority.  Each year newly eligible players are added
to the ballot, along with players from previous years.  If 75% or more
of the voters approve a player for the hall of fame, the player is
added.  If fewer than 25% of the voters approve a player, they are
dropped from the ballot, never to be reconsidered.

For Debian, which uses a ranked ballot (voters rank options in order of
preference) in order to support Condorcet voting, the choice to propose
was to treat ranking relative to the default "keep talking" choice as a
proxy for approval.  If a voter prefers to keep talking over an option,
it counts as a "not approved" vote.  If a voter prefers an option over
"keep talking", it counts as an "approved" vote.  To be considered a
successful supermajority option, a supermajority of voters must vote
"approved" by prefering it to "keep talking".  If it doesn't get a
supermajority approval rating, it is dropped from consideration.

The proposers of this amendment also feel that it is worthy to drop from
consideration any other option that is not approved by a minimum number
of voters or has more not-approved votes than approved votes.  Only
votes that have a minimum number of approved votes and are approved by
more people than don't approve it are considered in the cSSD process.

Follow-ups to debian-vote, please.

As an aside, where is the constituition located on www.d.o, and why
doesn't the search engine find any references to it at all?


Reply to: