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

Re: Constitutional voting, definition of cummulative prefererence

Raul Miller wrote:

>Hmm.. the constitution already states that votes are cast by email.
>[Which makes a lot of sense, when you think about the technologies
>involved -- email queues, web doesn't, and signing of email is a well
>established technology.]  And, personally, I'm not comfortable with
>"ranked higher" as a circumlocation.  But it's an interesting point
>you raise.

I agree that right now, the best way to collect ballots is to use an e-mail
interface.  However, amending the constitution requires major effort, so it
seems better to me to have terms and procedures defined with as much
generality as possible, while still remaining clear and unambiguous, in
order to minimise the need for future changes.  However, if an e-mail ballot
procedure is described in the Constitution as the only way to carry out
balloting, then this would also need to be changed for a more general
wording to be worthwhile -- I hadn't noticed that.

>> I am not exactly sure why you are defining 'cumulatively preferred' to
>> indicate transitive majority preference between options, so I can't say
>> certain whether or not this is a good idea, because I don't know what you
>> intend to use it for.
>I'm aiming for a "minimal change" fix for the apparent ambiguity of the
>current constitution.   I'm thinking about proposing an amendment to the
>constitution where "Dominates" is defined as strict cumulative preference
>(A is cumulatively prefered to B and B is not cumulatively prefered to A).

The only ambiguity I see with the current constitution is that it cannot
cope with circular ties.  Fortunately, these are rare, so this has never
been an issue for Debian (all previous elections have had a Condorcet
winner).  One thing I do as part of my involvement in the EM list is to
carry out simulation studies of voting methods, and these suggest that
circular ties will occur naturally about 5% of the time, so this isn't that
big an issue.  It's a minor bug which could result in some embarrassment and
confusion at some point, but it is not likely to affect anything in the
short run.

A minimal fix for this problem is to replace A.6.3 with A.6.4, and then
create a new A.6.4 which uses the Smith set to restrict the group of
potential winners that are then subjected to the STV count rule (or for
something even simpler, just leave it out entirely, in which case *all* the
candidates would be selected from using STV, if there is no Condorcet
winner).  This doesn't address the issue of mixing options with different
supermajority requirements, but I think Debian's existing rules can
unambiguously cope with that -- they're sub-optimal, but serviceable.  I
don't see anything wrong with the quota requirements as they're written.

>> 3) It is not necessary to define 'cumulatively preferred' unless it is
>> used as part of a voting method definition, and I'd need to see the
>> whole method in order to judge the merits of your system.
>As I indicated above, I'm considering the implications of explicitly
>specifying that an option "Dominates" another only where the first
>option is transitively preferred to the second, but the second is not
>transitively preferred to the first.
>[I'm aware that there are many alternate voting methods.  But, I think
>we need to at least consider options based on the "don't fix what
>ain't broke" approach.  If we completely rewrite large sections of the
>constitution we may create future problems which we won't notice for a
>year or two.]

I disagree with this approach.  If you redefine the underlying concepts used
in pairwise voting, you may be able to make minimal wording changes yet
create profound changes in voting results.  On the other hand, major changes
in wording can have little or no effect on outcomes.  Consider:

The existing 'Concorde' voting rule is a pairwise method which satisfies the
Condorcet Winner criterion.  The Schulze method (as well as almost all other
pairwise methods) also satisfy this criterion, and since there is a
Condorcet winner about 95% of the time, all these methods will agree on
results to the same degree.  Therefore, a change to Schulze would eliminate
the ambiguity without having any noticeable effect on outcomes (except in
the remaining 5% of cases, where Schulze would yield a winner, and Concorde
would leave people scratching their heads).  However, if you redefine the
pair-comparison relationship to mean something else, this could profoundly
affect results in *all* cases, not just when there are circular ties -- your
new method might not even satisfy the Condorcet criterion, and all of its
other properties would be completely unknown.  This *would* be a major, and
probably undesireable change.

It seems to me, given that:

1) Amending the constitution is a major undertaking, and
2) There is no urgent need for constitutional change

it is better to look at all the issues related to voting closely and
carefully, pick the *best* solutions after thorough discussion of the merits
of every alternative, and then propose an amendment that addresses all the
important issues.  It's better to take the time to fix this properly, than
submit a quick patch that breaks more than it fixes.

Norm Petry

Reply to: