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

Re: Norman Petry and I (Ossipoff) recommended CSSD, but Schwartz Woodall is a better voting system for Debian

>> But what if the B voters know for a fact that the A voters are
>> completely conscientious, responsible, and co-operative, and that the
>> A voters are sure to rank B over C?
>> Sure, I agree that there are a number of good reaons why the chicken
>> dilemma needn't be a problem. But, even if it isn't a full-fledged
>> problem, it remains a _nuisance_.
> For me to consider this a nuisance, I would have to see that there is a
> practical case where a rational group of voters might use this strategy to
> sway the election in their direction.  Clearly, there are cases where CSSD
> would reward strategic voting *if* a voting bloc had perfect knowledge of
> how everyone else would vote.

The C voters in my example don't really have any better information
than do the A voters. In a public political election, it might be
impossible for the C voters to organize a defection without the A
voters hearing about. More relevan to organizational voting, the A
voters could rank B 2nd with some probability less than unity, with
the effect of giving B partial 2nd place support--just enough to help
B defeat C if B is the bigger of {A,B}. Of course, as you said, the B
voters don't have perfect knowledge, or better knowledge than the A
voters have,and so the B voters would have to accept that they're
definitely risking a C victory if they defect. That sollution is from
Forest Simmons, and I call it "strategic fractional rating" (SFR). It
was originally proposed for Score and Approval, but it woud work with
CSSD as well.

That's why I cal the chicken dilemma a nuisance instead of an outright
problem. But, the situation can't be ignored by the A voters, and
that's why I say that it remains a nuisance, though not a problem.

>But is that realistic?

Sure. Members of a mutual majority might have a lot of rivalry among
eachother. For an example of riivalry and fractiousness among a group
who need to stand together,, look at the American progressive
political parties. And look at the set of voting-system reform
advocates. We need look no farther for examples and proof of the
fractious rivalry that can lead to the chicken dilemma.

Debian is an amicable and civilized organization, and I understand hat
Debian doesn't have a chicken dilemma. But in general, for public
political elections, and also for many organizations, the chicken
dilemma remains a likely nuisance.

> In your original
> scenario, the preferences are:
>  99: A>B>>C
>  2: B>A>>C
>  100: C>>(A=B)
> But how are the B voters to know this with certainty?

The example's assumption is that the B voters know for a fact that the
A voters  are thoroughly conscientious, co-operative and trusting.
Does that mean that the B voters won't take advantage of their trust?
No, it means that they will. Abusing other people's trust is what our
politics and  society are all about.

I admit that that is an idealizing, simplifying assumption . Of course
the B voters' predictive information is no better than that of the A
voters, and that's why the chicken dilemma is only a nuisance instead
of a problem.

> Even one voter
> preferring A>C>B (or at least, voting that way) is sufficient to undermine
> this strategy

Sure, but my exaample was intentionally written close-to-the-line, a
numerically extreme example.

> The reason we care about these properties of voting systems is that we want
> to avoid rewarding strategic voting.  I posit that a strategy that allows
> for a margin of error of <1% in the attackers' understanding of how all
> other voters will vote before it yields a pathological outcome instead is
> not a very rewarding strategy at all.

Sure, but, if the B voters are morally low enough (and in our society
that'must be regarded as a given) then the only thing that will deter
their defection will be the fear that the A voters are making good on
their deterrent threat to defect as well. The A voters have to
consider that, and I call that a strategic nuisance. A neat solution
for the A voters is the "strategic fractional rating" (SFR) that I
described above. That creates the necessary deterrent uncertainty for
the B voters. But the fact that the A voters must deal with the
situation qualifies it for the word "nuisance".

> All other things being equal, it would of course be better to address the
> chicken dilemma.  However, you bear the burden of demonstrating that all
> other things actually are equal.

I fully understand that Debian doesn't have a chicken dilemma, and
therefore doesn't need a chicken-dilemma proof method such as Schwartz
Woodall. I just felt that I should mention it.

>The method you propose has been evaluated
> with respect to a couple of important criteria (the Mutual Majority
> Criterion and the Condorcet Criterion);

...and with respect to the (usually) important property of freedom
from chicken dilemma.

> but what about other criteria that
> CSSD satisfies?  There are lots of criteria that are interesting to students
> of voting, and it's well known that some of them are mutually exclusive;

Quite so. There are very many criteria proposed. Anyone saying that
one criterion is more important than another should tell why.

The Mutual Majority Criterion and freedom from the chicken dilemma,
together, are a very powerful combination of properties. That
properties-combination means that a mutual majority has no reason to
do other than rank sincsrely. By doing nothing other than that, they
can _ensure_ that the winner will come from their MM-preferred set.
They can do that  by freely choosing among their MM-preferred set, by
sincere ranking. I call that the "rank-balloting ideal."  it's what we
want when we advocate rank balloting, but few methods can really
attain it. IRV, Woodall, Schwartz Woodall, and Benham achieve it for a
mutual majority.

Additionally, Woodall, Schwartz Woodall, and Benham don;t give
_anyone_ a need to do other than rank sincerely. A little temptation?
Maybe, but not a need.

That's saying a lot. That's something that just can't be said for most
voting systems.

> before making any changes to our voting system, we should understand the
> consequences fully, not just with regards to a couple of handpicked criteria
> that are superficially the most important.

But "handpicked" seems to imply that I chose the criteria after first
choosing the voting-system prooposal, and chose the criteria in order
to support the proposal. It was the other way around. Realizing the
nuisance that the chicken dilemma can, and surely will (in typical
electorates) cause, and realizing the powerful strategy-free benefit
of the Mutual Majority Criterion--provied that there isn't a chicken
dilemma--I was very impressed with IRV's possession of that
combination of properties. Then I realized that, via hybrids of IRV
and Condorcet, Condorcet's Criterion can be added to those desirable
properties, producing the most strategy-free method that you an get.
That was before Benham told me about James Green-Armytage's article.
Green-Armytage, in his article, states that conclusion that I stated a
few sentences ago, about IRV Condorcet hybrids.  ...the conclusion
that they offer an unequalled freedom from strategy need. He's right.

Sure, as you said, you get one criterion by giving up another.
Changing from CSSD to Schwartz Woodall, you'd be giving up Mono-Raise.
IRV opponents make much of Mono-Raise. it's tempting to play-up a
criterion that seems to support one's recommendations. I've given into
that temptation myself, when I used to look for ways to criticize IRV.
But, as I said, CSSD (like all Condorcet methods) fails several other
monotonicity criteria, inclusing Participation, Mono-Add-Top, and
Mono-Add-Unique top (defined in an earlier post here). So we shouldn't
place too much importance on Schwartz Woodall's failure of Mono-Raise,
when other Condorcet methods, including CSSD, fail so many
monotonicity criteria.

Sure, CSSD probably meets other criteria that Schwartz Woodalll
doesn't meet. For any 2 voting systems, that's pretty much a
certainty. But I've told why the Mlutual Majority Criterion, and
freedom from chicken dilemma are important, for the comlete freedom
from strategy-need that they confer on a mutual majority (MM).  The
IRV Condorcet hybrids, by adding Condorcet Criterion compliance,
thereby extend frreedom from strategy need to all majorities, however

I've fully told why I claim that those properties are important.  If
anyone else wants to claim that another property (possessed by CSSD
but not by Schwartz Woodall) is more important, then the burden of
proof is on them, to tell why their criteria are important.
Then,anyone having heard both advocacies, can judge for themselves
which criteria are more important.

> Where can we find public, third-party review and analysis of the method you
> propose (which seems to be a hybrid of other methods


 - so I'm not sure if it
> can properly be called "Schwartz Woodall" or not?)?

According to Green-Armytage, Woodall devised this method:

Do IRV till only member of the Smith set remains un-eliminated. Elect hir.

[end of Woodall definition]

For organizations, where there might be few enough voters for
pair-ties to happen, of course the Schwartz set is more exclusive than
the Smith set. Therefore, I proposed Schwartz Woodall. Schwartz
Woodall is my suggestion for a more deluxe version of Woodall, for
organizational voting. Schwartz Woodall is my name for the Schwartz
versioin of Woodall, in which the Schwartz set is substituted for the
Smith set.

But the name "Woodall" is used by Green-Armytage, for the above-defined method.

> Since the voting algorithm is enshrined in the Debian constitution, the cost
> of changing it is high; the burden of proof when arguing for a change is
> therefore high as well.

Quite so. And I fully accept and understand that Debian doesn't have a
chicken dilemma, and that changing the Debian voting system would
probably be a project that would unnecessarily take time away from
other work. I certainly am not claiming that Debian should change from
CSSD to Schwartz Woodall if Debian doesn't have a chicken dilemma.

It's just that, as one of those who initially advocated CSSD to
Debian, I felt a responsibility to tell about Schwartz Woodall.

Michael Ossipoff

Reply to: