At 07:53 PM 12/31/2008, Jan Kok wrote:
Click on the blue box to the left of the title "Implement Score Voting".
For those that may not have heard: Score Voting is a new, alternative
name for Range Voting.
Yeah. Small improvement. Big cost. You get a book out, Gaming the
Vote, by Poundstone, and lots of references now to Range Voting, so
what do you do?
You know what I think? I think that, like the name Approval, it's
misleading. These are "votes," they are *not* scores, just like
Approval votes are votes, not sentiments.
So, it's a setup for the strategic voting argument. "The voter
*really* scores the candidate at 90%, but in order to avoid
competition, votes strategically at 0%."
That's Bad, isn't it Bad to Exaggerate?
No, they are just votes, voters decide, from a mixture of utility or
expected satisfaction and probability, where to place their single
vote egg. All in one basket? Or some in this and some in that?
I imagine a series of candidates in a row, arranged in preference
order. True clones are right behind each other, occupying the same
position. In between each candidate and the next is a basket. I have
one vote, and I can distribute it between the baskets. The sum of
what I put in the baskets is one full vote.
Range Voting is *fractional* voting. Jan, you may not have noticed
that I recently realized that Oklahoma Bucklin was Range Voting,
phased in. (only certain values were allowed: 1 full vote -- first
preference --, 1/2 vote -- second rank -- and 1/3 vote -- third rank
--; the first and second ranks were exclusive.)
Or another way to conceptualize it is that the voter has, say, 100
votes. Each vote is a ball of a certain weight, all the same. The
voter drops the balls in the baskets, putting them where the voter
thinks they will do the most good. The balls represent votes in the
pairwise elections between all candidates above that position and
those below that position. (It's harder to describe than to do!)
So, for example, if you want to *totally* support A and B, and to
*totally reject* C and D, you will place all the balls in the basket
between B and C (i.e., the candidates are ordered A>B>C>D.) So this
is an approval style vote for A and B and against C and D. There are
100 balls in that basket and the vote for any candidate is the number
of balls in all the baskets below that candidate.
From the votes sincere preferences may be inferred. Not *all*
preferences may be inferred, but every imputed preference would be,
presumably, accurate, there is no benefit to not placing the
vote-balls that way. Preference strength is not determinable from the
ball placements because that is not the only consideration.
Essentially, the votes are votes in lotteries, and the voter wants to
"bet" the limited number of balls in such a way as to maximize the outcome.
As *votes*, the concept of sincerity, though, is off. Votes are
*decisions,* and decisions are not sentiments, they are actions.
I've been starting to promote the name Open Voting for Approval. It's
more accurate. It's simply Plurality with the vote-for-one-only
removed, so it is "Open." Range is then Open Fractional Voting.
Is Score better? Maybe. But it would have been better a year or two
ago, if that decision was going to be made. Now there is a cost to
it, and I am *not* convinced it is worth the cost.
Abstentions, i.e, *partial* abstentions, where the voter scores one
or more candidates but doesn't score some, *must* be counted as zero
rating (minimum), or a series of legal questions will have to be faced.
With Approval Voting, there is no proposal to use *average vote*,
with Yes/No scores for each candidate, and a voter who doesn't vote
on a candidate isn't considered. The precedent is that a partial
abstention can be neglected in determining if a ballot question
passes, but the result, the choice between two accepted candidates,
is the one with the most Yes votes.
I think it's a tactical error to go immediately for Score Voting. The
principle to be established is Open Voting, and the way to get there,
I have concluded, is to *recover* the old reform known as American
Preferential Voting, which was very popular, *and use it with runoffs*.
Preferential Voting was used in quite a few different places, both
Bucklin's method ("American Preferential Voting") and the Ware method
("English Preferential Voting," as it was known), and it was
abandoned, eventually. Why? *I don't know.* FairVote claims it was
because too few voters were adding additional ranks.
They blame that on Bucklin's LNH failure, but the fact is that
Bucklin does provide some substantial protection for the favorite
(better than pure Approval, for sure! which is to say, better than
Range as well), and further, what about the implementations that were
actually IRV or a form of it? (Some were two-rank only, for example.)
FairVote claims that STV was abandoned because of racism and the red
scare, but that probably only applies to multiwinner STV, at most,
not to the single-winner preferential voting applications.
Anyway, we've had a form of Open Voting (Approval) here, contrary to
what is often said. Which means we've had a form of Range, already.
It was sold as a way to find majorities. Did you know that American
elections, I've been finding, originally required majorities? I just
saw the old New Hampshire constitution: if a majority wasn't found,
the top two candidates (or something like that) were submitted to the
legislature, which chose. It wasn't until much later, in the
nineteenth century, that election by plurality was allowed. Same in
Massachusetts.
It is *still* the same in Vermont. Election of the governor requires
a majority; if there is no majority, the top three candidates are
presented to the legislature, which votes by secret ballot to
determine the winner. I forget if a majority is required in that election.
Pure Range doesn't define majorities, so it *isn't* a democratic
voting method! However, it is easy to make it one. (It's easy to
assume that midrating or higher is a vote for a candidate.)
So preferential voting was sold, as IRV is being sold now, as a way
to obtain majorities. Then as now, this was considered highly
desirable. FairVote has essentially lied its way into a series of
implementations, the language has been *explicitly* incorrect, such
as a claim in Santa Clara County, the first "success" in 1998, that
the winner will still be required to get a vote from more than half
the ballots, or a majority of total votes, etc.
Of course, there, it was merely enabling legislation, and didn't
actually describe the method. So it's possible that the
implementation actually could continue to require a majority or a
runoff. I would argue that the ballot measure (passed by 53%)
*requires* that the method continue to require a majority. If they
try to implement standard RCV, they will be violating a condition of
the measure. Unfortunately, the measure *also* claims that runoffs
will be eliminated.
*However,* Bucklin would do what the voters wanted, better than IRV.
It is a bit better at finding majorities from the votes, probably,
because it counts all the votes. My sense is that voters will rank as
sincerely with Bucklin as they do with IRV: in both cases, under
stable conditions, most voters won't bother with additional
preferences. Bucklin maintains what voters say they want most, as an
objection to Approval: to be able to vote for their favorite, as a
favorite, without thereby wasting their vote.
Bucklin easily becomes Range by using fractional vote values for the
ranks, as was done in Oklahoma.
Range, where the votes are considered ranks, and the ranks are
canvassed in descending rounds, adding new votes as they are reached,
until a candidate has a majority, satisfies the Majority Criterion.
Basically, nearly every argument that has been raised against
Approval and Range is satisfied or finessed with Bucklin, American
Preferential Voting. Invented in America, according to most sources.
However, one source I found, a document prepared for the
Massachussets Constitutional Convention, some time around 1918,
claims that Bucklin was first proposed by Condorcet! And that it was
used in Geneva briefly. Was that true or was it a confusion?
Condorcet did apparently propose a method and it apparently was used
in Geneva with "chaotic results," according to one source. I've come
to distrust most of these judgments.
I was saying for some time that we should be focusing on Approval as
the simplest reform; it establishes the Open Voting principle of
voting independently for each candidate. However, Bucklin, as
"instant runoff approval," addresses the political needs of American
voters more effectively. It was *very popular*, contrary to what some
have claimed. It wasn't voters who didn't like it; in Minnesota, a
loser sued. That loser clearly was not the best choice, by a decent
margin. The loser won. (Actually, I think it may have been a voter
supporting the loser who sued.) From the decision in Brown v.
Smallwood, we can tell that the decision was very unpopular, Bucklin
voting was *liked*.
So what happened? I'd say we need to know. There were 55 towns using
preferential voting by 1918 or so, including major cities like San
Francisco. There is lots of notice of the implementations, Bucklin
had "momentum." I've tried searching, so far nothing on the demise!
except for Minnesota, found unconstitutional on a basis that would
find all preferential voting unconstitutional, and Oklahoma, where
the method itself wasn't found unconstitutional, but a provision
tacked onto it that required voters to add additional preferences,
under some conditions, or their vote would not be counted. (As with
Australia, though full ranking wasn't required in Oklahoma).
Take note: tack the "average vote" concept onto Range, the same thing
could happen to it. The voting systems want the greatest number of
*voters* to support the winner, or at least a majority. Average vote
dumps that, doesn't even require a plurality. And plurality is
written into most state constitutions, I think. Standard sum-of-votes
Range probably doesn't have this problem. Average Range is a sitting
duck. Besides all the arguments that will come up in an
implementation debate. It's *weak*. Dump it!
So, Bucklin may have been dumped because it didn't fulfill the
promise of finding majorities. Ultimately it was replaced with a
system which did that better: top two runoff. *However*, with our
clear hindsight, we can see that a mistake was made. They should have
kept the Bucklin, and used the method as a primary in a runoff
system. And also as a runoff method itself, with write-ins allowed.
This is a Condorcet compliant method! But runoffs test preference
strength, so it is *de facto* Range as well. And it can be made
explicitly Range by using fractional votes, as with Oklahoma only
probably with a better vote distribution.
----
Election-Methods mailing list - see http://electorama.com/em for list info
