15. Dopp: Violates some election fairness principles
."
This charge reveals either a general lack of
understanding, or intentional
miss-representation. Every single voting method
ever devised must violate some "fairness
principles" as some of these criteria are
mutually exclusive. Dopp's example in appendix B
of "Arrow's fairness condition" (the Pareto
Improvement Criterion) completely misunderstands
the criterion, and gives an example that has no
relevance to it (and contrary to her
implication, IRV complies with this criterion).
IRV works essentially the same as a traditional
runoff election to find a majority winner. When
the field narrows to the two finalists in the
final instant runoff count, the candidate with
more support (ranked more favorably on more
ballots) will always win. Some theoretical
voting methods may satisfy some "fairness'
criteria, such as monotonicity, but then violate
other more important criteria such as the
majority criterion, or the later-no-harm criterion.
This is typical argument from FairVote. Read it
carefully. Without going into the truth of the
remainder of the paragraph, the remainder of the
paragraph confirms what Ms. Dopp wrote. Sure,
it's a possible argument that "all voting methods
violate some election fairness principles," but
... Ms. Dopps statement still stands. There are a
number of issues here, and it's something that
has fooled even experts, so please bear with me.
Arrow's theorem has been widely interpreted as
"no election method is perfect," or "all election
methods must violate at least one of a list of intuitively fair principles."
However, Arrow's theorem was actually very
limited, and Arrow made the decsion not to
consider as "voting methods" such basic methods
as Approval Voting and Range Voting. Arrow
recently actually repeated this, that he does not
consider Range a "voting method." This is because
it expresses preference strength information, and
Arrow concluded that there was no substance to
this. It's a very complicated debate, in fact. To
Arrow, all that matters is whether you prefer one
candidate to another, and how strongly you prefer
is irrelevant. Yet in ordinary human
decision-making, if we lived by the black and
white rules of pure preference, we'd be making
some pretty bad decisions! There is more recent
work on this that has not yet been widely
accepted, in particular a paper by Dhillon and
Mertensm, published in Econometria, Vol. 67, No.
3 (May, 1999), pp 471-498, purports to prove that
what they call Relative Utilitarianism, but which
is identical to the Range Voting proposed by
Warren Smith in his later work, is the unique
solution satisfying redefined Arrovian criteria
that allow for equal ranking and preference
strength information to be expressed.
Range Voting and Approval Voting both satisfy, in
fact, reasonable interpretations of all of
Arrow's fairness criteria; however, Arrow doesn't
consider either of them "voting systems," and
they do not meet the definition of voting system
used in his proof. (Actually, the proof doesn't
describe voting systems, as such, but merely the
conversion of a set of individual preferences
into an overall social ordering of options.
"preferences" means a strict ordering of all the
options. No equal preferences allowed, and no
consideration of preference strength.
This matter of preference strength is crucial.
For example, several possible voting methods,
excellent in many respects, fail the Majority
Criterion. Sometimes FairVote tries to equate
this with failing "Majority Rule," but, in fact, the two are quite different.
There are some differences of opinion as to how
to interpret the Majority Criterion. When Woodall
did his work on it, he was considering only
preferential voting systems, with no equal
preference allowed. And the input to the voting
systems he was describing was a strict preference order.
The Majority Criterion can be stated as: if a
majority of voters place a candidate at the top
of their preference listings, that candidate must
win. This is an example of a criterion that
seems, at first glance, to most, to be
intuitively fair and proper and, even, necessary,
and it sounds like majority rule, and, indeed, there is a connection.
This is the connection. If a majority of voters
express a vote to prefer A over all other
candidates, A must win. This is a variation on
majority rule. The essence of majority rule is
that no decision is made except by the explicit
consent of a majority of those voting. If a
voting system can choose a winner without the
consent of a majority of those voting, it
violates what I call the Majority Rule Criterion.
Does IRV satisfy the Majority Criterion? Sure. If
a majority of voters vote for a candidate in
first preference, that candidate immediately wins
the first round. And only one vote is allowed in
the first round, the election rules always
prohibit approval-style voting, so MC compliance is assured.
But does IRV satisfy the Majority Rule Criterion?
No. In fact, there is only one election method in
common use for public elections that does satisfy
it: top two runoff. (It *fully* satisfies it if
the rules allow write-in votes, if "vote" is
interpreted as in Robert's Rules of Order, and if
it is possible for the runoff to also fail if no
candidate gains a majority in the runoff. In most
runoffs this is possible, if extraordinarily
rare, but I don't know about the actual rules if
there is majority failure in the runoff. Robert's
Rules would say keep at it. Repeat the balloting
until you have a majority winner.)
Top-two runoff is *also* not a voting method by
Arrow's definition, because it isn't
deterministic from a single static set of
preference profiles. Voters can actually change
their votes! Robert's Rules, in fact, points out
that preferential voting "deprives" voters of the
ability to base later votes on the results of
earlier results. IRV is, in fact, a plurality
method, *unless you continue to require a true
majority." In Australia, in most Preferential
Voting elections, they do require an absolute
majority. They manage this by requiring *full*
ranking of all candidates, or the whole vote is
considered spoiled and not part of basis for
majority. Imagine that in District 9 in San
Francisco, with 22 candidates! This, however,
represents coerced votes, it is difficult to call
those lowest preference votes "consent."
Later-No-Harm is FairVote's favorite election
criterion. That's because the peculiar design of
sequential elimination guarantees -- if a
majority is not required -- that a lower
preference cannot harm a higher preference,
because the lower preferences are only considered
if a higher one is eliminated. But later-no-harm
is a quite controversial criterion, many think it positively undesirable.
Woodall, who named Later-no-harm, wrote: "...
Under STV the later preferences on a ballot are
not even considered until the fates of all
candidates of earlier preference have been
decided. Thus a voter can be certain that adding
extra preferences to his or her preference
listing can neither help nor harm any candidate
already listed. Supporters of STV usually regard
this as a very important property, although it
has to be said that not everyone agrees; the
property has been described (by Michael Dummett,
in a letter to Robert Newland) as "quite
unreasonable", and (by an anonymous referee) as "unpalatable".
Indeed. Later-no-harm interferes with the process
of equitable compromise that is essential to the
social cooperation that voting is supposed to
facilitate. If I am negotiating with my neighbor,
and his preferred option differs from mine, if I
reveal that some compromise option is acceptable
to me, before I'm certain that my favorite won't
be chosen, it is utterly ruled out, then I may
"harm" the chance of my favorite being chosen. If
the method my neighbor and I used to help us make
the decision *requires* later-no-harm, it will
interfere with the negotiation process, make it
more difficult to find mutually acceptable solutions.
Later-no-harm is actually one of the few common
criteria that IRV satisfies, along with the Majority Criterion.
So what about that Majority Criterion? Is it
desirable? Well, any method which considers
preference strength will fail the Majority
Criterion, it must. Suppose that 51% of the
voters have a trivial preference for A over B,
they really don't care, but if you ask them, they
would say they prefer A. The other 49% strongly
prefer B. Maybe this is a choice of foods, and
they are allergic to B. What's fair? Majority
Criterion, or maximized overall satisfaction with
the result. As this example was stated, the
choice of B has no significant harmful effect on
the majority and, in any healthy society, if they
are informed, say by a Range poll, of that strong
preference of the minority, and especially if it
is explained to them, they will quite cheerfully
vote, if directly asked, "Shall we choose B?", yes. Quite probably unanimously.
Majority Rule, strictly, involves asking a single
question that can be answered yes or no. However,
for efficiency, we do allow multiple questions;
but then the question arises, what if multple
conflicting choices both receive a Yes vote?
There is a standard legal answer for this: the
one that has the most Yes votes will prevail.
Approval Voting. But, technically, it fails the Majority Criterion.
*When* does it fail the MC? Only if there is more
than one option approved by a majority. In that
case, it is possible that the majority actually
preferred the option that received less Yes votes
than the other, because whenever a person votes
for more than one option, the preference between
those options is concealed (in Approval voting,
or those multiple conflicting Ballot Questions).
FairVote will be very quick to tell us that
Approval fails the Majority Criterion, but this is what that actually means:
In Florida 2000, suppose that voters could have
voted for more than one. Voters who preferred
Nader might also have added a vote for Gore. This
is really an alternative vote, because it is
never operative as more than one vote in any
pairwise election. Naturally, this violates
Later-no-harm, because, in theory, the extra vote
for Gore could cause Gore to beat Nader. I'm sure
that the Nader voters would have been very
worried about that, don't you think? But what
about the Majority Criterion? Well, the only
reasonable possibility at all would be that many
voters approved both Gore and Bush. More than
didn't approve either Gore or Bush. We can be
sure that there would be a lot of the latter, who
aren't going to vote for a major party candidate
come hell or high water. But the reverse, voters
who vote for both frontrunners in a partisan
election? Let;s say this would be rare and leave
it at that. And not only rare, but harmless.
Approval Voting, as I mentioned, isn't a voting
system by the definitions of Arrow's theorem, but
if the definitions are generalized in a
reasonable way, Approval does meet all the
conditions of Arrow's theorem. It should, it's a
Range method, the very simplest, and Range methods meet those conditions.
IRV can drastically fail to elect a candidate
preferred to the eventual IRV winner by a large
majority. It meets the Majority Criterion, yes,
but if fails Majority Rule, and badly, and, in
this case, not only Majority Rule, but the very
basic rule, the king of preferential voting
rules, the Condorcet Criterion. If there is a
candidate who is preferred to all other
candidates by a plurality of voters, considering
each pair of candidate separately, this candidate
must win. Approval fails the Condorcet Criterion
for the same reason it fails the Majority
Criterion. Essentially, it fails it for a good
reason, it does something better.
What does "better" mean? It's only fairly
recently that seriosu work started to become
widely known on this question. There is a method
of estimating the overall satisfaction of an
electorate with an election, and, in fact, it
shows Range Voting to be optimal precisely
because Range Voting, if we could somehow insure
totally accurate sincere votes, *is* the method
of measuring satisfaction. There is some very
serious math behind this, and a lot of work,
mostly by mathematicians and economists. The
political scientists mostly got stuck with
preference, but economists worked with game
theory and how to make optimal choices.
What is the point of all this? If we are going to
reform elections, there are much better methods
to choose than instant runoff voting, and, it
turns out, they are simpler and cheaper to
implement. Approval costs practically nothing:
just count all the votes. Bucklin voting deserves
another look: it was Widely used in the U.S., and
it was popular, and real voters, apparently,
don't pay much attention to the Later-No-Harm
criterion. The "harm" in Bucklin only occurs if
your favorite doesn't win by a majority in the
first round. The only difference, really, is that
Bucklin doesn't eliminate any candidates. It just
counts all the votes. It's quite like Approval,
but ranked. "Instant runoff approval." It is more
efficient at finding majorities than IRV, because
IRV does *not* count all the votes. (When an
election reaches the last round without having
found a majority of all the votes cast, there are
the lower preferences of the remaining two
candidates that have not been uncovered yet.
These votes are never counted. -- but the San
Francisco reports do show them, so it is possible
to do Bucklin analysis on those San Francisco IRV elections.)
Contrary to the attack on her integrity from
FairVote, Kathy Dopp does explore these issues in
her report. The one point, though, that is often
overlooked, is that IRV is being used to replace
top-two runoff, a far superior method from the
point of view of democratic process, and one that
can easily, with little or no expense, be made
even better, i.e., more efficient at finding
majorities without requiring a runoff, as well as
more likely to choose the best candidates if
there must be a runoff. Hybrid methods are quite
possible that would perform essentially ideally,
satisfying the requirements of democratic process
-- which Plurality and its sister IRV don't --
and all that it takes is the political will to
start examining the alternatives with clear eyes
and open minds. We can start by looking at how
existing methods are performing. Not much study
has been done on top-two runoff. And not much
study has been done of how IRV is actually
performing. That needs to be remedied. I can
guarantee our readers this: FairVote is not
interested in objective analysis of how IRV is
performing, it's on a mission, and it does not want to be distracted by facts.
----
Election-Methods mailing list - see http://electorama.com/em for list info
