At 09:29 PM 6/9/2010, Abd ul-Rahman Lomax wrote:
http://irvfactcheck.blogspot.com/p/there-is-lot-of-miss-information.html
<http://irvfactcheck.blogspot.com/p/there-is-lot-of-miss-information.html>Myths
and Facts about IRV
Continuing from Part 1, where the response to 1.2, first part of the
first paragraph, was begun. That paragraph is repeated here for context.
1.2 Is instant runoff voting prone to strategic voting?
No. There is another side to voting that makes IRV simpler than most
other voitng methods, such as Plurality, Approval, Range, etc. While
every voting method is potentially subject to manipulation by
strategic voting in some situations, IRV is uniquely resistant to
such strategy (see
<http://econ.ucsb.edu/graduate/PhDResearch/electionstrategy10b.pdf>James
Green-Armytage analysis for more) Under most voting methods, a
potentially beneficial voting strategy can be recognized by at least
some voters (who may gain an advantage over other voters). Thus,
voters may face a dilemma deciding whether to engage in strategic voting.
I provided a link to what is probably what they were referring to:
http://fc.antioch.edu/~james_green-armytage/vm/svn.pdf
The comment that "strategic voting" causes some voters to "gain an
advantage over other voters" is misleading. Strategic voting, indeed,
may be how the voters can bypass the poor results of a method,
resulting in an advantage to *most* voters. In the center squeeze
situation, a result with IRV that is less preferred to another
outcome by two thirds of the voters, can be fixed by some of the
voters, who support the candidate who can be predicted to lose, but
come in second place, voting for their second preference in first
rank. It needn't take many. In a perfect center-squeeze situation,
the candidate in the "center" is the first choice of (almost) a third
of the voters, and is the second choice of all the other voters,
sincerely. In Burlington, it wasn't that balanced, but the Democrat,
the candidate in the center in Burlington, was preferred as shown on
the actual ballot, to all other candidates, by a majority. It is this
kind of outcome that caused election experts to long ago dismiss IRV
as impossibly flawed, and it only survived in Australia and a few
places for local political reasons. It supports a two-party system,
making it safe from the challenges of upstart parties.
For example, in a simple plurality election there is the
"lesser-of-two-evils" problem where voters often realize that voting
for their favorite candidate, who is likely to get very few votes,
may deny their second choice candidate enough votes to beat out that
voter's least preferred choice. Such a voter might find the
plurality election decision to be extremely difficult.
Not usually. It's only difficult under certain fairly unusual
situations, and, except for those situations, the difficulty isn't
about the election itself, it's about the importance of expressing a
sincere preference. If you are a Green voter in in Florida in 1980,
you probably knew what you were doing, you knew that if you voted for
Nader, you were risking the election of Bush, and, my guess, you
*accepted that* because you believed Nader when he said it didn't
make a difference. This is a situation where, indeed, IRV makes
voting somewhat easier. But there is quite a cost, in the long run,
not only in tabulation expense, but also in overall satisfaction with
outcomes. The dirty little secret: most all voting systems, in most
elections, will come up with the same result. It really is the
exceptions that are what we need to be concerned with. Plurality will
usually choose the best winner, it is not totally stupid that this
method is in such wide use. Top Two runoff fixes, like IRV, the
spoiler effect, which only arises when there is a minor candidate who
bleeds off more votes from one side of the major party divide than
the other. Did that happen in 2000? I'm not sure at all. Maybe. Maybe
not. There were other minor candidates besides Nader, some on the
right, and there were other major problems with that election. And
Nader was campaigning on a Tweedle-dum and Tweedle-dee platform.
Would Nader voters in 2000 have added votes for Gore if the method
had been IRV? It's not completely clear. But, I'd agree, probably.
But so too would supporters of candidates on the other side have
added votes for Bush. We'll return to this election below.
Under IRV, this voter's ballot will automatically transfer to that
second choice candidate if the first choice is at the bottom, and
the second choice candidate can win with that transferred vote.
Sure. But the first choice might have won with votes from other
voters. There are two ways to approach this, both superior to the IRV
system, and both known from long before IRV. Condorcet methods will
look at all the pairwise elections, looking at how all voters have
ranked the candidates. The Democrat would have won in Burlington. And
Bucklin, which is a kind of "instant runoff Approval voting," and
which was actually also invented by Condorcet, all before 1800, if
I've got that right, looks for the candidate who is most "approved."
Voting choices in Bucklin are quite easy.
This makes the voter's task much simpler with IRV.
Sure. At the cost to the voter of the possibility of the first choice
actually winning! The way I state it is that, to protect this voter
from having to decide who else to approve, as in a Bucklin system (or
straight Approval voting, where the choice is indeed more difficult
-- but the voting method is terminally simple and cheap), the voting
system takes the first choice out back and eliminates him or her. And
none of this addresses the basic democratic principle of not making
any decision without the support of a majority of those voting. With
repeated ballot, the real Robert's Rules method, you vote for one,
who is your favorite from among all those whom you consider realistic
possibilities. Period. And if nobody gets a majority, the election is
repeated. Without any actual eliminations. These elections are
typically handled by voters using a blank piece of paper and writing
the name of the choice on it. Voters can write any name they please.
The voting system does not restrict or control them. That should be
understood as the background, it is a highly sophisticated voting
system that, among other things, is Condorcet compliant if a majority
is required to win. Neither center squeeze nor the spoiler effect
bother it. Because it takes as many "rounds" as needed, it is highly
intellligent, amounting to recursive processing as voters make
compromises, and it's easy to improve the efficiency of it. Approval
voting, for example, works perfectly with repeated ballot, and
Bucklin can collapse many rounds of repeated Approval voting into a
single ballot.
Bucklin used in a runoff system would be *extremely* easy to vote.
Just vote sincerely! If you just want to vote for your favorite,
that's fine, but you might have to face the inconvenience of a runoff
election. Your choice. If you would prefer the election of a second
choice to a runoff, then vote for another, at a lower rank. Some
voters will vote for many, and with Bucklin and other Approval
methods, you can actually vote "anybody but...."
The default strategy is always to honestly rank candidates in the
order of choice.
This is a tautology. It says nothing more than "the default strategy
is to honestly rank...." Definitely, that is not always the best
strategy, and can be quite poor, in one particular situation, and
that is where there are three major parties in partisan elections.
Burlington, Vermont. IRV was a Bad Idea for Burlington in the first
place, and voting systems theorists could certainly have told them,
and some tried.
Of course, voters may still face strategic dilemmas with IRV in
some rare situations, but these are far less common than under most
other voting methods.
The "rare situations" are indeed rare, overall. But the fact is that
the need for IRV at all is rare. What is the "need" for IRV? FairVote
hit upon this sales pitch: find majorities without expensive runoff elections.
Does it work? It turns out that, with nonpartisan elections, which is
the large majority of implementations and actual elections taking
place with IRV, if there is no majority in the first round (which
means that with Plurality, there would almost certainly have been a
majority also, all it takes with both systems is for the voter to
simply vote for their first preference, and the rest of the
preferences don't matter), there never is a majority of votes found.
FairVote has covered this up; originally by not mentioning it at all,
and the San Francisco voter information pamplhet contained a clear
error: it stated that the winner would still be "required to gain a
majority of the votes." But the actual language of the code revision
struck the requirement for a "majority of the votes." It should have
been obvious, but ... who reads the actual language? Many don't. In
North Carolina, where I lived for a time, I tried to get the actual
language of state propositions. Even at the library, they thought I
was crazy.... In California, they do send the actual language out,
but who reads and understands all the implications.
Again and again, in FairVote propaganda, the fact that an "IRV
majority" is usually not a real majority, when there are runoff
rounds, has been covered up. To be fair, they may not have realized
this fact. In Australia, in most places, voters are required to rank
all the candidates, or the ballot is "informal." So, naturally, every
election is won with what they call an "absolute majority." Even that
language would be misleading here, because in many places here (and
in Roberts's Rules) all cast ballots with marks are considered part
of the basis for "majority," even if spoiled in some way. But in
places where full ranking is optional, some places in Australia are
like that, majority failure when an election goes into "runoff
rounds" becomes common.
More recently, in response to repeated criticism, FairVote has
started to claim that exhausted ballots in IRV elections, ballots not
containing a vote for one of the top two, are "like" top two runoff.
It is an entirely new, made-up argument. Runoff elections are
separate elections, with a different set of voters. People decide
whether or not to vote in a real runoff election, knowing who the
candidates are, and having had additional opportunity to learn about
them and consider them. If they don't care, they don't vote, but if
they care, they do. When runoff elections represent a major
difference for many voters, they can and do turn out in even higher
numbers than the primary. FairVote claims that IRV is "fairer"
because runoff elections have lower turnout. That's highly deceptive.
When I pointed out that runoff elections sometimes have higher
turnout, or as high turnout, as primaries, they laughed and
derisively claimed I was crazy. But it depends on conditions. In
Cary, NC, for example, the primary was a special election, held in
October before the general election in November. If nobody got a
majority of the votes in the primary, then the runoff was held with
the general election. Turnout in primaries and runoffs, then, were
about equal, which I found somewhat surprising. It means that more
people than might be expected were turning out to vote in the
primary! Obviously, they cared enough to vote!
The load of blarney that FairVote has been selling places like that
is that they can have their majorities without "expwnsive and
inconvenient runoff elections." It is essentially a lie, by now,
since they no longer have the excuse of ignorance.
For example, under Range and Approval voting, giving any support
to a second choice may cause that voter's first choice to lose
(violation of the Later-No-Harm Criterion), causing some voters to
strategically truncate their true preferences. IRV complies with
the Later-No-Harm Criterion, and is thus immune to such strategic calculation.
And this is one of their biggest deceptions . Let's start with
Later-No-Harm. This is a criterion invented by Woodall, probably
desperate to find something supposedly good about the Alternate Vote.
When Woodall's paper was being reviewed, a reviewer wrote that this
criterion made him ill.
Why? Well, any voting system that satisfies Later-No-Harm is a system
that prohibits the voter from revealing possible compromises until
the voter's first choice is utterly eliminated. Yes. If you tell your
neighbors that you might accept Red as the color to paint that fence,
instead of your favorite Blue, you might end up with Red! So is the
solution to never reveal a possible compromise until you are sure
that there is no possibility of getting your favorite? This is what a
LNH compatible system *requires.* Surely this is a tad ... selfish?
Voting is a method by which people find broadly acceptable
compromises. LNH compliance turns it into a method where people vote
for and hold out for their first preference, not revealing their cards.
And the damage from that cuts both ways. Suppose that a group of
neighbors have all voted for their first preference for that fence
color. No preference has a majority. So what do they do? Well, one
way this would be handled directly is to count and report all the
first choice votes, then ask people to vote again! That, in fact, is
standard parliamentary procedure! People will start to make
compromises. If the voting method is approval (which I've seen
actually used by organizations for this, where each choice is voted
on separately, and people would vote for choices they found
acceptable), it's pretty easy, with repeated ballot, to gradually
lower your expectations, what's called the "approval cutoff." People
can do this slowly or quickly, or even at the beginning, it's up to them.
What Later-No-Harm compliance means is that you are protected from
"harming" your first choice by your first choice being eliminated by
the method, your vote is otherwise concealed when other methods look
at all the votes to determine the best compromise (according to the
method). So instead of *you" supposedly harming your first choice,
the method does, preventing your first choice from gathering second
choice votes from other voters until they, too, are eliminated, which
may be too late to keep your first choice from being eliminated.
Those votes from others aren't counted at all, in the formal IRV
method, they are completely negelcted, because the candidate has been
"eliminated."
It's not true that your addition of another approval, in Approval
voting, "harms" your favorite, though. What has actually happened is
that you have, by adding the additional approval, equally "helped"
both candidates. You only do this if you are willing to thereby elect
that second choice. Your vote did not "harm" your favorite, because
if you ballot is struck entirely, it would not cause your favorite to
win, that is, not with approval.
With IRV, your vote for a candidate can cause the candidate to lose.
How's that for "harm"?
Range Voting is a system where voters reveal their "relative
satisfaction" with the possible election of each candidate. Perhaps
they rate candidates on a scale of, say, 0-10. Quite obviously, if
you give any ratings other than 10 for your favorite and 0 for
everyone else, you might possibly help the candidate, you give a vote
to, to win. Range is like holding an Approval election where you can
cast fractional votes. Will you only bullet vote if you have the
choice? ("Bullet voting" means to vote only for one candidate,
regardless of the method. Pluralty requires bullet voting.)
FairVote asserts that voters will "strategically truncate" because of
this possibility of helping some other candidate besides your
favorite. Some will. Which ones? *The ones who have a strong
preference for their favorite." But wait, if the vote was an accurate
expression of preference, it was sincere, so it wasn't "strategic."
This kind of contradictory thinking is underneath much of FairVote's
facade. People will, in fact, not vote insincerely, but they will
also make decisions based on what they perceive as realistic.
To go down to the nuts and bolts of social choice theory, there is
generally no way to make people's "sincere preferences"
commensurable, i.e., to assign some absolute number to their
satisfaction, and then to add these numbers up to find an overall
social satisfaction that actually maximizes this. Is my "half
satisfaction" equal to your "half satisfaction"? But if we assign
everyone one vote, and allow people to express their "degree of
satisfaction," over some set of choices, it is fair to treat these
expressions as commensurable. After all, they are voluntary.
But what we see from critics of Range is the idea that people will
"exaggerate" their preferences in order to gain some advantage. From
one perspective they certainly will, but this doesn't damage the
outcome *seriously*. More accurately, this could only cause a poor
outcome if most people vote very stupidly, which is true for any voting system.
The classic comment is that people will supposedly bullet vote in
approval because of LNH failure of the method. This neglects the
actual situations which voters face. The most common situation is one
where there are only two frontrunners, plus minor candidates who
don't really have any hope. This, in fact, is the only situation
where IRV actually works with any reliability, fixing the "spoiler
effect" -- to the vast detriment of minor parties, which can almost
never win with IRV. They don't win in Australia, and they have minor parties!
In that situation, if you favor a frontrunner, you vote just for the
frontrunner, and nobody else, unless maybe you like a minor canidate
*almost* as much and don't care if your vote helps the other win, on
the off chance you were wrong about "frontrunners." That, by
definition, is at least two-thirds of the voters or so, or quite a
bit more if there is only one minor candidate. Then there are the
supporters of the minor candidates. These are the ones who may add an
approval for a frontrunner. These voters are not particularly worried
that their additional vote will "harm" their favorite. So we can
expect, with Approval, that most voters, under two-party conditions,
will "bullet vote." It is harmless. It isn't "strategic voting," it
is honest preference.
Now, there is another method that is in the Approval or Range class,
and that's Bucklin. With Bucklin, you submit a ranked ballot. The
votes are counted in rounds like IRV, except that each ranks' votes
are added, in turn, to the totals for the candidates. You can vote
for just one if you like. Or you can add approvals at lower ranks,
and you can skip ranks (which is why it's really like Range voting,
especially if one can vote for as many candidates as one chooses at
any rank, something which -- just allowing multiple candidate votes
in third rank origially -- allowed Bucklin to find true majorities
with *many* candidates and only three ranks on the ballot.)
The ranks are counted, in sequence, until a majority is found. (This
means a majority of ballots contain a vote for the winner; there may
be many more votes cast than ballots, but the majority sought is a
"majority of voters," not, to use the language of Brown v. Smallwood
in Minnesota, a "majority of marks." It's odd that the MN Supreme
Court, tossing out the very popular Bucklin method, correctly got the
standard, but then proceeded to complain that there were more marks
than voters, as if they had not just written what they wrote. That
court had an agenda.
The advantage of Bucklin over Approval is that the voter does not
have to equate two candidates where the voter actually has a
significant preference. The Nader voter can vote Nader>Gore or even
Nader>.>Gore (skipping the second rank) if they really want to
emphasize the gap, without harm. Nader still has two rounds, then, in
the last case, in which to win before the voter's ballot causes equal
consideration with Gore.
Absolutely, some voters will refrain from adding additional
approvals. These are the ones who strongly prefer their favorite!
Those are, in fact, sincere Range votes, just counted in a different way.
What Bucklin does is to *defer* counting your lower preference votes
until it's clear that your favorite is not going to win by a majority
(before your lower ranked votes are counted). With Bucklin, the
operative phrase is "not going to win by a majority," whereas with
IRV it's "has been eliminated." Tell me, which would you prefer as a
candidate? To risk the possibility of losing to another because of
additional approvals, but at the same time to have the opportunity to
gain those approvals yourself, and not be eliminated, or to be
eliminated before additional votes for you can be considered?
IRV favors candidates with the most first preference votes. If we
have a situation with three balanced candidates, as to first
preference, it's a toss-up which is eliminated. If the eliminated
candidate was *everyone's first or second choice," preferred by a
two-thirds majority to each of the other two candidates, too bad.
This is the consequence of elimination and of complying with Later No
Harm. No wonder that reviewer felt ill.
----
Election-Methods mailing list - see http://electorama.com/em for list info
