[EM] Approval Voting vs Instant Runoff Voting

[Richard Moore]

-- In election-methods-list at y..., donald at m... (Simmons' Approval Voting) wrote:

> Don: It all depends on what a person's agenda is. If the agenda is to give
> extra undue help to a lower candidate so that that candidate can slip thur
> the back door to win the election, then you are correct to feel that
> Approval Voting has some significant advantage in this respect over Irving
> (IRV).

If "lower candidate" means "the candidate approved by more voters than
any other candidate" and "back door" means "a fair and monotonic
election method" then yes, Approval Voting does this. I don't know how
to translate "extra undue help" into anything that makes better sense of

the above paragraph.

> Approval Voting does subsidize the lower candidates while Irving
> insists on treating all the candidates and all the parties and all the

> voters with equality. The word `all' includes everyone, not just the
lowest
> candidates and their supporters.

I'm not sure what Don means by "subsidize". Voters cast votes for
candidates they approve of. They do not cast votes for those they do not

approve of. Why call this expression of voter preference a "subsidy"?
And
then, which are the "lower/lowest candidates"? The ones that come in
last
according to some other method? Using one method's results to judge
another
method has the effect of elevating the first method to a standard.

> If we are ever going to have a better election method, it will need to

> be a method that is better for everyone. Irving will be of help to all

> factions.

You can't help all factions simultaneously. A single-winner election is
a constant-sum game -- for the candidates (though not for the voters,
which is one thing that makes the topic so interesting).

> I agree that candidate C should be the winner, but I do not agree that

> this is the way most people would have voted. I have this theory that
goes:
> If and when Approval Voting or Condorcet are ever imposed on the
people,
> these methods will change automatically into Irving, as a result of
the way
> the people will vote and/or not vote. This is the Trump Card
thatIrving
> holds over Approval Voting and Condorcet. For that reason, I have no
fear
> of these two methods, if they are installed, so what, the people will
> change the election into an Irving election.

How and why would they do that, Don? You say this will happen
automatically. I'm sorry, I just don't understand.

> Under Approval Voting, most people in your example will vote for only
> C, no other choice, because most people will learn that, in the
Approval
> method, other choices they make will be used to help defeat their
number
> one most preferred choice. If the people don't see that right away,
someone
> will make sure they are informed.

Well, someone -- either an IRV supporter or some political group who
feels threatened -- will probably try to see that the voters are thus
MISinformed. But the statement seems to imply that voters only want to
see their most preferred choice elected at any cost. In this view, they
would take the risk of seeing their most feared candidate elected, just
to ensure that their second choice doesn't sneak up and defeat their
favorite. But would the voters really vote this way in an approval
election if the candidates were, say, Roosevelt, Truman, Stalin, and
Hitler?

> This is a common mistake made by supporters of Approval Voting. They
> think most voters will be making more than one choice. Not so, the
> supporters of the major factions have no interest in deciding which of
the
> lower candidates should be last. On the other hand, the supporters of
the
> lower candidates should and will make lower choices because they will
have
> an interest in deciding which candidate should be first.

Sounds like a scenario in which there are monolithic voting blocs rather

than individual voters. Why characterize a voter who votes for one of
the major candidates and one of the minor candidates as a supporter of
the minor candidate and not a supporter of the major candidate, when it
is clear the voter has given equal support to both?

> In an Approval Voting election if there happens to be a candidate with

> a majority of the first choices, it is possible for this candidate to
lose
> the election. This is because the winner in an Approval Voting
election is
> determined by all the choices, not just the first choices. With only
two
> candidates, the student body was facing the very real possibility of
having
> one candidate seeming to win via the first choices and the other
candidate
> winning via Approval Voting, the addition of first and second choices.
Best
> to avoid this public relation conflict, best to avoid Approval Voting.

Such an outcome might happen if a sufficient number of voters from one
side vote AB while the other side's voters vote only their candidate.
Suppose the actual preferences are: A 55, B 45, but 15 of the A voters
vote AB instead, so the election gives B the victory:

    A 40
    AB 15
    B 45

But all that has really happened is that 15 voters voided their ballots.

They could just as easily have voided their ballots in the plurality
election. There are ways to cast void ballots in IRV. So what is the
point of this story?

> Consider a four candidate election:
>           40 A,  30 B,  20 C,  10 D
> Now, it was not the method that caused candidate A to receive forty
percent
> of the votes and for candidate D to receive only ten percent of the
votes.
> No, it was the voters that did that, because that is their job, they
were
> merely sticking to business. It is the business of the voters to favor
one
> candidate over the others, it is not the business of any election
method to
> favor any candidate over any other, but that is just what happens when
the
> lower choices are added in with the first choices. Take a look at the
lower
> choices.
>           60 A,  70 B,  80 C,  90 D,
>     The lower choices favor the lowest candidates over the highest
> candidates. This is not the intent of the voters.
>     When we add the lower choices to the first choices we get:
>         100 A,  100 B,  100 C,  100 D,
>     We must be careful how many of the lower choices we use. Lower
choices
> negate the intent of the first choices. The more of them that are
added
> into the calculations the more the voter's intent is negated. Approval

> Voting is the worst method in this respect because it uses all the
lower
> choices.

Huh? Approval voting uses only those choices -- lower or otherwise (the
distinction is silly) that the voters make. The argument completely
misses
the point of approval voting. A voter in an approval system exercises
strategy
by setting a threshold, based on how the candidates are distributed in
his own
view of the political spectrum as well as the apparent popularity of the

candidates. He can still vote sincerely in exercising that strategy. If
the voter's ranked preferences are A, B, C, D, then he can vote A, AB,
or
ABC. He will never be compelled to vote AC or AD or ABD for strategic
reasons.

> The first choices are the voter's favored choices. The lower choices
> are the reverse, they are the voter's disfavored choices. The lower
choices
> are the leftover choices and they favor the lowest candidates. This is
not
> rocket science, anyone should be able to see that when we added in all
the
> lower choices, the lowest candidates were assisted more than the
higher
> candidates. Adding all the choices together gives all candidates the
same
> chance of winning, regardless of how the people voted. We may as well
use
> sortition.

But you DON'T add the disfavored choices. The voter doesn't mark those.
If he does, he obviously doesn't have very strong preferences. He is
like the AB voters in the student body election example.

> You have got to accept that a candidate must get enough votes in order

> to win. Stop advocating some magic method that will leverage your
> candidate's few votes into a win.

Don must not be talking about approval anymore here. Approval doesn't
"leverage [a] few votes into a win". Without the approval of the largest

number of voters, a candidate does not win. Why is letting a candidate
win based on some artificially constructed "majority" better than this?

> >      In the case of an election of four or more candidates and three
that
> > are close together, these two method may not always elect the
correct
> > candidate.  We will need something better than Top Two Runoff or
> > Supplementary Vote, we will need Irving.

"Top Two Runoff" (or Real Runoff) is actually better than IRV. It still
meets the Condorcet loser criterion (heck, any lame method that selects
two candidates arbitrarily and holds a runoff between those two does
this). But, once the two finalists are known, voters can abandon their
strategies and vote sincerely. IRV does not give them this opportunity.
So Real Runoff actually meets the sincere Condorcet loser criterion. Of
course, a Real Runoff is more costly and more time consuming than a
single election.

> > The candidate with the lowest
> > number of first choices will be dropped. If this candidate happens
to be
> > your first choice, fear not, your vote will not be dropped. Your
vote will
> > be salvaged and transferred, as an instant runoff, to your next
choice, and
> > transferred again if necessary until your vote ends up on one of the
last
> > two candidates so that you can have a voice in the final runoff of
the
> > election.

"Fear not"? This scheme for "salvaging" votes is the source of IRV's
aliasing problems. No thanks.

 -- Richard Moore