[EM] (with paragraph spacings) The two extremes of voting-system strategy-advantages

[Michael Ossipoff]

(I'm re-sendiing this in plain text, in the hopes that it will post
with the paragraph spacings)

Of course there are all sorts of standards and criteria by which
voting systems can be compared and regarded. But I'm not the only
person whose interest in voting systems is about strategy.

I suggest that there are two classes of methods that represent two
extremes of voting system strategy advantages. ...and that that
voting-systems distinction and division is significant for classifying
voting systems.

I refer to these two classes of voting systems:

1) Approval and a few strategically-similar methods (Score, ICT, and
Symmetrical ICT, and maybe others).
.
2) IRV, its Condorcet hybrids, and Approval-IRV (AIRV)

Those two classes of methods represent two extremes of voting system
strategy advantages.

The 1st class, I've named "Approval etc.", or Approval&c, further
abbreviated to A&c.

The 2nd class, I'll call "IRV etc.", or IRV&c, further abbreviated to I&c.

(I'm using the name "IRV" instead of AV or Preferential Voting,
because IRV is now best known by that name, and is offered
by that name in the platforms of at least 5 U.S. political parties.)

Approval&c:
---------------

Other than its compliance with all of the monotonicity and consistency
criteria, Approval's main criterion advantages are its compliance with
the Favorite-Betrayal Criterion (FBC) and Later-No-Help (LNHe).

Symmetrical ICT meets FBC, and an insignificantly weaker version of
LNHe, which I call 0-info LNHe. I've defined it here, and at
electowiki.

But maybe I should repeat it here:

0-info LNHe (ZLNHe):

In a 0-info election, voting for one or more candidates in set S
shouldn't decrease the probability that the winner will be from S.

To vote for a candidate is to vote that candidate over at least one
other candidate.

(In earlier postings, I've given two general, universally-applicable,
definitions of voting X over Y)

[end of ZLNHe definition]

Like LNHe, ZLNHe relieves voters of the need to vote for unacceptable
candidates, greatly easing the task of voting when there are many
candidates.

Symmetrical ICT additionally brings compliance with CD. Symmetrical
ICT automatically avoids the chicken dilemma. I suggest that
Symmetrical ICT (SITC) could be regarded as Approval with CD.

Approval, Score, ICT and Symmetrical ICT, I refer collectively to as
Approval etc., Approval&c, or A&c.

Symmetrical ICT meets the Condorcet Critrerion when CC is defined in a
general way (arguably the more legitimate way).

But all A&c tend to elect CWs.

Though Approval and Score don't meet CD, they have a good way
(strategic fractional ratings) of dealing with chicken dilemma. For
that and other reasons (described in previous postings here) the
chicken dilemma won't be a problem with Approval and Score (or, of
course with CD-complyng ICT and Symmetrical ICT).

Approval and Score, and, to a large extent, all A&c, have great
stability. I've often referred to Approval as a solid and reliable
handtool, as opposed to idiosyncratic labor-saving
contraptions.

A&c are the methods of choice where FBC is needed, as it is under
current conditions.

Approval's opponents want to fully vote X over Y, and Y over Z. They'd
like to be allowed that, but the catch is that most methods that allow
that give strategic incentive or need to do so insincerely, which
really defeats the whole purpose of it.

Approval lets you reliably and fully vote a preferred set over an
unpreferred set. The importance and power of that shouldn't be
underestimated. You can easily, simply, fully vote all the acceptable
candidates over all the unacceptable candidates, and that will be
reliably and simply counted. And that's what matters in a u/a
election. With various kinds of experimental test-voting, and with
probabilistic strategic fractional rating, it's possible to begin
choosing among one's preferred set too.

Strategy in Approval:

If it's a u/a election, then it's especially simple; Just approve the
candidates who are acceptable.

Say it's non-u/a:

Some people express a need to vote in a way that is optimal, in terms
of the actual (unknowable) configuration of other people's votes. But
is that necessary? Maybe it's enough to vote in a way that's optimal
in terms of your perception of the situation. You can do that by
approving the candidates you like, the ones you trust. If it's a u/a
election, then it's especially simple; Just approve the candidates who
are acceptable. But here I'm talking about non-u/a.

If you want to make it unnecessarily more difficult, you could, before
approving a candidate, ask yourself such questions as "Does s/he feel
more like a protection, or a menace, if I approve hir?", or " Are the
better candidates really so good that you don't want that compromise?
Are the worse candidates really so bad that you want that compromise?"
"Which feels greater, the threat of someone worse winning, or the
promise of someone better winning?". Or "How far does it feel like i
need to compromise--who's likely the best I can get?" ...etc.

In that way, too, you're voting in a way that's optimal in terms of
your feelings, your perception of the situation. That might not sound
as good as voting in a way that's optimal in terms of the actual
unknowable configuration of other people's votes, but why not? After
all, the other voters don't know more than you do, so you're not at a
disadvantage. And if everyone approves those whom they like, then we
get the candidate liked by most.

Optimality in terms of your likes, feelings or perceptions, as opposed
to the actual unknowable configuration of other people's votes, is a
completely different way of regarding the election, and how to vote.
But it's just as valid, and works just as well, for the individual
voters and for society.

In summary, Approval is a lot better than people think it is.

Additionally, I've spoken about how mis-estimates about relative
faction-sizes, mis-estimates shared by all the voters, result in
Approval tending to elect CWs, mitigating the results of a
mis-estimate made by you.

For instance, if you don't approve Middle, because you mistakenly
think that Favorite has a majority, then the Worst voters, having
access to the same information you do, aren't going to think that
Worst is bigger than Favorite, and so they'll approve Middle. Your
mis-estimate didn't cost you.

But many want the "rank-balloting ideal"--They want to fully protect
their preferred set (as they so easily can in Approval), while still
having an easy, strategy-free choice _among_ their preferred set. They
want entirely strategy-free voting. I don't think that's necessary.
Approval, or other A&c, would be fine, under any conditions.

But the rank-balloting ideal is achievable, for some voters--the ones
in a mutual majority (MM). (Of course Gibbard & Satterthwaite pointed
out that it can't be reliably available to everyone).

IRV&c:
----------

The popularity of the rank-balloting ideal can be judged by the great
popularity of IRV. As mentioned above, at least 5 U.S. political
parties offer IRV in their platforms. IRV is the only alternative
voting system offered in a party platform.

As noted above, the rank-balloting ideal is achievable, at least for
MM-members--and that's good enough.

The MM voters' strategy-freedom in IRV, and the other IRV&c, is the
other, opposite, strategy-advantage extreme for voting systems.

Of course every advantage comes at a price--some disadvantage, or the
loss of some other advantage

In the case of IRV, its strategy-freeness for MM-voters comes at the
cost of favorite-burial need for non-MM voters, mostly caused by lower
Condorcet efficiency, and the prohibition against equal ranking. That
isn't undemocratic or unfair, because there's nothing wrong with
government by a cohesive majority.That's ok.

The failure to elect a CW can, itself, count as a disadvantage, when
it displeases CW-preferrers, causing them to side with the non-MM
voters in a vote to replace IRV with something more Condorcet
efficient. The disadvantages named in this and the previous paragraph
could result in a majority voting to replace IRV with a more Condorcet
efficient method. That's ok too.

It isn't that IRV isn't a good method (for the Green scenario). It's
just that it could be vulnerable to replacement with something that
better elects CWs.

But, other than that, not electing the CW isn't as bad as some seem to
think it is. If the MM vote sincerely, then IRV chooses from the MM
preferred set. How important is it which member of that set is chosen?
To give an example, I just want to elect a progressive. I don't care
which one. I'd be glad to approve them all in Approval. Well, if I'd
help them all in Approval, because I just want one, any one, of them
to win, then why should I start caring a lot which one wins when the
method is IRV? It's enough that IRV would choose from the MM preferred
set.

...But, as I said, that might not be ok with the CW preferrers, when
the CW isn't chosen. Because that would displease them, and the non-MM
voters, then IRV might get replaced. Those 5 political parties want
IRV in spite of that, and there's nothing wrong with their finding out
from experience that they'd prefer something more Condorcet-efficient.
That's ok.

Some people get all hysterical about failure to choose the CW, but if
it bothers a majority, they'll change the method. Don't worry. It's
going to be alright.

When IRV doesn't choose the CW, when it chooses among the MM preferred
set, it's often because the favored wing are more numerous than the
CW-preferrers. What's unfair about that?

For Green scenario conditions, IRV is a lot better than some people think it is.

Nevertheless, that CW-elimination disadvantage, with the resulting
replacement-vulnerability for IRV can be avoided.

I've spoken of the Condorcet-IRV hybrids, of which James
Green-Armytage described four, in his article at
http://econ.ucsb.edu/~armytage/hybrids.pdf

A simpler one is Unbeaten//IRV:

If there's a candidate with no pairwise defeats, then elect hir. If
there isn't one, then do IRV among all of the candidates.

(Of course, if there are more than one unbeaten candidate, then you
could do IRV just among those)

It might be tempting to say to do IRV among the Smith set. That's
Smith//IRV, one of James' hybrids. But James points out that
Smith//IRV fails Mono-Add-Plump and Mono-Append.

All four of James' hybrids meet the Smith Criterion.

Another way of avoiding or mitigating IRV's disadvantage would be what
could be called Approval IRV, or AIRV:

Equal ranking allowed. If you rank several 1st choices, they all get
an initial whole vote from you.

When all of your candidates at rank N are eliminated, then all of your
candidates at rank N + 1 each get a whole vote from you.

In my brief IRV definition, that amounts to saying that a candidate
tops a ranking if s/he is one of the candidates at that ranking's
highest rank position that has uneliminated candidates.

Though AIRV isn't as good as the Condorcet-IRV hybrids, I'll include
AIRV in IRV&c (I&c), as one of the ways to avoid IRV's disadvantages.

Another hybrid:

Dennis Hill said to, at each stage of IRV, exempt the current CW from
elimination.

In contrast, Benham said to elect that CW when one appears during IRV.

I don't know which hybrid, of those mentioned by James, and any
others, is better, but, collectively, they're my favorite methods for
the Green scenario.

That's mostly because 1) Rank balloting and the rank-balloting ideal
are of great interest to progressives; and 2) Experience with polling
suggests that people do better with Score than with Approval, and that
they do better with rankings than with Score; 3) Of course we'd all
like the luxury of the rank-balloting ideal.

Personally, as an individual voter, I'd have no objection, even in the
Green scenario, to Approval, Score, ICT or Symmetrical ICT, because
choosing among the progressive candidates wouldn't be so important to
me.

------------------------------------------------------

Anyway my point in this post is that A&c and I&c are the opposite
extremes of an important strategy-advantage distinction. ...and that
that distinction is significant for classifying voting systems,
because of the universal agreement about the desirability of the
rank-balloting ideal.

Though I refer to them as extremes, I'm not saying that there are
methods between them. Those "extremes" consist of the strategically
desirable voting systems.

Of course there are methods that don't qualify for either
classification, and I don't say that those are "between" A&c and I&c.

-------------------------------------------------------

IRV brings MM voters strategy-freeness at a cost, as described above.
The cost is worth it, but it can be avoided, as described above, by
the hybrids or AIRV.

IRV doesn't really have a strategy-requirement, for MM voters to be
able to make IRV work. Sure, a mutual majority would have to be voted
as such--but why wouldn't it be?

The methods that avoid IRV's cost the (hybrids) likewise bring their
own cost. Some degree of ethics or consideration of consequences is
needed. As is so often the case, with power comes a requirement for
responsibility, if you are to benefit from it. I think that the voters
will qualify for successful use of the hybrids.

AIRV mitigates IRV;s disadvantage by allowing Approval-like voting.

Under Green scenario conditions, there's a case for IRV, the hybrids, or AIRV.

In fact, of course, even under Green scenario conditions, there's a
case for A&c too.

Michael Ossipoff