[EM] Approval vs rank methods

[Dave Ketchum]

On some of this we seem to be talking different languages, and thus not
communicating - but there are a few points that may be worth more effort.

On Fri, 11 Nov 2005 05:59:06 +0000 MIKE OSSIPOFF wrote:

 > Dave--
 >
 > You wrote:
 >
 >
 > [replying to my Approval handtool analogy]
 >
 > Tools vary in their abilities:
 >      Plurality is excellent if approving of a single candidate is all I
 > desire.
 >      Approval fills a gap - I can approve, equally, of one or more
 > candidates.
 >      Sometimes I want more than either of them is capable of, and desire a
 > more capable tool.  Caring not where you divide tools between handtools 
and
 > machines.
 >
 > I reply:
 >
 > ...but you want something whose input is all of the voter's pairwise
 > preferences. And presumably something that makes good use of those 
pairwise
 > preferences, by some important standard. That requires a device, an
 > automatic machine. That's what you're saying you want.
 >
 > When you collect that information, you either have a good automatic
 > majority-enforcement machine, or you have a shabby piece of shit like IRV.
 >
 > I'd said:
 >
 > Rank methods are majority-enforcing machines. A machine can be 
well-designed
 > and high-quality, or it can be a cheap and defective machine like IRV.
 >
 > You replied:
 >
 > While we can agree on wanting better than IRV, since we know better is
 > available, does IRV deserve your nasty words?
 >
 > I reply:
 >
 > I wouldn't say it otherwise.
 >
 > You continued:
 >
 > Easy enough to describe a possible ugly failure, but how often might such
 > happen in real elections?
 >
 > I reply:
 >
 > Well, someone said that it would have happened in Debian's most recent
 > election, if they'd been using IRV instead of Condorcet(wv).
 >
 > Say there are 3 candidates, on a 1-dimensional political spectrum, with 
one
 > between the other two. The middle candiate is a CW if the 1st choice
 > popularity of all 3 candidates varies by no more than a factor of two.
 > That's a sufficient but not necessary condition. All it takes is for 
Middle
 > to be bigger than the difference between Favorite and Worst.

Trying:
       35 F
       33 M
       32 W

Difference between F and W is 3, M is much more, F is CW, and IRV will
agree.
 >
 > And then, when that's so, when Middle is CW, what is the probability that
 > Middle will be eliminated first, if we have no information about their
 > relative popularity? The probability is 1/3.

Trying:
       35 M
       33 F
       32 W
Difference between F and W is 1, M is much more, M is CW, and IRV will
agree that CW wins ONE HUNDRED percent of the time.
 >
 > And even if the middle CW is favorite of more voters than any other
 > candidate is, that CW is sure to lose if favoriteness-strength tapers 
at all
 > gradually away from the median point. That's not a contrived scenario.
 >
 > Then, the low-popularity candidates at the extremes will be eliminated
 > first, ane their votes will transfer inward, till candidates other than 
the
 > middle CW build up accumulated vote totals that eliminate the CW.

You HAVE NOT described these voters as offering any second choices.
Without there being second choices there cannot be any transfers, and
cannot be anything in what you describe to cause IRV to award to other
than the CW.

Further, unless you specify otherwise, second choices could go to, and
thus strengthen, the CW.  In fact, for 1-dimensional, if M is CW AND is 
favorite of more voters than any other, then F or W has to be loser per 
IRV, and their second choices would more likely be M than the opposite 
extreme.
 >
 > Several of us have posted such examples to this mailing list.
 >
 > So, to answer your question, IRV's problem will happen a lot.
 >
 > I'd said:
 >
 > If the voters are as incompetent as they seem to show that they are, 
they're
 > going to need a good majority-enforcing machine so that they won't have to
 > incompetently make the all-or-nothing strategic decisions that the 
handtool
 > Approval would require of them.
 >
 > You reply:
 >
 > While we can debate how much the voters might bother to learn, when they
 > have no opportunity to use such knowledge, should we assume that none 
would
 > learn if and when it started to make a real difference?
 >
 > I reply:
 >
 > Not with certainty, no. As I said, the voters haven't been tried with
 > Approval, and no one can say for sure if they'd be unable to stop 
voting for
 > corrupt lesser-evil sleaze if their favorite outpolled their greater-evil.
 > But what I've been saying is that the voters' behavior so far doesn't
 > justify a lot of confidence in their competence.
 >
 > You continued:
 >
 > Also, how can I learn to use Approval to rank candidates according the
 > acceptable vs only-tolerable qualities they display.
 >
 > I reply:
 >
 > You can't rank candidates in 3 tiers (acceptable, only-tolerable, and
 > intolerably unacceptable) with Approval. I didn't mean to give the
 > impression that the voters are incompetent because they don't know how to
 > rank candidates in 3 tiers with Approval.
 >
 > It's the other way around: Because they're incompetent, it's probably 
better
 > if they don't have to make the choices that Approval demands of them, the
 > choice of which of their pairwise preferences to vote.
 >
 > You'd said:
 >
 > Approval does not let competent voters vote their decisions full strength.
 >
 > I'd replied:
 >
 > Corrrect: Approval doesn't _let_ people vote pairwise preferences full
 > strength: It _forces_ them to.
 >
 > You reply:
 >
 > Do you use this refusal to see what is written when you debate with Markus?
 >
 > I reply:
 >
 > You need to be a little clearer about what you mean.
 >
 > I said that Approval forces people vote full-strength, all-or-nothing
 > preferences. It does. It doesn't allow them to. It forces them to. I stand
 > by that statement.
 >
 > You say I didn't notice what you wrote. No, it's that you didn't write 
what
 > you meant. You said that
 > Approval doesn't let people vote their decisions full strength. On the
 > contrary, every pairwise preference that someone votes in Approval is
 > full-strength, in the strongest all-or-nothing sense. Full-strength in the
 > exaggerated sense of top vs bottom.
 >
 > What you meant was that Approval doesn't let people vote all of their
 > pairwise preferences. I didn't reply to that (in that sentence), 
because you
 > didn't say it. Later, however, I replied to that, because I assumed 
that it
 > was what you meant. No, Approval doesn't let them vote all their pairwise
 > preferences. Just the important ones. And, unlike IRV, Approval counts 
every
 > pairwise preference that you choose to vote.

When I go to the dictionary, "does not let" translates into "forbids" - 
FAR FROM "forces to".
 >
 > You continue:
 >
 > It only wastes everyone's time.
 >
 > I reply:
 >
 > You mean like debating soimething that we don't disagree on? I, too, 
prefer
 > (the best) rank methods for public elections. We don't disagree on that.
 >
 > You speak of wasting time. We've gone over the same ground a few times.
 >
 > As for Markus, can you name an instance in which I disregarded what he 
said,
 > except when I finally got tired of replying to his parrot-like 
repetition of
 > answered statements?
 >
 > I'd said:
 >
 > Approval, which I also call Set Voting, lets you vote one candidate-set 
over
 > another. That lets you vote the pairwise preferences that really matter.
 > Anything more would be a frivolous luxury--except that, as I said, our
 > public election voters need that luxury.
 >
 > You reply:
 >
 > It would be a luxury if the cost exceeded the expectable benefits.
 >
 > I reply:
 >
 > No, then it would be a prohibitively costly luxury. But it can be a
 > frivolous or unnecessary luxury without being a prohibitively costly 
luxury.

Back to the dictionary for "luxury":  "a condition of abundance or great 
ease or comfort".  I think of you claiming that a car being able to be 
driven at over 50 mph would be a luxury, for I could travel 200 miles at 
50 in part of a day.  Ability to do 65 mph I class as a necessity since, 
with a 65 mph speed limit on many roads I use, it would be dangerous to 
drive on such roads at 50.

Back to voting:  Ability to do rank voting is extended ability.  If worth 
the added expense it does not qualify as a luxury; if we never had more 
than 2 candidates in a serious election it would be a useless luxury.
 >
 > You continued:
 >
 > Most of us seem agreed that rank falls on the affordable side of luxury.
 >
 > I reply:
 >
 > Don't be so sure. I'd say that a lot of us hope that a good rank method 
can
 > be enacted. Some of us have talked about why that could be a problem: 
There
 > are innumerable ways to count rank ballots. We'd have to show people that
 > our proposed way is the best, or at least that it's good enough to go 
ahead
 > with enactment, without (never-ending) studies being conducted, though 
there
 > are infinitely many other ways to count rank ballots. We'd have to wade 
into
 > the bog of showing why IRV is crap, and why our proposal is better than 
IRV.

Agreed that there are many details to debate.  That does not prevent 
agreeing that a worthy variation should be choosable.
 >
 > In contradistinction, there's only one way to count Approval ballots or RV
 > ratings: Add them up. RV wins the prize for familiarity and 
popularilty. RV
 > is the best bet for a public proposal.

This I do not see.
 >
 > I, too, would hope that the best rank methods, MDDA, MDDB,
 > MDD,ER-Bucklin(whole), and MAMPO, would be accepted and enacted. Maybe the
 > enactment of a good rank method is too much to hope for. But of course if
 > people are willing to bounce back with a backup proposal, then it wouldn't
 > hurt to try a good rank method before trying RV or Approval.
 >
 > You wrote:
 >
 > The debates over rank method details sometimes get a bit deep, 
depending on
 > plotters knowing unknowable details and having unbelievable control.
 >
 > I reply:
 >
 > Maybe sometimes, but not when I write about it. No, the problem that I 
write
 > about is when the voter _doesn't_ know what s/he needs to know, and 
decides
 > to play it safe by voting Lesser-Evil over Favorite.
 >
 > You wrote:
 >
 > Again, IRV counts most [pairwise preferences], and its backers claim it
 > counts enough.
 >
 > I reply:
 >
 > Its backers are full of shit. How consoling is it for you if IRV counts 
many
 > or most people's preferences but ignores a crucial pairwise preference of
 > yours?

Agreed IRV has problems, though not as serious as you claim.  There can 
also be unhappiness with resolution of a Condorcet cycle.
 >
 > I'd said:
 >
 > Exactly. And the incompetents that I've referred to don't do a good job of
 > choosing which of those pairwise preferences to vote. That's why I want 
them
 > to be able to vote all their pairwise prefences, and have them all 
counted,
 > by a method with majority-enforcing criterion compliances.
 >
 > You reply:
 >
 > I was not talking of voter incompetence
 >
 > I reply:
 >
 > Of course not. I was talking of incompetence.
 >
 > You continued:
 >
 > , for Approval restricts their possible actions.
 >
 > I reply:
 >
 > As I clarified above, I'm not saying that the voters are incompetent 
because
 > they don't know how to vote 3-tiered preferences in Approval.
 >
 > It's the other way around: It's because they're incompetent that they need
 > an automatic majority-enforcing machine that uses all of their pairwise
 > preferences, rather than a hand-tool that requires them to choose which of
 > their pairwise preferences to vote.
 >
 > You continued:
 >
 > Still, I would not claim Approval incompetence, assuming it behaves as
 > promised.
 >
 > I reply:
 >
 > I never said that Approval was incompetent, or questioned your 
competence to
 > use Approval.
 >
 > You continued:
 >
 > What is left is designer incompetence, for not letting voters express 
their
 > affordable thoughts.
 >
 > I reply:
 >
 > You're saying that Weber and Ottewell are incompetent because they didn't
 > design a method that will work well for idiots? Blame the idiots, not the
 > Approval inventors.

I have no idea as to the designers of Approval.  If they wanted to get 
past Plurality yet did not see rank as practical in their environment, 
they did a proper intermediate step.

Actually I see nothing working well for idiots, for I expect nothing can - 
what I wish is giving power to those who can absorb a basic education.
 >
 > Approval is a beautifully elegant method, and it's only because of 
questions
 > about voter-competence that I don't call Approval an excellent choice for
 > today's public elections.
 >
 > You sound like someone complaining that Harley-Davidson was incompetent
 > because it designed and sold to you a motorcycle that didn't have
 > training-wheels.
 >
 > I'd said:
 >
 > Yes, sincere rating is more work. But no one forces you to do that 
work. You
 > can just give maximum points to the candidates you'd vote for in Approval,
 > and give minumum points to the others. That's your best strategy. And
 > another reason to use that Approval strategy is that Approval is easier to
 > vote.
 >
 > You reply:
 >
 > If this is your recommendation as to how voters should act with RV
 >
 > Yes it is. (If that's what I said).
 >
 > You continued:
 >
 > Yes, if you complain, as you just had done, that voting in RV is a lot of
 > work, then I recommend voting RV as if it were Approval, to make the job
 > easier.
 >
 > You continued:
 >
 > ..., why should it bother to exist?
 >
 > I reply:
 >
 > 1. Because it's much more familiar and popular than Approval or any rank
 > method

Where did all this supposed familiarity come from?

Did those who called it familiar understand what they said?

 > 2. Because it likely will give somewhat better results than Approval in
 > today's public elections.

Assuming it does, that is not complete proof of value.

 > 3. Because some might want to use RV's ratings to vote sincerely.
 > 4. Because, like Approval and the best rank methods, but unlike IRV,  RV
 > meets FBC.
 > 5. Because, unlike IRV, RV meets WDSC, in addition to FBC.
 >
 > I'd said:
 >
 > I disagree there: RV doesn't give any problem to counters. They merely 
have
 > to add up each candidate's points, and declare as winner the one with the
 > most points.
 >
 > You replied:
 >
 > Ok, so the counters do not try to guess
 >
 > I reply:
 >
 > Correct. They don't.
 >
 > You continued:
 >
 > ..., leaving the whole communications headache on the voters' weak
 > shoulders.
 >
 > I reply:
 >
 > Take the difficult choices off the voters' weak shoulders by using MDDA,
 > MDDB,
 > MDD, ER-Bucklin(whole) or MAMPO instead of Approval or RV.
 >
 > You'd said:
 >
 > [with rank methods] Voters can indicate order of their preferences.
 >
 > I reply:
 >
 > A necessity for our incompetent voters. Otherwise a frivolous luxury. As I
 > said, Set Voting  (That's Approval) lets you vote one candidate-set over
 > another. Those are the important pairwise preferences,  and that should be
 > enough.
 >
 > You reply:
 >
 > Proper topic is whether capable voters are allowed to express their
 > important thoughts
 >
 > I reply:
 >
 > Answer: Yes. With Approval, capable voters are allowed to express their
 > important preferences. Incompetent voteres are allowed to likewise, 
but, for
 > them, that doesn't help much.
 >
 > But no, with Approvla, capable voters (and, worse, not-so-capable ones)
 > aren't allowed to vote all of their pairwise preferences.
 >
 > You continued:
 >
 > ..., not whether some voters may be incapable of better thinking - for 
which
 > more adequate education may be practical.
 >
 > I reply:
 >
 > I disagree again: The capabiliies of the vulgar masses is something that
 > affects voting outcomes.
 >
 > Education? Maybe honest &/or democratic media. Failing that, the best 
voting
 > system we can enact would do a lot of good. Some are workig to improve
 > media. Here we're talking about the benefit of a better voting system.
 >
 >
 > You'd said:
 >
 > Counters can read these preferences in the same language as the voters 
used
 > for writing.
 >
 > I reply:
 >
 > That's true with Approval and RV too. And I remind you that Approval 
and RV
 > receive information that rank methods don't receive: Preference 
strength. In
 > that regard, Approval and RV are more expressive than rank methods.
 >
 > You reply:
 >
 > You lose me here.  You have suggested above ZERO ability to express
 > preference strength with Approval
 >
 > I reply:
 >
 > Would you please quote the sentence in which I said that, and tell the 
date
 > of the EM posting that contained that statement?

Quoting from the above, IN THIS POST:

>> I said that Approval forces people vote full-strength, all-or-nothing 
>> preferences. It does. It doesn't allow them to. It forces them to. I stand 
>> by that statement.

They cannot express amount of strength with Approval - only zero or infinity.
 >
 > You continue:
 >
 > - only two sets with zero preference within a set, and uncontrollable
 > maximum strength between sets.
 >
 > I reply:
 >
 > As I said, Approval measures the strength of your pairwise preferences,
 > because you vote the ones that are most important.
 >
 > You continued:
 >
 > Agreed you say voters can express strength in RV and that counters have no
 > responsibility for understanding there - leaving how voters determine what
 > to say a puzzle.
 >
 > I reply:
 >
 > Let's avoid giving voters that puzzle by enacting MDDA, MDDB,
 > MDD,ER-Bucklin(whole) or MAMPO instead of Approval or RV.
 >
 > But, as a practical matter, RV is a lot more gettable, and, despite the 
fact
 > that Approval and RV don't help incompetent votes as much as we'd like, RV
 > may be what we can get.
 >
 > I'd said:
 >
 > You think that Approval doesn't allow you to express as much because it
 > doesn't let you indicate the direction of all your pairwise 
preferences. But
 > Approval lets you express their strength: You vote a preference if it's
 > important enough to you. Rank methods let you express the direction of all
 > your preferences, but that's all. They don't let you express the 
strength of
 > your preferences.
 >
 > No one denies that the best rank methods do the best job of enforcing
 > majority rule. But who says that's all that matters?
 >
 >
 > Agreed there are so many possibilities, and some offer so little value for
 > what they offer as to deserve "luxury" as a descriptor
 >
 > I reply:
 >
 > I disagree again. If a method doesn't offer anything of value, there's
 > nothing luxurious about that.
 >
 > You continued:
 >
 > , and some are so weak as to voters being able to express their 
thoughts as
 > to be useless, that we need more careful thought.
 >
 > I reply:
 >
 > Other than Plurality, all the methods are good at letting voters express
 > preferences. For today's public election voters, the luxury of being 
allowed
 > to vote all their preferences might be a necessity.
 >
 > Mike Ossipoff
-- 
   davek at clarityconnect.com    people.clarityconnect.com/webpages3/davek
   Dave Ketchum   108 Halstead Ave, Owego, NY  13827-1708   607-687-5026
             Do to no one what you would not want done to you.
                   If you want peace, work for justice.