[EM] Kristofer, 4/18/12

[Kristofer Munsterhjelm]

On 04/18/2012 10:25 PM, Michael Ossipoff wrote:
>>> You see, there are methods for which we can assure people that it's
>>> been proven that no one can benefit from favorite-burial.

>> Yes, such methods exist, but you have to give up quite a bit to get FBC.

> You have to give up Condorcet's Criterion. What else do you have to give
> up, in ICT, to get FBC?

I haven't run ICT through my criterion compliance program, mostly 
because the codebase is not altogether good enough and I don't have the 
time to work on it. The kind of work for which I get paid takes up too 
much time.

But I'm pretty sure that ICT fails clone independence. Clearly CT does: 
just construct a case where the Smith set contains all the candidates, 
then you can use the simple vote-splitting example that shows regular 
old Plurality fails clone independence.

You've also said that ICT fails the Plurality criterion (which I persist 
in saying is not about Plurality, the method). You've also said it 
elects C in Kevin's MMPO example. You don't consider that important, but 
you've said that it could cause real trouble for ICT's proposability.

I'd also *think* that ICT fails Condorcet loser, but I don't know that 
for sure. The proof would set up a system where ICT's analogous Smith 
set differs considerably from the real Smith set, so that the Plurality 
winner within the analogous Smith set is the Condorcet loser in practice.

On a side note, now that I remember, I think the FBC you're talking 
about is not the strong FBC (which almost no method passes), but the 
ordinary FBC. If so, then that doesn't mean that betrayal doesn't pay 
off. It simply means that for all situations where betrayal could pay 
off, there also exists some other strategy that gives the voter the same 
amount of (or more) strategic additional power. And if that's the FBC 
you're talking about, then it doesn't mean nobody benefits from favorite 
betrayal, and the actual degree of protection would depend on how simple 
the non-betrayal strategy is.
Of course, if you're referring to the strong FBC, then this paragraph 
does not apply. Does ICT pass the strong FBC?

> You continued:
>> In Approval's case, you have to give up the graduation of the ranked
>> ballot and do part of the parsing yourself.

> Approval's strategies are simple. Condorcetists and IRVists are missing
> that.

Yes, "frontrunner plus" is pretty simple. However, it also adds another 
component to the game: the polling loop. The voters need feedback to 
know where to put the threshold (or they can guess, which will give them 
less power if they're wrong). That feedback then should be rolled into 
the next iteration of polling, to which the voters react *again*, and 
the polling adjusts, until equilibrium.

Simple? Perhaps, but I can't help but feel it's a hassle. And if the 
feedback goes just a little wrong, the method will fail precisely when 
it's needed the most: with candidates of similar strength.

> But it's true that the unlimited rank methods allow a gradation of
> preference that isn't expressed in Approval. The benefit of that is
> largely or entirely illusory for nearly all rank methods.

> Concider Condorcet, for instance. I've mentioned this subject before,
> but I'm probably saying it in more complete detail now.

> Condorcet has several strategy problems that undermine the goal and
> purpose of rank-balloting:

> All of these strategy problems will make a joke of Condorcet's criterion
> and the supposed advantage of better expresivity. How much
> do those things count for when people are engaging in these strategies?:

> 1. Condorcet, like Approval, is not defection-resistant. Lack of
> defection-resistance is Approval's only actual problem. In other
> words, Condorcet retains Approval's only problem worth mentioning.
> The defection problem can be dealt with perfectly well in Approval,
> in (at least) the five ways that I listed and described in a previous
> post. It can probably be similarly dealt with in Condorcet. The point
> is that Condorcet shared Approval's problem. Rank balloting doesn't
> get rid of it.

It is if more people are honest. If voters want to be honest in 
Approval, what choice do they have? Should they take dice to the polls 
so that they can probabilistically emulate Range? Somehow, I don't think 
they're going to do that.

True, if everybody decides to bury or truncate to get as much as they 
can, to the effect you get a sort of internal Approval election, then 
there will be an unstable tipping point like in Approval. On the other 
hand, if that's not the case, then Condorcet can glean more information 
from the rank ballots than from Approval's one-or-nil ballots.

(And in Burlington-style elections where there's an n-way contest but 
it's not too close, ranked voters under good systems can give their 
actual preference. Yet in Approval they'd be unsure without polling data 
- at least I would.)

> 2. Favorite-burial due to FBC failure. I've talked a lot about why FBC
> will be a problem even in Condorcet. You argue on that matter later,
> below, and I answer your arguments there.  So, here, I'll just ask, How
> much do Condorcet's Criterion and the rankings' free expressivity mean
> when people are burying their favorites to maximally help compromises?

When everybody are burying, not a whole lot. When people are fully 
honest, then there's no maximal burying. In between, the benefit is also 
in between those edge cases. In the very worst case, people can do the 
sort of countermeasures that you say can be done in Approval -- but they 
don't need to care about them if they don't find themselves close to 
that worst case (unlike Approval).

> 3. Offensive burial. With only 3 candidates, Condorcet well-deters
> offensive burial. With more than 3 candidates, that deterrence pretty much
> evaporates. I can find some losing candidate, someone who definitely
> won't pair-beat mine, and my faction can make hir beat the sincere CW.
> When that's being done, how much do Condorcet's Criterion and
> expressivity mean?

Eh? If there is no cycle, then you need a faction that's at least a 
majority to make that losing candidate beat the sincere CW. So someone 
must have already established a cycle, and even then, your faction has 
to outweigh the faction that voted according to their preference of the 
sincere CW over the loser.

> Besides, whether or not offensive burial is rampant, you can bet that
> opponents will make much of it. And they'll have a lot more airtime
> and print-space than you will. And remember that you can't prove that
> offensive burial will be rampant. You can say that you think it won't be.
> Opponents and media, etc., can continually emphasize that no one knows
> for sure, and that we'd be taking a chance on a method with which
> we don't know what would happn.

So we let experience make our case. I'll talk about that later, but for 
now, I'll just say that IRV, even with its 2.5 party problem, has been 
tried in various places. That problem is worse than offensive burial, in 
my opinion, because IRV picks the wrong winner even when given mostly 
honest ballots.

> In MCA, MTA, etc., the middle rating is only for candidates barely
> qualifying for an approval in Approval, and candidates almost
> qualiifying for an approval in Approval. In ABucklin, I suggest that the
> whole range of rank positions below top and above bottom should be reserved
> for those same candidates whom you'd middle rate in MTA or MCA.

So those are still Approval-like, in that you're supposed to portion out 
ranks according to polling information? I think that's quite different 
from what people consider ranked ballots to imply. I know that G-S and 
Arrow means we can't have a method that induces the voters to only rank 
in the "X above Y implies you prefer X to Y, nothing more" manner. Yet 
to have the intended (non-strategic) sense of operation to be different 
from that... strikes me as odd.

> You wrote:
>> We know from Burlington that compromise isn't absolute. In the 2009
>> election, voters voted in a way suggesting a close race. (Then IRV
>> punished them for it.) If the election had been Condorcet, the right
>> candidate would have won. Wouldn't that encourage those who didn't
>> compromise to keep not compromising, and suggest to the compromisers
>> that compromise wasn't needed? Couldn't that, after enough elections,
>> show the voters that they don't need widespread strategy?
> [endquote]
> Sure, I don't deny that. I just doubt that it people will stop burying
> favorites under one of "the two choices",  and so people will never have
> that experience in elections for national office.

Not unless they had already got off the overcompromising habit, no.

> You continlued:
>> FBC is obviously a much stronger assertion. When you're dealing with a
>> method that passes FBC, you know (and can tell the voters) there's
>> absolutely no need to betray a favorite. There's no need for the kind of
>> feedback I gave above. Yet if the scenario I gave is realistic, that
>> means one doesn't need to dismiss other methods like Condorcet for
>> public elections.
> [endquote]
> But what you said was based on Burlington. Municipal elections. Voters
> are a lot more sincere in municipal elections than in elections for
> national office.

Wouldn't you have to start there? Not even FairVote tried to go right 
for the Presidential prize. They've tried to get IRV passed in several 
local areas, and have succeeded in some. (They're now seeing backsliding 
several places, but at least they got it tried.) Similarly, the PR 
leagues that got STV passed in New York started locally. They didn't try 
to change the composition of the Senate or the House of Reps directly.

For something as wide-reaching as altering a voting system, I think the 
leap from no change directly to presidential or national reform would be 
tall indeed: impossible or nearly so to cross. By contrast, a gradual 
approach lets the system prove itself. IRV failed to prove itself in 
Burlington. Condorcet might have.

A gradual approach isn't perfect, either: the push to PR was countered 
and proportional representation eventually quashed. Perhaps that 
happened because PR worked too well. Yet I don't think PR would have had 
any chance if the organizations had tried to go for the big, national 
changes first.

> You continued:
>> But if the voters are like RBJ, for instance, they may prefer a ranked
>> ballot to an Approval or Range-type ballot; or they may think that (if
>> the concept was explained to them) the CW should always win because he
>> has a majority backing him against anyone else.
> [endquote]

> Though most people new to voting systems aren't interested in Condorcet
> complexity, some people do like rank balloting. Of those latter, nearly
> all want Borda or IRV. But even if they preferred Condorcet, it wouldn't
> matter, when media and authorities insisted that the new proposals are
> unpredictable, may have unknown adverse & undesirable, disastrous,
> consequences.  ...and when they say, "This will need a lot more study."
> Some people initially liking rank balloting won't be enough to overcome
> that.

You keep saying that the authorities will claim "this will need a lot 
more study", block change, and also claim the new proposal is 
unpredictable. Yet IRV did enjoy initial success. IRV, with its 
attendant nonmonotonicity and sudden shifts, is a lot less predictable 
than Condorcet, but there weren't really any opposition pieces focusing 
on its chaotic behaivor before Warren (and Gierzynski) wrote about the 
various paradoxes shown in the Burlington election.

According to Robert, the IRV repeal campaign in Burlington instead 
focused on the "obvious nature" of the Plurality system, i.e. "vote for 
your favorite and that's it". Arguments of the sort that other methods 
are unnatural would happen no matter what method you picked - just look 
at the web poll comments on Approval, which range from  "Approval is a 
liberal ploy, will they count votes for a single candidate more than 
once too, hur hur" to "this violates one man one vote".

Given that those kind of arguments will appear for any method, the best 
we can do is to make the outcome seem right and not have the method act 
too strangely. Again, Robert suggested (as have I, earlier) that if the 
CW won, because a majority would prefer the CW to everyone else, no 
repeal campaign could unite against a single candidate. If Montroll had 
won, the Wright voters would have said "well, it's still better to have 
Montroll than Kiss", while the Kiss voters would have said "well, it's 
still better to have Montroll than Wright", which would seriously weaken 
the repeal.

> You continued:
>>   Throwing away FBC and not getting anything in return is of course
>> silly, but throwing away FBC and *getting* something in return might not be
> [endquote]
> Ok, sure. As I said, if I could decree the voting system here, I'd
> decree that it be ICT.
> But take a closer look at the word "getting", in your above sentence. We
> can't get anything more complicated than Approval.

Why not? Why couldn't Condorcet (or some other method) take IRV's path, 
but avoid IRV's 2.5 party problem and thus continue further?

> You continue:
>> at least if FBC failure need not be as serious as you say.
> [endquote]
> I've seen its effect on Condorcet voting, in a presidential Condorcet
> Internet poll. I only observed one voter,so this is "anecdotal".
> But where is the burden of proof? Should we enact Condorcet (as if it
> were possible) instead of Approval because we don't know if a
> significant number of voters will favorite-bury?
> And if someone would favorite-bury in a staw poll, where nothing's
> riding on the outcome, then there's all the more reason to expect
> rampant favorite-burial in an actual election.

We don't know if a significant number of voters will favorite-bury after 
less competitive elections show that they don't have to. If you could 
get Approval passed, like that, at the national level, that would be a 
different matter.

I would compare evidence: your one voter against the various 
organizational Condorcet elections. But that wouldn't work (I don't know 
whether anybody actually favorite-betrayed in those elections, just that 
they seem to give good results), and it wouldn't address the issue (as 
you've said, since you consider organizational elections to inherently 
have less favorite betrayal).

Instead, let me just say this: I've given a reason for thinking that 
voters may wean themselves off favorite betrayal (just like they got 
used to it), and that I think the way you'd go about doing the reform 
(from the bottom up) would mean this effect would indeed take palce. Can 
you wave a wand and have Approval instated on a national level without 
doing it locally first?

>> Perhaps one can show that in an u/a setting, Condorcet passes FBC; and
>> you could reason that when voters are no longer thinking in u/a terms,
>> they have already moved beyond Plurality-style reasoning.
>> [endquote]
> That hadn't occurred to me. But if it's understood by all to be  u/a
> elections, then there would be very little reason to have anything other
> than Approval.

Stepping stones. u/a FBC helps the electorate get out of their 
overcompromising mode. Then, once they're out of it, they can move past 
u/a - because the candidates can start campaigning in ways that don't 
rely on "us vs them".

> The matter of Condorcet and FBC in a u/a electoin hadn't occurred to me,
> and I don't know--Would Condorcet pass FBC then? Maybe; I don't know,
> because I haven't considered that matter before. If it would, then, as
> you said, maybe, under u/a conditions maybe voters in Condorcet would
> get over their overcompromise habit. It's a new idea to me.

It might pass. I think it'd be easier to prove for a Smith method: show 
something like that you can't get a "honest" cycle with u/a (except 
among acceptables", and then show that you don't have an incentive to 
start a cycle strategically. I'm not sure how you'd do the details, though.