[EM] Kristofer, April 3, '12, Approval vs Condorcet

Kristofer Munsterhjelm km_elmet at lavabit.com
Mon May 14 15:06:18 PDT 2012


On 05/03/2012 11:19 PM, Michael Ossipoff wrote:
>
> Kristofer:
>
> It's necessary to distinguish between _two_ Condorcet disadvantages that
> I spoke of:
>
> 1. Condorcet's FBC problem, when Condorcet is in use.
>
> 2. Condorcet's enactment problem, due to being a complicated
> rank-balloting contraption.
>
> You addressed #1, but not #2.

We're going in circles, but I can repeat what I have said. Sure I have. 
My proposal #2 is this:

Start locally. Let the voters' experience with Condorcet become its 
strength. You have said that people do not compromise as heavily under 
local elections, so let's make use of that.

In short, Condorcet's enactment problem is solved by applying IRV's 
strategy, but to do it right. It won't be easy, of course. Maybe we'll 
need a Richie, maybe not -- the proportional representation leagues 
didn't. But Condorcet will provide good results with voters who take the 
chance to not compromise, and it will keep on doing so even beyond IRV's 
2.5 limit.

In a way, #1 and #2 are interrelated. #1 shows how Condorcet can get 
around not technically meeting FBC. The feedback dynamic suggested in #1 
can then be used in #2 for enactment itself.

At this point you've said that Condorcet will get held up by people 
saying 'it is too complex, it needs more study'. Why? IRV didn't; and 
IRV is more complex than Condorcet. IRV's setbacks have, at least as far 
as I can see them, been primarily along two points: first, arguments 
that IRV is a trick, or is unamerican, or only Plurality is good enough 
for us (which also applies to Approval, see board posts); and seond, 
arguments that it behaves strangely - and the latter didn't really 
surface until Warren started pointing out the flaws in IRV.

> As for #1, anyone can say, regarding any criterion, "That is very
> important." If you lived here, you'd know that favorite-burial is
> _the_ problem. A combination of media deception and overcompromising
> favorite-burial are essential to  preserving the phoney "2-party-system".

Compromising is a problem because Plurality compromise incentive is so 
obvious. But, as I have said, even under Plurality, the voters sometimes 
say "forget about strategy", even on the national scale. Remember Ross 
Perot.

I don't think the case is as clear cut as you say, either. Many other 
Americans on the list support Condorcet even though they know that it 
doesn't strictly meet FBC. If it was obvious that compromise incentive 
is the big thing to the extent that nothing short of FBC compliance 
could salvage a method, they would have held other positions.

> Yes, that's right. Without favorite-burial incentive, you can kiss-off
> the media-fabricated two-party system. The Democrat and Republican
> parties would be history.
> Democrat and Republican politicians and party officials would have to
> find other scams, other than politics. Maybe they can go into the
> junkbond or S&L business, or something.
>
> We mustn't underestimate the degree that their electoral survival
> depends on favorite-burial.

Sure, and if you could find a method that passes IIA, you would 
eliminate every kind of strategic nomination incentive. But IIA is very 
strict, probably too strict.

If you limit yourself to FBC, then you're going to have a hard time 
finding a cloneproof method, for instance. Range is "cloneproof" in the 
sense that if rating differences between clones go to epsilon, then it 
doesn't matter how many clones there are, but ranked systems? Very hard 
indeed.

(Strong FBC is even more strict - you pretty much have only Approval and 
Antiplurality. I think someone wrote a paper about strong FBC earlier, 
and that there were some Bucklinesque and non-majoritarian methods that 
also passed it. However, you don't claim you need strong FBC, so that's 
a bit besides the point.)

> So, you or anyone can call _any_ criterion the important one, but I've
> told, above, why I consider FBC to be important.
>
> As for #2 above, Condorcet's enactment problem, it's the one that I've
> been discussing a lot. I don't like to repeat, anymore than people like
> me to repeat. It's a lot of work. But I'm going to repeat this:
>
> Rank methods are complicated contraptions. People won't know that
> they don't have unintended, unexpected consequences that could be
> disastrous. Media pundits, commentators, tv anchormen, newspaper and
> magazine editors and writers, talk-show hosts, and hired university
> authorities will say, "You don't know what that voting system will
> do. We can' change our voting system without being sure that the new
> one won't be a disaster. This needs a lot more study." And then, of
> course, it will never happen.

And as you keep repeating that, I keep repeating: where were the "this 
needs a lot more study" calls for IRV? Why did IRV, chaotic and 
nonmonotone in its effects, not "never happen"?

> That can't be said of Approval. Approval is the minimal fix of
> Plurality's problem. Nothing other than the elimination of
> Plurality's ridiculous forced-falsification requirement. I've told
> why it will be obvious to all that that can only be an improvement.

But will it last, or will you get a situation analogous to the 1800 tie 
between Jefferson and Burr? If the latter, couldn't that lead to 
backsliding?

And before you say that Condorcet, too, has the chicken dilemma: first, 
it is not as obvious. Second, it requires more strategy on each side. 
Third, despite how you keep repeating how "Condorcetists are in denial", 
there *are* Condorcet methods that handle the chicken dilemma. Or so you 
have said, yourself.

Really, I suppose what my intuition is going by at this point is: we 
have precedent that Approval has a very tense chicken dilemma situation, 
and that it has harmed Approval. We don't have any precedent for a 
similar situation harming ranked ballot methods that have been tried. 
Where they have been reverted, it has not been because of chicken 
dilemma situations.

> Remember that Condorcet's winnability problem has nothing to do with
> FBC. It's the "complicated contraption" problem that I spoke of above--a
> problem that Approval, uniquely, doesn't have.

And presumably a problem that IRV had, or Bucklin, or that even more 
complex thing used in New York, STV. Where were the "complicated 
contraption" voices then?

And where did the complicated contraption voices get top-two runoff 
disbanded? Limited voting? I don't think it's unique to Approval.

> You continued:
>
> It [Approval] has the defection problem
>
> [endquote]
>
> So does Condorcet, to the same degree. So let's not use that to compare
> those  2 methods--except that it's why I say that Condorcet doesn't
> significantly  improve on
> Approval.

To the same degree? No. If voters just rank and don't strategically 
consider whether to defect or not, there's no problem -- but in 
Approval, they can't "just Approve" without using implicit strategies 
outside of the method itself. Honest voters don't have to put any cutoff 
anywhere in a ranked method. They still have to put the cutoff somewhere 
in Approval. Sure, they can use strategies to guide where that cutoff 
has to be set, but they still have to set it.

Let's be a little more formal about this. Consider left-right politics 
with voters and candidates on a line, voters preferring candidates 
closer to themselves. Now, let's assume some candidates are close enough 
and have a common base by enough that two of them have nearly the same 
amount of Approvals.

To get the right winner in Approval, the voters have to fine-tune so 
that they don't get the other close candidate. No big deal, maybe? But 
in Condorcet, the voters just rank all the candidates and everything 
works properly (Black's singlepeakedness theorem). No matter how "almost 
as good" the other candidate is, that's not "to the same degree".

Or to put it in another way: Approval has a semi-honest/honest problem 
where Condorcet has none. When everybody are rational game theorists, 
then "Plain Condorcet" (i.e. not CT, Smith,Plurality, Smith,DAC or 
similar) is just as bad as Approval, true. But when the voters just want 
to get out there and vote... then it's not "to the same degree". It is 
not "to the same degree" when the fraction that votes honestly can 
shield against the effects of those that don't, either.

> That's so, regardless of why voters gave their approval. That is _their_
> business, and their business only. Don't worry about why. Just
> note that the winner is someone to whom the most voters have given an
> approval, by marking hir "Approved". And recognize their right
> to do so, without accounting or justifying it to anyone.

As you said, yourself: the same argument can be applied to Plurality. 
You then went on saying that the difference is that in Plurality, the 
voter's voting power isn't fully his.

Nobody is forcing the voter to vote a particular way in Plurality, so I 
presume you mean that the voter feels constrained to pick within a 
particular subset of the votes because otherwise, something bad will 
happen. In Plurality, the constraints are usually "vote for one of the 
two, or the other guy wins".

In Approval, the situation is a little different. The voter has a subset 
of votes among which it's only reasonable for him to vote (e.g. if he's 
Democratic, he wouldn't pick the vote that Approves the Republican but 
not the Democrat), but it's not always very clear what he should pick. 
For this, he can take a chance that his dithering (rendering of 
preferences to Approval style) will be good enough, or he can consult 
polls to dither more effectively. However, he has to make more effort, 
and significantly more the more close the election gets. This is about 
noise more than about trying to cancel out the other camp's strategy.

In both cases, however, the voter is trying to attain a given result by 
carefully considering the effects of what he does. He shouldn't have to 
do that. Yes, G&S dashes that hope in the absolute, criterion 
satisfaction sense, but within common settings, he shouldn't have to do 
it either. If he wants to "play the game", sure, then it makes sense 
that strategy should take some power of mind, but if he doesn't... if 
it's a left-vs-right, he should just rank and get the right result. If 
someone does offensive strategy, perhaps then he has to make use of 
counterstrategy, but not before.

> But that statement about operational approval was only part of what I said.
>
> I also pointed out that  there are many ways a voter could choose whom
> to approve. In all of those ways, the voter will always approve candidates
> s/he likes better than those s/he doesn't approve.
>
> Usually, maybe nearly always, people should, and will, just approve the
> candidates whom they like, trust, or consider deserving of their
> support. No strategy there.

So if the Greens become larger, Nader voters will know when to vote 
Nader alone or Nader+Gore depending on their trust?

> If it's a u/a election (there are unacceptable candidates who could
> win), then we'll simply approve all the acceptables, and none of the
> unacceptables.

Right.

> In the rare instances where we don't  just know whom to approve because
> we like or trust them, I've described some strategy suggestions.

Yes, you've described how to reduce preferences to a two-level ballot by 
the use of either preference alone or by external data (such as polls) 
-- what I've called "dithering" in analog to how you can turn a 
grayscale picture into black-and-white.

I'd rather not have to do it, though.

Hmm, now that I look at it, I might see a better idea of where we are. 
If people are honest or semi-honest (enough are honest that offensive 
strategy can't gather sufficient strength or work), then the dithering 
becomes additional burden on the voters and so isn't needed. If, on the 
other hand, the voters readily look to take advantage of each other, 
then dithering and poll synchronization would be better than having to 
go all "defensive strategy, offensive strategy, counterstrategy, 
counter-counter-strategy".

If so, then we return to the point I made right at the top. I've given a 
suggestion for how to get voters to stop doing static strategy - 
"just-to-be-sure" strategy - by showing how the method doesn't need that 
amount of static strategy. For dynamic strategy (coordinated by parties, 
like the vote management that failed under STV in New York), once static 
strategy is dealt with, the variety of candidates and parties would, I 
think, help take the edge off the "it's us or them, I have to make use 
of strategy" climate. At least that is the case here. We have many 
parties and strategy doesn't really come into it. It would be nice to 
have some sort of transfer of votes so I could vote for a minor party 
without having my votes go to waste, but I don't refrain from voting for 
the minor party even so.

[So it's a half-empty vs half-full thing, and that might be why we keep 
disagreeing without getting anywhere. When we look upon a situation 
where "doing dishonest thing X sometimes helps but usually has no effect 
either way", you think that voters will do X because it *may* help -- 
half-empty. I think that they'll generally not do X unless it's obvious 
they need to or knowing when it works is really obvious -- half-full. 
It's like that with FBC, and it's like that with ranking Approval style 
in MJ. I also think I see it in how we talk about the chicken dilemma in 
Approval: you consider it the case of strategists having to figure out 
where to put the line, but I consider it in the case of mostly-honest 
voters facing noise that could throw the election the wrong way, or of 
having to make use of external algorithms when they could, in certain 
cases (e.g. 1-D) Just Rank under Condorcet.]

> You continued:
>
> To make it really obvious, consider a similar pitch for Plurality. "In
> Plurality, the candidate who is favorite of the most voters win!".
> Sounds reasonable enough, doesn't it?
>
> [quote]
>
> No, it doesn't. Not to someone who lives here. Plurality dosn't elect
> the candidate who is favorite to the most voters.

That's the point. You know the voters are burdened and constrained in 
Plurality, because you've seen the elections and the dynamics that 
constrains the voters. But there haven't been all that many Approval 
elections. For all you know, there might have been more 
1800-presidential-election cases.

> What is Approval's distortion?

See above. Approval's distortion happens when people dither imperfectly.

> In a non-u/a election,where it's only gradations of liked-ness, your
> "distortion" has no meaning, because there's no objective
> falsification. The voter likes all of hir approved better than any of
> hir unapproved.

That's stretching it a bit. Consider again a contrived method which is 
like Approval but you're forced to Approve at least half of the 
candidates. Your response could be used against it, too: "distortion has 
no meaning because the voter still likes all the Approved above all the 
not-Approved". True, but the limitation (uncertainty in the case of 
Approval, uncertainty plus external constraint in the case of the 
contrived method) still degrades the result compared to what you could 
have got.

> You continued:
>
>   If everybody's a true game theorist of the sort that would give the
> other guy 1 dollar in the dictator game, then use an iterated method
> because you're only going to get a good result if the method has a good
> game-theoretical equilibrium.
>
> [endquote]
>
> No, Approval, ICT, and maybe a few other methods would do fine with
> strategic voters. Yes, Approval has the defection problem, but I've
> described
> 5 ways of solving it. ICT pretty much does away with the defection
> problem. I suggest that ICT actually delivers on the promise of rank
> balloting.

Would they? I was considering an extreme case, where everybody's 
rational and plays the voting method like a game. ICT handles 
compromising, true, but it's not strategy-proof. You can still bury in 
ICT: rank the worse mainstream candidate below all the "fruits and nuts" 
because you want him to get as few votes as possible in the Plurality 
tiebreaker.

As I'll mention in another post, I think ISmith,DAC/DSC (where ISmith is 
to Smith what ICT is to CT) would do better than ISmith,Plurality (ICT) 
here.

> You continued:
>
> But in between? That matters, and that's where you'll find most
> practical elections.
>
>
> [endquote]
>
> We know that U.S. voters are strategic. Range's full merit won't be
> realized in our electorate. Instead, with only some voters honest,
> they'll be taken advantage of. There's no reason to use the more
> complicated Range instead of Approval.

(Incidentally, I think Range has too obvious a strategy. I'd prefer MJ.)

> You continue:
>
>  From above, the ranked methods are better than Approval because if
> there's a significant fraction of honest voters
>
> [endquote]
>
> Oh yeah? Millions say that they hold their nose and vote for the Democrat.
>
> They'll hold their nose and vote hir alone  at top, in Condorcet,
> because that's the only way to _fully_ help hir against the Republican.

Like they held their noses and voted for Bush instead of Perot in '92? 
Like some smaller fraction voted for Nader in 2000 even in a swing state?

Here's my feedback point. All we need is that the method shows that not 
all of those who compromised needed to.

>>   Further, if favorite betrayal is to pay off, the group doing it have
>> to be large enough that, by voting Compromise over Favorite, they
>> reverse the direction of (Favorite > Compromise). Thus, that seems to be
>> more the kind of strategy that would be directed from party central than
>> something any individual voter would come up with, perhaps with the
>> exception of putting the viable opponent last.
>
> [endquote]
>
> Of course that isn't true. Voters now favorite bury without
> organization. They think that it's their own idea. Yes, they're told to
> by media and other misinformed progressives.
> ...as they will be when Condorcet is the method.

And again. If the Condorcet calculations show that Nader got 2% but Gore 
still won, then the 2% know they can vote for Nader without spoiling it. 
It won't be only 2% next time.

> You continue:
>
>> And this is really where I make my feedback claim. If it takes a lot of
>> people to make strategy work, at least in <= 3-major-candidate or u/a
>> elections, then the overcompromising voters can kick their habit. You
>> seem to consider the voters to be closer to the game theory economist
>> position than I do, in that they would compromise "just to be sure.
>
> [endquote]
>
> It isn't a supposition. They are known to do so.

They are known to do so under a method that more or less forces them to. 
A very limited sample is also known to do so in a few Condorcet polls 
without experience of how Condorcet works.

Yet the Range voters in Warren's Range polls didn't strategize all the 
time - the results were far from identical to the Approval polls he also 
did.

> You continue:
>
>> (and would similarly min/max rank in MJ "just to be sure").
>
> That isn't about FBC.  Some would vote min/max. Some would vote
> sincerely. Result:  strategists take advantge of sincere voters. Not
> good, unless my co-factionalists are the strategists.

See the pattern I mentioned above, about half-empty / half-full. My 
response here would be: *may* take advantage of. The median may still 
fail to change. There's that difference again.

> You continued:
>
>> In short, I say that if people see no change to their overcompromise,
>> then they will slowly stop compromising. In matters of probability,
>> there are always *some* who don't compromise, and it'll spread from there.
>
> [endquote]
>
> I've acknowledged that maybe Condorcet can overcome its FBC problem,
> somewhow, through long use. But Approval doesn't have the problem
> in the first place.

Nope, but it has others, and without FBC failure being as condemning, 
one might reasonably pick the Condorcet basket of advantages over the 
Approval one.

> You continue;
>
>> Finally, I say that I don't dispute that Approval is better than
>> Plurality. I am, however, not certain that the change to Approval is as
>> simple as you say
>
> [endquote]
>
> Ok, how is it more complicated than I said. Pretend that you're a
> Plurality-defender. Let's hear your best argument for claiming that Approval
> wouldn't improve on Plurality.

Please read again: "I say that I don't dispute that Approval is better 
than Plurality".

If I were a Plurality defender, I'd probably aim low. I can't exactly 
use Tammany's "Stalin Frankenstein" line, but I'd claim Approval to be 
un-American (or something much alike it). I'd further claim that 
Plurality has been good enough for us all this time, and Approval is a 
Democratic plot to weaken us all, and that we upset the two-party system 
at our own peril. I'd further add, like FV does, that "Approval doesn't 
even guarantee that the majority winner will win" (given a certain 
definition of the Majority criterion).

But if you mean what my argument is for Approval not being as easy a 
change to get through, I'll take your example of slavery as something 
that was changed on a national scale. True, slavery's a good example of 
where conservatism was wrong (and so might have been a good counter to 
the disingenuous me-as-Plurality-defender above, if that disingenuous 
person would try to get at the people who like reform in the first place 
at all). However, abolition was not done as a single sweeping national 
change. You had your free states and your slave states and there was a 
significant "scuffle" about it (to say the least). In any event, there 
was no national abolition before the free states' abolition. It started 
in smaller areas -- so a state approach to electoral reform has more in 
common with this than has a national approach.

So find the right method -- one that has some chance of withstanding the 
challenges -- and then go from local to national. As a side effect, 
it'll smooth out the FBC problem.




More information about the Election-Methods mailing list