[EM] Comments on some rank methods

MIKE OSSIPOFF nkklrp at hotmail.com
Tue Apr 10 13:20:22 PDT 2012


You wrote:

On 04/09/2012 11:31 PM, MIKE OSSIPOFF wrote:
> I've said seemingly contradictory things about IRV. It's particularly
> flagrant FBC failure makes it entirely inadequate for
> public political elections, more so than Condorcet, which, too, is
> inadequate due to FBC failure.
You keep saying anything without FBC is automatically a no-go. How do 
you know that?


It is a country-specific observation, regarding the electorate of the U.S., where I
reside. I don't know that about any other country, though there is evidence for it in various 
countries where people are used to Plurality voting.

Yes this is "anecdotal", but I've personally observed favorite-burial in Condorcet voting.

A safe and prudent rule is "Never underestimate the voter's inclination for resigned over-compromising
give-away, if there's any chance that it could help a compromise against someone worse."

You continued:

Condorcet elections seem to work well where they're used: 
in the Pirate Parties, in Debian, on Wikimedia, and in smaller 
organizations (see the Wikipedia article on the Schulze method for more).


Of course. I said in my answers for Democracy Chronicles that Condorcet is fine for organizational
elections, where the lesser-of-2-evils problem has nothing resembling the magnitude that it has
in public political elections.

You continued:
IRV's compromise problem is particularly bad. 


Most definitely.

You continued:

In a Burlington scenario, 
the voters (and/or parties) immediately know they didn't compromise well 
enough. In contrast, that doesn't happen in Condorcet, because it picks 
the CW.


Quite so. The FBC problem is _qualitatively_ worse in IRV than in Condorcet. But
it's still present in Condorcet. If I had to choose between Condorcet and IRV,
for our public political elections, with the electorate that we have, of course
I'd choose Condorcet.

But I'd rather have Approval, ABucklin, or ICT, to name a few better methods.

I believe that ICT has tremendous promise. As I'm going to say in a separate posting,
there's a sense in which Approval is the best method, due to its FBC-compliance, its
unique optimizations, and its simple and clear strategy (things that I mentioned in my
Declaration signature).

But of course it depends on what you want, and you might want more expressivity, for a
gradation of preference. I question the value of that, but I myself would use, in an Approval
election, an option allowing that. So I certainly understand that goal. I suggest that ICT
is the best, among the methods attempting that goal, due to its FBC compliance and its
defection-resistance (actually nearly a defection-proofness). ICT deals with defection more
neatly than do the conditional methods (but I still like them due to their usability as options
in Approval). And defection in ICT can only help a candidate if s/he is the top-favorite among
the undefeated candidates.

For a method other than Approval, to improve preference-level expressiveness, I'd choose ICT.

You continued:

I suppose you could argue I'm making the same mistake as the IRVists. 
The IRVists say "oh, we need a way of not making small third parties get 
in the way of the major parties - here's a fix". You might argue I'm 
saying "oh, we need a way of the method not messing up when third 
parties get reasonably large - here's a fix, in the form of Condorcet 
compliance". But if I'm just proposing a fix and the fix isn't 
substantial, then we should be seeing backsliding from Condorcet where 
it's used, and I don't know of any such.

Condorcet is probably fine for most organizations. In that implementation, there won't
be the intense "lesser-of-2-evils inclinations" that is necessary to induce favorite-burial
in Condorcet.

> And I also said that IRV would be a fine method, were it not for the
> public's inclination towards resigned, cowed overcompromise,
> and their very sad and disastrous lowering of standards for
> acceptability. "Vote for the least bad of the corrupt candidates,
> because they're the
> winnable ones".
I disagree. Look at Burlington with its center squeeze again. The 
Burlington voters didn't strategize (much or effectively). Yet instead 
of picking the candidate closest to the median voter, IRV threw that one 
away because it didn't have enough first-place votes and picked the 
largest wing instead.


Never choose IRV as a way to attain the voter median. For that, I recommend ICT
for public political elections. Condorcet(wv), maybe Beatpath, would be ok for organizations.

(But I remind you that it's been pointed out that Approval will result in a candidate configuration
equilibrium with a candidate at voter median, and a candidate there will win at Myerson-Weber 
voting equilibrium.)

But, even in organizational or committee elections, I feel that it's important to use
a method that is public-suitable, in order to provide use-precedent for a good public
voting method.

I don't deny that IRV has center squeeze, or that center squeeze is undesirable.

You continued:

Sure, it's better than Plurality. It's not as polarizing as Plurality, 
either, but it does have some element of polarization in picking the 
largest wing.


Yes, but how bad the effect of that is, depends on the electorate.

You continued:
How about this zero-info: Rank all the acceptables in alphabetical 
order. Rank the unacceptables in random order after the acceptables.

[of course, in that situation, there'd be no need to rank the 
unacceptables, unless the rules require ranking everyone]

Yes, that sounds like a perfectly good way to vote in 0-info u/a 
IRV elections.

You continued:

ranking the acceptables in the same order, you focus the first 
preference votes on a few candidates so that compromise is maximized


I hadn't thought of that, and I like it.

You continued:

; by 
ranking the unacceptables randomly, you discourage strategy that 
attempts to compromise with your pattern (because there is no pattern).


But, as I mentioned above, you could also just not rank them, unless the
rules require ranking everyone.

You continued:

If you have information, then rank the acceptables in order of first 
preference votes


Sure, that's a good rough measure of their ability to take victory away
from an unacceptable.

You continued:

, then the unacceptables in reverse order of first 
preference votes.


...if the rules require ranking everyone.  You'd be preferentially helping the unacceptables
least likely to benefit from that help.

But, by the time your vote gets to the unacceptables, the acceptables must have all been
eliminated. So, if you're going to rank the unacceptables, wouldn't it be best to rank
them in order of preference (in reverse order of unacceptability)?

There's a good case for not bothering to rank them, if the all-important thing is 
keeping an unacceptable from winning. But, by ranking them, you could at least try
to help the least bad of them against the worst of them.

Anyway, IRV's complete defection-proofness and its Mutual Majority Criterion compliance are powerful
advantages, for an electorate that had decent judgement about acceptability and compromise-deservingness.

I'm not saying that I'd choose IRV then, only that it could be worth considering then. 

If voters only care about winning by mutual majority, then they could just all rank sincerely in IRV.

If they want to maximize their expectation, then they'll sometimes favorite-bury, in the strategies that
we've described. But if their judgement is any good, then it won't be the kind of dismal overcompromise
that we're used to. 

But I emphasize that IRV is no good whatsoever, completely inadequate, for the kind of electorate that actually
exists here, for public political elections.
Mike Ossipoff

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20120410/9026109c/attachment-0004.htm>

More information about the Election-Methods mailing list