[EM] Re: equal rankings IRV

[Bart Ingles]

Kevin Venzke wrote:
> 
> Bart,
> 
> --- Bart Ingles <bartman at netgate.net> a écrit : >
> > > The simplest FBC failure scenario I can think of looks like this:
> > >
> > > 6 A
> > > 2 B
> > > 2 C=B  <<< (sincere is C>B)
> > > 3 C>B
>
> [...]
> > This is why consider approval voting a better system than
> > ER-IRV(whole).
> 
> I don't disagree with that.  I'm only arguing that ER-IRV(whole) is much better
> than FPP, IRV, ER-IRV(fractional), runoffs, etc., in avoiding the Duverger scenario.

We seem to agree that ER-IRV(whole) is better than fractional but worse
than plain Approval, but apparently differ on where we place it within
that range.  I'm inclined to say midway between, if the candidate fields
were relatively uninfluenced by the voting method.  But I'm thinking it
might be closer to plain IRV in partisan races, where nomination
strategy is more likely.  It might be better if you also did away with
elimination.

If I understood the optimal zero-info strategy I might change my view
one way or the other.  But for full-info strategy, I can't think of a
good reason to rank more than one candidate first. 
 

> > On the other hand, the fact that they voted B=C might indicate that they
> > were inclined to vote strategically anyway.  It's clear that these two
> > voters would have been better off voting B>C, so maybe they would have
> > voted that way.
> 
> Maybe, but in my view that would involve insincerity whereas compression
> wouldn't.

That's my point-- with this method, the optimal strategy is insincere
(if polls are available), and with approval voting, the optimal strategy
is to use compression.


> > > I think the voters risking the election are those who DON'T use approval
> > > strategy.
> >
> > I don't think that's a valid generalization.  The statement is true
> > enough for the more extreme voters, who voted C>B, but in this case the
> > voters who DID use approval strategy were the ones who blew the
> > election.
> 
> I don't see how one can say that it was the C=B voters and not the C>B
> voters who blew the election.  The C>B vote, on its face, is worse strategy
> than the C=B vote.  The Favorite=LesserEvil vote is not capable of blowing the
> election unless some voters hide their support for the LesserEvil.

You could say they both blew the election.  I'm assuming there was some
reason that the C=B and C>B voters voted differently, such as the C>B
voters being so extreme that they couldn't bring themselves to
compromise.  Or they sincerely believed that C had a better shot at
winning.

On the other hand, the C=B voters compromised, but since they didn't use
optimal strategy, they compromised for nothing.


> >    Note that the C>B voters were ranking sincerely, while the
> > B=C voters were apparently attempting to use strategy (they just didn't
> > use a strong enough strategy).
> 
> So if you voted sincerely, you're not responsible for blowing the election??

Maybe, from a utilitarian standpoint.  If they had a strong preference
for C>B, and a weak one for B>A, they may not have had much incentive
for voting C=B.  


> > Another thought-- has anyone calculated the best zero-info strategy
> > under ER-IRV(whole)?  Is it to rank sincerely, use approval strategy, or
> > something in-between?
> 
> I imagine it is mostly approval strategy, but perhaps some very low utility
> compromises could go in second place.
>
> I'm fairly sure it could be shown that if you strictly rank, the risk of causing
> your compromises to be eliminated would be much more expensive than whatever utility
> cost you pay by not differentiating among your "approved" candidates.

It would be interesting to find out for sure.  Even if just for the
three-candidate, zero info case.


> > The problem is that the C>B>A voters mistakenly believed that ranking B
> > second was enough.  This was a kind of a "sucker bet", much like giving
> > a partial rating under Cardinal Ratings.  Then again, the C=B>A voters
> > were also commiting a "sucker bet".  It's not that they were wrong-- you
> > can be wrong in approval voting too-- it's that they were'nt maximizing
> > their chances.
> 
> This "sucker bet" idea is why I think you should not strictly rank in zero-info.
> Although I don't see how C=B is also a sucker bet.  Is it always a sucker
> bet to use compression strategy?

Probably not for zero info, but I can't think of a good reason for it
under full-info.  But it seems reasonable that even with zero info,
bullet voting would be preferred in more situations than it would be
under Approval.


> Well, I thought the simulations would assume that everyone used a certain strategy.
> When everyone puts a lesser evil in first (even if tied), there is no FBC problem
> as far as I can see.

If you have incentive to rank a lesser-evil first (above your favorite),
that is by definition an FBC failure.  The "even if tied" case would
violate strong FBC, but that criterion seems too strict to worry about.


> [...]  In ER-IRV(whole), the party has to ask for less than
> it would in Approval: It needs all its supporters to either equal-rank B=C or to
> rank B>C or C>B.  From this perspective, it's the 2 B voters who have wrecked the
> result.
> In Approval, the party needs everyone to vote B=C.  Probably easier to ask C
> to bow out, don't you think?

With the above example, maybe, but only because there are no swing or
A=B voters.  Although your point might tend to balance the argument.  I
wouldn't be totally shocked to find that it's close a wash in the
zero-info case.  But full-info seems to favor the same strategy as plain
IRV.  The real-world scenarios are likely in-between.

Bart