Inventing preferences causes strategy dilemmas

[Mike Ossipoff]

Steve Eppley writes:
> 
> Mike O wrote:
> >Steve E wrote:
> >> Here's an illustrative example, a 3-way race between Good, Evil1, and 
> >> Evil2.  One voter's ballot: {Good > Evil1 = Evil2}
> >> I chose the candidate's names from this one voter's perspective.  
> >> Other voters might see things differently, producing a close election 
> >> and maybe a circular tie.  Not counting ballots like this against both 
> >> Evil1 and Evil2 in their pairing might throw the election from Good 
> >> to one of the evils.  
> >
> >Good. I fully admit that different results are gotten when we
> >count un-expressed preferences. If you voted only for Good, and
> >no one else, your truncated ballot could create a circular tie,
> >even though Evil1 was Condorcet winner, and would have won
> >had you voted it 2nd.
> 
> By Condorcet winner here, I assume you mean the beats-all-pairwise
> candidate.  I'm not sure why this is relevant, since whether or not

When a Condorcet loses to the truncator's candidate, unless that
Condorcet winner is protected by drastic defensive stratgegy, I
call that relevant. I consider it important to avoid the need to
do that.

> Evil1 beats-all is independent of x.  I think what's relevant is
> whether the effect of x on who wins circular ties is enough to cause
> voters to change their votes (vote defensively).  So let's go on to

It's not so much the question of whether they will, but rather whether
they'll sometimes be sorry they didn't. I don't want the truncator's
candidate to steal victory from the Condorcet winner because those
prefering the Condorcet winner to him didn't use drastic defensive
strategy.


> your next paragraph.
> 
> >If we count you as not voting any preference that you didn't
> >_willingly_ vote, then Evil1 can't have a majority against it. If
> >we invent a preference vote by you for Evil2 against Evil1, then
> >you (unwillingly) can be making Evil1 be beaten with a majority
> >ranking Evil2 over it. 
> 
> There's nothing in the definition of Condorcet that explicitly cares
> about a majorities.  Is this relevant?  (Is it related to an emergent
> property of Condorcet's method?)

Correct. It's relevant to a property of Condorcet's method. A Condorcet
winner can never be made to have a majority against it due to truncation.
The truncators can't do anything about the majority against their
candidate, and therefore can gain nothing by truncation if those
prefering the Condorcet winner to the truncators' candidate don't
defect from voting the Condorcet winner over that other candidate.

That isn't a lot to ask, because no method can help people who need
the compromise but who refuse to rank the compromise.

> 
> Counting the Evil1=Evil2 vote against both Evil1 and Evil2 doesn't
> change the winner of the Evil1 vs Evil2 pairing, but if Evil2 beat
> Evil1 pairwise it would increase the size of this Evil1 defeat.  If
> this defeat is Evil1's worst defeat then counting the vote this way
> could affect the result of the circular tie breaker.

Most definitely. It would affect in a way that is adverse, from the
standpoint of getting rid of need for drastic defensive strategy.

> 
> >Result: As you said, Good might win. That's bad. It means that your
> >truncation has stolen the election from Condorcet winner Evil1. It
> >means that I would have to quit saying that no defensive strategy is
> >needed unless order-reversal is attempted.
> 
> To help me think about this more clearly, I'm going to substitute
> Dole=Good, Clinton=Evil1, and Nader=Evil2.  Let's try the old example:
>   46 Dole
>   20 Clinton
>   34 Nader>Clinton>Dole
>                              x=0           x=.5            x=1
>   Dole loses to Clinton    46 to 54      46 to 54        46 to 54
>   Clinton loses to Nader   20 to 34      43 to 57        66 to 80
>   Nader loses to Dole      34 to 46      44 to 56        54 to 66
> 
> In this example Clinton wins with x=0.  Dole wins with x=.5 or x=1.
> 

Correct. You've just shown an example where truncation steals the
election from a Condorcet winner with x=.5, something that couldn't
happen with Condorcet's method, as I define it. 

I emphasize that any value of x other than 0 can't be called Condorcet's
method if Condorcet didn't say anything about inventing un-voted preferences.
And I believe that he didn't say anything about that.

> >The possibility of mere truncation would be enough to make it
> >necessary for Evil2 voters to rank Evil 1 equal to their favorite,
> >insincerely. Drastic defensive strategy needed even without anyone
> >using order-reversal.
> 
> This decision by Evil2 voters wouldn't hinge on having implausibly
> precise knowledge of the expected votes, would it?  In this case, an
> expectation of massive truncation by the Dole voters?  


Quite right: Defensive strategy _does_ depend on having good information.
That's why I say that it brings with it "strategy dilemma". The voters
in a position where they might need defensive stragtegy to prevent
the worse-evil from winning won't know whether or not to use the
drastic defensive strategy that could give the election away to that
lesser-evil.

So to answer your question: Do the voters using defensive strategy
need good information in order to know what to do? Yes they do. And
they likely won't have it. That's the problem.

> 
> In this example, the tally using x=.5 is very close.  The worst
> defeats for each candidate are within a few votes (or percentage
> points) of each other.  In the past, you've said it's okay to ignore
> scenarios where implausibly precise knowledge is needed to affect
> the voting.  This isn't one of those scenarios?  

Certainly not. I've said that when _offensive strategy_ is risky
and requires better polling information than that possessed by the
intended victims, then that offensive strategy isn't really a problem,
since it's well-deterred in Condorcet's method.

But the fact that, with your proposal (x=.5) voters wouldn't know
how to vote, whether or not they should use defensive strategy--
that isn't a good thing. That's a strategy dilemma. Avoiding that
is our whole purpose in single-winner reform (The goal can be
expressed in various ways, & that's one of them. We all agree with
that goal).

***

Defensive & offensive strategy are different, and shouldn't be
lumped together into a single category of "tactical voting"
or "manipulation". The academics do that because either they're
blissfully unaware of the concerns of voters about the lesser-
of-2-evils problems, or because they're dishonest.

Yes, we want to make offensive strategy difficult & well-deterred.
No, we don't want dilemmas regarding the perhaps-necessary use of
defensive strategy. Creating a situation where voters who might 
need defensive strategy can't know how to vote is not a desirable
accomplishment for a method. It's what we all want to avoid.

> 
> If you were a Nader voter here, you'd vote Nader=Clinton>Dole just
> because the Dole voters *might* truncate en masse?  You'd protect
> Clinton even though Nader has a good shot at winning?  What if a few

The problem is that, as you yourself say, sufficient information might
not be available to know what to do. To answer your question literally,
no I wouldn't vote Clinton = Nader. But guess what: Lots of people would.
Lots of people, in November, are going to insincerely vote Clinton > Nader.

The problem is knowing how good a shot Nader has at winning, and 
, in case he doesn't, how much Clinton needs the Nader voters. I'm
the 1st to admit that Nader voters might not have that knowledge.
As I said, that's the problem. That's the problem with Copeland,
and with x=.5


> Dole voters vote Nader>Clinton or a few Clinton voters vote
> Nader>Dole, which is all it would take for Nader to win?  Are 
> you saying these possibilities are less likely than the massive
> truncation?

I'm saying that that detailed information can't be known to the 
Nader voters, and so, with x=.5, they wouldn't know what to do
if there's a good liklihood of Dole truncation. I'm saying
that if x=.5 the lesser-evils problem isn't gotten rid of
as it is in Condorcet's method.

> 
> Perhaps we need a slightly different example, where the Nader support 
> is much less, to illustrate why the Nader voters would be strongly 
> tempted to ranked Clinton=Nader if x>0:
> 
>   48 Dole
>   24 Clinton
>   28 Nader>Clinton>Dole
>                              x=0           x=.5            x=1
>   Dole loses to Clinton    48 to 52      48 to 52        48 to 52
>   Clinton loses to Nader   24 to 28      48 to 52        72 to 76
>   Nader loses to Dole      28 to 48      40 to 60        52 to 72
> 
> If x=.5, here the Nader voters have a stronger incentive to abandon
> their hopes for Nader and insincerely rank Clinton=Nader, if they
> have enough predictive knowledge.  They may foresee a close race
> between Dole and Clinton even though Clinton loses to Nader pairwise.  
> So I guess I'm down to one question: does is take an implausible
> level of information for the Nader voters to resort to drastic
> defensive strategy, if x=.5 for the equally last-ranked pair? 

Often yes. As I said, that's the problem. I don't want a strategy
dilemma where voters can't know whether or not they need defensive
strategy. Where they have to choose between risking victory for the
greater-evil, or risking giving away the election to the lesser-evil.

> 
> --Steve
> .-
> 


--