[EM] Strategic vs Dishonest

[Kristofer Munsterhjelm]

On 20.09.2021 20:55, Forest Simmons wrote:
> I know this is picky semantics to some people, but to me strategic
> voting does not imply dishonest voting. Is Approval voting dishonest? It
> certainly requires some kind of strategy. Suppose that your sincere
> preference is A>B but optimal strategy leads you to approve B but not
> A.... does following that strategy make you into a dishonest person? I
> would call it insincere strategy but wouldn't go so far as to call it
> dishonest. Does voting for a lesser evil make you dishonest? Are poker
> players who bluff dishonest? Not in my opinion because bluffing is a
> premise of the game.
> 
> On the other hand false witness ... smearing or telling lies about a
> candidate? Or false reporting of actual or fake polls? Or unilateral
> defection from an agreement? To me that crosses the line!

I would say that there are three levels of instrumental/sophisticated
voting.

The first level is sincere/honest: it's voting in a way that answers
what the ballot asks (e.g. what are your scaled utilities, what are your
preference), without giving much thought to how to adapt youre vote.

The second level is adapting your vote to the circumstances, but without
deliberately trying to maximize your chances of winning. Optimal
Approval voting goes here.

And the third is preference reversal, burial, chicken, etc.: defecting
in the prisoner's dilemma.

I guess that most voters would prefer the first type, and that in
civilized society, feel at least some desire to not go to level three.
If the method is sufficiently awful (e.g. Plurality), they may feel like
they have no choice, but I suspect that advanced methods can get quite a
way towards practical strategy immunity by requiring coordination for
strategy to pay off. Keeping honest voters honest plays to the natural
desire to stay at level one (or two) rather than going to level three.
And I think something like Borda fails to put up such a threshold, while
(at least some) Condorcet methods pose enough of a threshold that voters
don't feel inclined to wander off to level three on their own. (Compare
the discontinuous behavior of minmax compared to the continuous behavior
of Borda scores.)

Of course, the natural inclination towards honesty would differ based on
the society in question. Some societies are more dog-eat-dog and voters
would feel less of a problem with squeezing the most power out of their
ballots at any expense.

As for Approval, the problem of it and (to a lesser extent) cardinal
methods that try to escape IIA is that they get rid of level three by
mushing together levels one and two. A voter who just wants to cast a
honest ballot and then go home is forced to contemplate where to put the
approval cutoff based on information that may or may not be current.

As Robert Bristow-Johnson likes to ask: "can you tell me whether to
approve of my second choice or not, without engaging in strategy?". A
honest voter who just wants to vote can't "just vote", because that
decision is inherently a matter of strategy.

The trade-off of making level three ineffective may not be worth it in a
society where the vast majority prefers level one anyway. Because a
voter who, after much consideration, settles on approving only of Best,
only to later find out that Worst won because not enough voters approved
of Good, is going to be pretty annoyed. Meanwhile in Condorcet country,
you need a significant number of level-three voters to get the same outcome.

> So one set of strategic ballots to decide who the finalists will be, and
> another to choose between them ... no dishonesty in my book, even if the
> first set of ballots elicits insincere strategy as expected. If you took
> an oath promising to use only sincere strategy, that might be a
> different story,  unless the oath was under extreme duress!
> 
> So let's not be too quick to use the judgmental term "dishonest" when a
> softer term "insincere" will do.
> 
> Just my opinion about shades of traditional English usage of these terms!

That's a good point. Calling things sincere/dishonest/whatever can be
useful in making it uncomfortable to go to level three. But the names we
give the levels don't change the behavior of the methods, nor to what
degree they incentivize or force voters to use one level or another when
voting.