[EM] Voter strategising ability

[Kristofer Munsterhjelm]

On 07/19/2014 04:36 PM, robert bristow-johnson wrote:
> On 7/19/14 4:04 AM, Juho Laatu wrote:
>> On 12 Jul 2014, at 22:34, Gervase Lam<gervase at madasafish.com>  wrote:
>>
>>> Given the above, I really find it hard to see a good proportion of
>>> voters doing the correct strategic calculations.
>> Yes, _in_typical_public_elections_ voters can reliably implement only
>> some very simple strategies.
>>
>
> really, the only "strategy" (if i were to call it such) that voters ever
> employ in public elections (which use, in most cases, single mark
> ballots) is that of "compromising".  sometimes voters vote for their
> second or third choice, instead of voting for their first choice,
> because they expect that their second choice has a much better chance of
> winning than their first choice.
>
> the whole purpose of the ranked ballot was to remove that burden of
> tactical voting from voters.  and the whole problem with IRV or STV or
> Alternative Vote or Preferential Vote or whatever they're calling it, is
> that the tabulation procedure sometimes resolves the election
> incorrectly (as it did in Burlington Vermont in 2009).  in my opinion,
> if there is a Condorcet Winner and the election is resolved to elect a
> different winner, that is decidedly a mistake since a winner was chosen
> when a greater number of voters actually marked their ballot that they
> preferred someone else to be elected.

I think this is the core of the issue. Whether or not (ordinary) 
Condorcet is good enough for you depends on what strategies you think 
the voters will employ, and to what extent.

If they're mainly compromising (which means voting for the lesser evil), 
then Condorcet will protect you, because Condorcet is pretty resistant 
to compromising. See Green-Armytage's strategy paper for more about that.

On the other hand, if you think they'll also bury (push likely 
competitors down in the ranks), then ordinary Condorcet doesn't help you 
because it is susceptible to burial (again, JGA's paper goes into this 
in greater detail). If you're really worried about burial, you might not 
even accept anything less than FBC (weak or strong, depending on *how* 
worried you are).

Discussion about which kind of strategy is most likely to happen can go 
on forever without data. Even if there is data, it is quite easy and/or 
tempting to explain it away as not being representative of what would 
happen under an ordinary election. As long as that's possible, it's 
really hard to convince someone who is worried about burial not to be, 
or vice versa.

(There's also the related discussion of how widespread strategy can be 
and how complex the coordinated strategy can be. Again, it's really easy 
to go nowhere.)

Green-Armytage has found Condorcet methods that resist both burial and 
compromising; however, you have to compromise (heh) Condorcet's 
criterion compliances in order to get the empirical dual resistance. 
Some people would like the method to be "forever good", even once the 
political landscape becomes more nuanced due to the protection the 
method initially provides, and therefore think that the loss of 
criterion compliances is unfortunate. I think I'm like that, at least a 
bit. Condorcet-IRV or (possibly) first-preference Copeland might resist 
both burial and compromising because it in essence frustrates strategy 
once the strategizer gets into the IRV regime, while preserving 
Condorcet's benefits when there is a CW. But what about when the voters 
legitimately get into the IRV regime because there are many real options 
to choose from and a Condorcet cycle forms? If all we then have is IRV, 
that's not very good.

> and Range or Score Voting and also Approval Voting, require voters to
> strategize in the voting booth because they have to decide how much to
> score their second choice (or whether to approve their second choice).
> that is not strategy-free.

In my opinion, the main problem with Approval is that it forces even 
those who want to vote honestly, to have to strategize. In a 
Gore-Nader-Bush scenario, every honest Nader voter has to ask himself 
whether or not he should vote for Gore or for both Gore and Nader. In 
contrast, in Condorcet, he can choose not to participate in the strategy 
game if he values honesty in itself.

Others have disagreed with me. Abd, I think, takes the position that 
what I call strategic Approval essentially comes easier to people than 
honesty, so there's not much of a burden. Yet others say that nobody 
will be so stupid as to give his vote away, and so every voter in every 
method will employ strategy (or will be told to do so by the parties), 
hence there's no benefit to Condorcet in this manner either.

The latter argument is like my argument against Range. I think Range 
strategy is just too easy; anyone can think of it and there's no 
protection against offensive strategy except offensive strategy of one's 
own. The voters will thus feel compelled to execute the strategy lest 
they lose. The second argument above is just that this holds for ranked 
methods as well, so just about everybody will be distorting their 
ballots just as badly as I imagine they would under Range.

Though this does sometimes lead to weird responses from others, like 
that Range is bad because it's too sensitive, yet MJ is bad in the other 
direction because it only pays attention to the median voter. But there 
seems to be a definite balance: you can't have both strategy resistance 
and sensitivity to honest ballots at the same time. The trick lies in 
finding the frontier (the best possible trade-offs) and then picking one 
point according to your idea of what kind of strategy will be employed.

... or just go with any ranking system better than IRV and adjust it 
later. Some variant of the New Zealand approach to voting reform (first 
ask whether there is to be a change, *then* what the change should be) 
might be better in the end.