[EM] Strategic Voting and Simulating It

[Raph Frank]

On Thu, Oct 16, 2008 at 2:23 AM, Kevin Venzke <stepjak at yahoo.fr> wrote:
> Each voter's sincere utilities are determined randomly and independently,
> which is problematic because it does not produce realistic scenarios. It
> would be better to combine voters into factions, although it would be no
> easier to reach consensus regarding what simulation is realistic.

I think Warren has some simulations where the voters are spread over
1-2 axes.  (He can comment).  My understanding is that there are lots
of different distributions.

> Two frontrunner candidates are determined essentially at random, and this
> information will be used by the strategic voters. This is problematic
> because usually when a candidate is called a "frontrunner" this means
> there is a perception that this candidate is likely to win, before any
> strategy is used. When the frontrunners are determined randomly, this is
> not realistic unless you hold a very pessimistic view about how a
> candidate becomes a "frontrunner" in real life.

In plurality, it is self reinforcing.  The top-2 are front runners
because they are supported by the party, and thus are likely to win.

Party support makes them likely front runners and that then makes them
actual front runners.

In such a situation, random doesn't seem entirely unrealistic,
electing 2 candidates via honest PR-STV might better simulate the
primary system.

> The strategic voters, in any rank ballot method, will simultaneously use
> favorite betrayal compromise strategy and also burial strategy. There is
> no calculation, or awareness of the specific election rule. The strategic
> voters seem schizophrenic in that they are sufficiently paranoid about
> losing their compromise choice that they will abandon any actually
> preferred candidate, but at the same time they are sufficiently reckless
> that they will rank the worse "frontrunner" dead last even though in
> methods where this can be an effective strategy, it also creates a major
> risk that the winner will be a candidate that nobody likes.

It is a problem that strategy has to be based on assumptions of the
simulator.  However, maybe my suggestion in the other thread would
help.

> No other strategies or information sources are simulated. Equality of
> preference and truncation are not implemented, so that many popular
> methods cannot even be tested without ignoring their capabilities.
> Nomination (dis)incentives can't be examined either.

True.  It might be possible to test nomination strategy directly.  For
example, the 2nd place candidate might be cloned with a slight offset
in utilities and the election ran again.