[EM] Any unusual/bad/overlooked methods or lotteries? For a simulation

[Kevin Venzke]

Hi Kristofer,

--- En date de : Mar 8.3.11, Kristofer Munsterhjelm <km-elmet at broadpark.no> a écrit :
> > I'm working on another simulation. It is for
> 3-candidate elections and
> > allows these ballot types (if the method also allows
> them):
> > A (bullet vote)
> > A>B>C (strict)
> > A=B>C (tied at the top)
> > A|B>C (middle candidate ranked but "disapproved")
> > 
> > This should be enough to handle most rank or 2- or
> 3-slot methods.
> > 
> > The voters have some intelligence, and polling
> opportunity, so a method like "elect the guy with the most
> last preferences" should do just as well
> > as Plurality.
> > 
> > They can be a little superstitious, especially with
> random methods: e.g.
> > Random Ballot isn't perceived as strategy-free. And
> there can be e.g.
> > burial in IRV from voters who never managed to be
> harmed by it.
> 
> How is this superstition (on the one hand) and intelligence
> (on the other) implemented?

Well, I don't want voters to be superstitious. It's an accident. They are
trying to find the best way to vote. But they may latch onto something
that is not sincere, because they have no concept of that.

> Also do voters who favor one of the candidates have greater
> intelligence than those who favor others?

I'm not sure where this thought comes from, but I think the answer is no.

It is bloc-based, and a bloc votes together. So a large bloc could be
said to be able to "coordinate" better. If there are few blocs (which is
how I currently test it) it may not be very fair in this sense.

> If you had a
> computer that could strategize on behalf of every voter, you
> would in effect have a DSV method, and the DSV method might
> not be all that bad. 

It is sort of like that. It's based on polling. There's no difference
between a poll and the actual election. So, voters have a good idea how
other voters think they should vote, in the sense of how to respond to
those votes. Voters may in effect decide on a "mixed strategy."

It's possible to imagine that at the last minute voters in reality could
decide to switch their polling vote to something else for the actual
election. Could that be realistic? Those voters have to communicate their
plan to each other without anyone else learning of it. My sim assumes
that is not possible.

> However, if one of the
> parties/candidates are better at coordinating their voters
> and executing strategy, they may snatch the victory from the
> "honest" winner. Thus, the worst case in strategy might be
> when all strategize (in mutually limiting scenarios like
> Plurality's lesser evil situation), or when only some do,
> and if you want to check the impact of strategy, you might
> want to check both cases.

I think I've mostly covered this, but let me know if not.

I am very interested in what strategies are used. Four are detected:
Compromise (involving reversal), compression, truncation (in the sense
of bullet-voting), and burial. ("Push-over" strategy is possible but would
be detected as compromise, and I'm not sure how to get around that.)

Occurs to me that there's no "abstain" strategy, which might be 
interesting. A little tricky to add though.

> > Anyway, I'm interested in methods that might pose a
> challenge to my voters (such as perhaps deterministic
> methods that fail majority favorite;
> > I have very few of these), or methods that might
> actually be good...
> 
> Every positional method except Plurality fails majority
> favorite. Range fails it as well, and Approval might if you
> interpret it a certain way.

Yes, all those are implemented, as well as strictly-ranked Borda. As an
example of something odd: Once when playing with a simulation years ago
I came up with the idea of taking the pairwise comparison between the
approval winner and the candidate with the greatest opposition (approval
or pairwise, I don't remember) to him. It sounded nice but it failed
majority favorite.

> You might also test Random Pair and Hay, if that's feasible
> within your simulator. Both are strategyproof. Random Pair
> picks two candidates at random and elects the one who beats
> the other Pairwise. Hay is described here: http://www.spaceandgames.com/?p=8 and there's a (very
> complex, cloneproof?) iterated version at http://www.panix.com/~tehom/essays/hay-extended.html .

Random Pair is on the to-do list, but I don't think Hay is feasible if
it requires a very fine ratings ballot. Can we do Hay with 3-slot ratings
(normalization required)? I'll check.

Another example based on Forest and Jobst's paper: What if the first-
preference winner wins if he gets (say) 2/3rds of the vote, else Random
Ballot? I'm pretty sure this would stump my voters with the number of
polling iterations I currently use.

Thanks for your thoughts.

Kevin