[EM] Hybrid Beats-All/Approval v. Straight Approval

[Bart Ingles]

Rob LeGrand wrote:
> 
> On the other hand, the best Condorcet methods, while imperfect,
> usually make it extremely difficult for a voter to take advantage of
> voting insincerely no matter what information he has.  I'd rather have
> a method that doesn't depend on polls.

Then see my previous post regarding random drawings.  I wonder what
would be the expected utility of such a system?  I would expect it to be
pretty good, since the more popular the candidate, the more "raffle"
entries he should receive.

I'm not sure that I buy the idea that Condorcet is all that immune to
strategy, though.  True, most of the time this strategy is beneficial in
that it tends to weed out low utility Condorcet candidates, but this
also means it can be manipulated.  If you want to sabotage a lukewarm
Condorcet candidate, try to make the polls show a close race between two
other front-runners, so that voters for those candidates will be tempted
to bullet-vote.  Or if you want to help the Condorcet candidate, try to
convince the voters that one of the front-runners is way ahead.

Also, with ranked balloting, voters in a position to use order-reversal
strategies may be relatively rare, but when they exist they would have
greater influence than if truncation were the only insincere strategy
allowed.



  On the other hand,
> Condorcet's
> (and CR pairwise's) requirement of new voting equipment is a huge
> disadvantage; the U.S. government spends far too much taypayers' money
> as it is!  Approval's simplicity can't be beat.
> 
> > Someday, someone with a lot of time on his/her hands should dig
> > through these archives and publish all of the good examples and
> > counter examples in a book like the classic book of examples and
> > counterexamples in topology.
> 
> Good idea; it would make a fun project.  But note that most Condorcet
> methods perform significantly better in my sincere-ballot simulations
> than Approval, which should be seen as a *very* rough approximation of
> CR.  I think we should be very careful drawing conclusions from
> individual contrived examples.

How did your simulations differ from Merrill's?  Were you using position
in policy space to determine voter utilities, etc?

I wonder how Condorcet would have done if some plausible strategies were
tried -- such as truncation for low-utility candidates.  Or you could
assume the voters will try to deduce the likely Condorcet winner based
on position in policy space, and then decide (based on utility) whether
to support or sabotage that Condorcet candidate.

The latter suggests an enhanced strategy for Approval as well, since
there probably is no such thing as a zero-info election if the voter is
able to place the candidates on a policy continuum.  For example, the
voter could then bisect the space between candidates in order to get a
rough idea of how many voters will prefer each candidate and which one
is the likely Condorcet candidate.