[EM] Approval, polling and strategy

[LAYTON Craig]

This is my third (and final) installment.  I'm sorry for taking up so much
space in your in-boxes, but I don't intend to repeat these arguments often,
and I wanted to get them all together in one go.

Approval, more than any other voting system (except probably Cardinal
Ratings), relies on strategy based on polling results.  As I pointed out in
my previous post, people will generally vote sincerely in Ranked Pairs (it
is easier to talk about a particular method than going on about Condorcet
completion methods generally, especially as a few of them aren't that good).
The advantage is that there is only one way to vote sincerely.  As Martin
Harper pointed out, the most common form of insincere voting will be (lazy)
truncation, and this won't generally hurt the voter or effect the election
result, as long as they get as far down as numbering their favourite
compromise candidate.  This becomes less true the more complex the election
gets - say five or ten candidates with a good chance of winning & a
multi-dimensional issue space meaning an exponential number of voting
factions / patterns.

But this type of situation is even worse for Approval, especially if we
consider the kind of roll polling can play in an Approval election.
Firstly, polls are innacurate.  Election results deviate from polling
results significantly, perhaps in part becuase they influence the outcome,
both strategically and otherwise (there is at least one recent Australian
case of polling actually influencing the result of a state election, to the
point where the polls were unbelievably innacurate, predicting a comfortable
win for the incumbent party, but resulting in a landslide victory for the
opposition).

Polls are a sample of what people think.  Like any sample, they are prone to
sampling errors, and perhaps more than other types samples, they are prone
to dynamic changes (by the time the polls results are published, the people
who were polled might well have changed their minds).

Unlike elections, polls are not closely policed.  There have already been
cases of parties using polls to affect people's opinions (from the way the
questions were asked), and it isn't too much of a stretch to imagine parties
fudging results, or more likely, just not trying too hard to iron out
sampling errors that might give them a favourable outcome, or conducting
surveys in partiuclar geographic areas (say) to give particular results.  I
don't want to get too *big party conspiracy* here, there is enough paranoia
about politics as it is, I simply want to examine the 'real politik' of an
approval election.

The reason I think this is important is that polls directly effect results
in Approval.  If the supporters of candidate A can be convinced by polls
that the candidate doesn't have a chance of winning, then the candidate
won't have a chance of winning, becuase A supporters will vote a compromise
candidate, costing candidate A the election if the polls are innacurate (for
whatever reason).  I trust the difficulties with this scenario are obvious,
as are the inherently undemocratic properties of elections decided in this
way.

By comparison, Ranked Pairs has very few of these problems.  It is certainly
still prone to some sorts of polling effectations (I don't think that's a
word), and some risk of insincere strategy, but, on the whole, it decides
elections on a discrete set of sensible propositions - which candidate is
preferred to which by the electorate.  If candiate A is preferred to every
other candidate by the electorate, then candidate A wins (in almost every
case).  The average SU scores are high and consistent, results are
understandable, it is clone independent, monotonic, and generally strives to
elect the most popular candidate in such a way as to give each voter as far
as possible the tools to have the maximum impact on the outcome, and express
all of their preferences at the same time.

If anyone is interested, I can give a criticism of Cardinal Ratings (I
believe the problems are slightly different), but otherwise, I'll leave it
at that.  Thanks for reading (if you got this far).