[EM] Does IRV fail the Smith criterion?

[MIKE OSSIPOFF]

R. Moore wrote:

>I. Instant Runoff Voting -- How Well Does It Work?
>
>Instant Runoff Voting (IRV) works well when only a few rounds of
>elimination take place.

IRV can mess-up when there are only 3 candidates, and one elimination
(followed by someone being elected by possessing a majority of the votes). I 
mean that it can mess up by failing Condorcet's Criterion,
WDSC, SFC, &/or Monotonicity.

WDSC & SFC are defined at http://www.electionmethods.org

Additional voting system information is available at:

http://www.barnsdle.demon.co.uk/vote/sing.html


>The IRV winner
>can only be guaranteed to defeat one other candidate in a head-to-head
>matchup.

That's something that we've been trying to explain to the IRVies,
with little success.

>II. IRV and Aliasing
>
>IRV belongs to a class of election methods that I like to call "aliasing
>methods". An aliasing method is any election method in which candidates are
>eliminated without considering all expressed rankings of that candidate.

I compare that to proceeding with an important project or job by
acting irrevocably after looking at only a small fraction of the
available information. You wouldn't do that in an important project,
and you don't want to do it in an election count.


>I believe aliasing methods are non-monotonic.

The important thing is that IRV is nommonotonic for sure.

>III. Improving IRV: RIRV

When Tom Round of Australia first proposed the candidate withdrawal
option, it occurred to me to make that withdrawal automatic, by 
automatically dropping the last remaining nonwinning candidate, and
then repeating the count. I too called it RIRV, but the 1st "R"
stood for "Repeated". That was probably around '94 or '95. I'd hoped
that RIRV, or some other method repeated in that way, could turn out
to be the best, to make strategy guarantees that Condorcet can't make.

It turned out to not be so. Other than that, RIRV's properties weren't
studied. But of course, since it's being proposed now, it's worth
studying its properties, its criterion compliances. It would be surprising 
it RIRV met SFC, much less GSFC, but it's worth checking out.

By the way, don't expect the IRVies to accept RIRV. They refuse every
mitigation compromise. Apparently their motto is "Only the worst, for you".


>V. Improving IRV: SIRV

SIRV has been proposed a number of times. All that can be said for it
is that it isn't as bad as IRV. Of course it meets Smith, which
means it meets Condorcet's Criterion, Condorcet Loser, Majority Loser,
and Mutual Majority. But lots of methods do that.

Some of us have asked for more from a voting system, which is why
the defensive strategy criteria were written. They're defined at
the electionmethods website referred to above.

Consider the simple 3-candidate example, which is the most convenient
to discuss. Say it's candidates A, B, & C. B is the middle candidate.
If A pairbeats C, then the A voters can safely make a strategic circular
tie (by not ranking B, or by insincerely ranking C over B). Doing
so won't make C the BeatsAll winner. At worst it would make a
strategic circular tie.

But if A pairbeats C, and if A isn't the candidate with fewest 1st choice
votes, then that means that C can't win. That means that the same people
who can safely make a strategic circular tie are also safe in the
IRV tiebreaker. The A voters can't worsen their outcome by making
a strategic circular tie; they can only improve it. That's a big
strategy disadvantage for SIRV. SIRV fails all of the defensive
strategy criteria.

Mike Ossipoff

_________________________________________________________________
Get your FREE download of MSN Explorer at http://explorer.msn.com