[EM] IRV vs Condorcet vs Range/Score

[Raph Frank]

On 10/13/08, Chris Benham <cbenhamau at yahoo.com.au> wrote:
> CB: Not necessarily, no. I have in mind something like (with sincere
> ratings):
>
> 49: A(99)  > C(2)  > B(1)
> 48: B(99)  > C(2)  > A(1)
> 03: C(99) > A(98) > B(1)
>
> C is the CW, but A is a much more stable, much higher SU, arguably as
> fair or fairer winner.

Right, the problem here is that condorcet has no concept of strength
of opinion.  It, in effect, converts your above ratings into something
like:

49: A(99)  > C(50)  > B(1)
48: B(99)  > C(50)  > A(1)
03: C(99)  > A(50) > B(1)

A: 150
B: 101
C: 199

So, it is reasonable to assume that C is better, based purely on the rankings.

Also, in your example, a majority prefer C to anyone else.  In a
pairwise election, he would win against A or B.  This is an example or
a weak preference of a majority overriding a weak preference of a
minority.

I wonder if differential turnout would help.

> 49: A(99)  > C(2)  > B(1)
> 48: B(99)  > C(2)  > A(1)
> 03: C(99) > A(98) > B(1)

If C and A are the only potential winners, then the B group would have
a lower turnout., as they don't care much which of them wins.  If only
3 of them decide not to bother to vote, then A will win.

Similarily, the C group would have a lower turnout.

The combined effect is that A will win.

Ofc, given how close the A/B race is, they will likely turnout anyway,
so the result will be somewhat random.

>
> Perhaps you (Aaron) could answer my direct question about what you
> mean by  "Smith/IRV".

There is a bottom 2 runoff version of IRV.

Instead of eliminating the candidate with the fewest votes, the 2
lowest candidates are checked against each other and the one who loses
the pairwise comparison is eliminated,

A condorcet winner by definition cannot be eliminated, so it is IRV
but it always picks one of the Smith set.

Maybe that is what he meant, it is sometimes called BTR-IRV.

Warren has a page on it.  It apparently fails some important criteria.

http://scorevoting.org/BtrIrv.html