[Election-Methods] RELEASE: Instant Runoff Voting (Chris Benham)

[Abd ul-Rahman Lomax]

At 12:23 PM 6/23/2008, Juho wrote:
Let's assume that our voter has fixed preferences of A=543, B=123,
>C=99. The election method is either Range with values from 0 to 10 or
>Approval.
>
>If the election has two candidates, B and C, the voter might vote in
>Range B=10, C=0 and in Approval B=1, C=0. We can assume that the
>voter will normalize the vote and rank one of the candidates at 10
>(or 1 in Approval) and one at 0.

Right. However, let's do the full normalization for the total 
candidate set, first, otherwise the absolute utilities, which may be 
on the Heaven-Hell scale, could be misleading. Normalizing, we have
A: 10.00
B:  0.54
C:  0.00

In other words, B, compared to A, is almost as bad as C.

>If the election has three candidates the voter might vote in Range
>A=10, B=1, C=0.

Maybe. More likely, because it doesn't make much difference to the voter:

>  Or if A and B are the strongest candidates then maybe
>strategically A=10, B=0, C=0.

>  In Approval the voter might vote A=1,
>B=0, C=0. Or if B and C are the strongest candidates then maybe A=1,
>B=1, C=0.

If it were me, I might be buying tickets out of the country. That is 
*really* bad. *Sincere normalized rating, unmodified by election 
probabilities, is almost zero.*

Voters with utilities like this, if they believe A doesn't have a 
prayer, tend to not vote. If this were Plurality, they would not vote 
for B under just about any circumstances. If it were Optional 
Preferential Voting, they'd truncate. A preference increment of 1/20 
range is quite possibly not reliably determinable. I think that 1/10 
is hard to tell. The reasons for wanting higher resolution range have 
to do with an ability to express preference with less effect 
strategicically. So, given utilities like those above, I might vote A 
100, B 1, C 0. If I could tell a difference.

>The sincere opinions/utilities A=543, B=123, C=99 were valid in all
>the cases but typical voter behaviour in Range and Approval was to
>normalize the vote and maybe to vote strategically depending on who
>the strongest candidates are. The ratings given to the candidates
>varied although the opinions/utilities stayed the same all the time.
>
>This changing behaviour may sometimes lead to one of the candidates
>being a spoiler.

Plurality, if voters vote sincerely, guarantees the spoiler effect. 
It's part of the method. Range allows something different, but nobody 
coerces voters. *Voters* can decide to act in ways that mimic the 
spoiler effect, but it's not intrinsic to the method, and if voters 
vote with any reasonable understanding at all, there is no spoiler 
effect with any significant frequency with Range or Approval.

How many voters in 2000 would not have known that Bush and Gore were 
the frontrunners? The decision in Approval is quite simple: if you 
want to influence the election, vote for one of the frontrunners, 
period. Indeed, it gets tricky when there are three frontrunners, but 
that is vanishingly rare in the U.S.

If it had been Approval, would Nader still have been a spoiler? 
(Assuming he was, there is some possible controversy about that, 
though I do accept the reality of it myself.) Suppose he had not run. 
How would his supporters have voted? The question is, if they 
believed his arguments or voted for him because they already felt 
that way, that there was no important difference between Bush and 
Gore, they might not have voted at all. It's entirely possible that 
the Nader candidacy increased turnout, for people who wanted to vote 
for him. The real question is how many Gore votes disappeared into 
the Nader rabbit-hole.

Probably some. That election, in Florida, which was itself a crisis 
point, exquisitely sensitive, was very close, the official margin 
(never mind the actual votes) was 0.009%. But then we'd have to look 
at other possible spoilers in the other direction. Buchanan. Brown. 
Both of them might have taken Bush votes.

Here is the point: With Range or Approval, voters would not be 
*forced* to make the choice. They still might. Same with IRV (which 
can generally be considered to fix the minor candidate version of the 
spoiler effect, just about its one good feature). If voters are 
convinced by Nader that there is no difference between Bush and Gore, 
they might well truncate with IRV just as well. Indeed, there were 
some arguments from this side that it was better for Bush to win, 
because it made things worse than thus people would be more motivated 
to seek better solutions. (This is a very dangerous argument, and 
never mind that it contradicts the argument that Bush and Gore are 
the same. But I've heard people make this very argument, in person. 
I'll admit it makes me angry: they are responsible, then, for the 
deaths of those who needlessly died because of their desire to make 
things worse. The ends do *not* justify the means. The means *are* 
the ends, that is all we have: what we do. We don't control "ends.")

What if Florida had an early open Presdiential Primary, maybe in 
October, with a runoff on the November election day if needed? It 
would have been needed, in this case since neither major candidate 
gained a majority. And then we would not be guessing about who'd have won.