[EM] Tom Ruen talks about bullet voting in Condorcet

[Instant Runoff Voting supporter]

To: "Instant Runoff Voting supporter" <donald at mich.com>
Subject: Re: Withdrawal is merely parts of a Condorcet election.
Date: Sat, 30 Dec 2000

Hi Donald,

Thanks for your thoughts. I've been thinking the same about cycles.

First, about using real names, I've concluded it is overall a bad idea too.
I thought it helped communicate better with scenarios close to real
elections, but unfortunately then I get people denying my argument as
unrealistic because they look at the their expectations of real people and
supporters and can't get past my premise! I will try to not do this any
more.

I also agree my IRV-withdrawal method is of limited use. It can correct for
the spoiler effect from IRV elections, but when pair-wise preference cycles
exist, my method simply picks a different winner than IRV, not necessarily a
better one. If a pair-cycle is verified, then Condorcet is needed to fairly
resolve cycles.

I want to show you my most amazing noncyclic IRV spoiler election: CB=49%,
AB=26%, BA=25%

Candidate C beats both A and B in a 3-way race, but loses to both A and B in
separately pair-races! Condorcet can correctly eliminate C first and allow A
and B to compete directly for the win. However in IRV, B is eliminated first
(Despite B having deeper support than A), and A wins a narrow victory
against C.

At least in cases like this, when no cycles exists, my IRV+Withdrawal gives
the same answer B as Condorcet.

My IRV-withdrawal method also has another problem. It may encourage bullet
voting.

See these two elections: (Same as above, but 25%, 26% reversed)

Version 1: CB=49%, BA=26%, AB=25% (Fully ranked)
IRV 1:
Round 1: C=49, B=26, A=25, eliminate A
Round 2: C=49, B=51 (B wins?)
Withdraw C
IRV 2:
Round 3: B=75, A=25
B wins!

Version 2: CB=49%, BA=26%, A=25% (Bullet for A)
IRV 1:
Round 1: C=49, B=26, A=25, eliminate A
Round 2: C=49, B=26 (C wins?)
Withdraw B
IRV 2:
Round 1: C=49, A=51
A wins!

This example shows in my method, ranking B second cost A the election in
version 1! This means bullet voting in my method is a good strategy for
voters. That's bad since it means voters may neglect their lower preferences
to aid their favorite! We would rather have all preferences available,
however slight,
for the best results.

What about Condorcet?
Version 1: CB=49%, BA=26%, AB=25% (Fully ranked)
Condorcet pairs: A:51>C:49, B:75>A:25, B:51>C:49
B beats both A and C parwise.
B wins!

Version 2: CB=49%, BA=26%, A=25% (Bullet for A)
Condorcet pairs: A:51>C:49, B:75>A:25, C:49>B:26
Cycle - eliminate A for being weakest in first ranks.
B wins!

Well, at least in this example, Condorcet didn't give the election to A
supporters because they bullet voted, although I don't yet know in general
if Condorcet resists offering benefits to bullet voting.

Can you find any cases where bullet voting changes a candidate from a loser
to a winner in a Condorcet method election? Can you disprove this
possibility?

 Thanks for listening!

 Tom Ruen