unsatisfied with implications of Condorcet method (fwd)

[Rob Lanphier]

Here's my response, with light editing in [] (I caught a thinko in my
original response)
---------- Forwarded message ----------
Subject: Re: unsatisfied with implications of Condorcet method

This is a really good question, and one I'd like to take to the Election
Methods list, if you don't mind me forwarding your mail.  More comments
below...

On Thu, 23 May 2002, xxxxxxxxxxxx wrote:

> I was interested in your interactive demo of Condorcet's method, and I
> decided to try it out with some real world numbers.  I am a member of
> xxxxxxxxxxxxxxxxxxxxxxxxxxxxx, a notoriously factionally divided
> organisation.  One recent election for a single position saw three
> candidates - a moderate left-winger, a fairly "harmless" centre
> candidate, and a more extreme right-winger.  The left-wing candidate had
> the support of a binding faction (the Left - my faction), which means
> that we all decide to vote in a way which will best further our
> political aims - to pull party policy towards socialism.

> Because the right wing candidate was an extremist and had marginalised
> himself, he only had the votes of a small number (5) of hard-core
> supporters.  These all preferred the centre candidate to the left.

> The centre voters (16) all preferred the left to the right, as our
> candidate was more moderate.

> The election was conducted under preferential, STV style voting.  We
> (the Left) therefore decided to preference second to the Centre
> candidate, to ensure that the Right candidate didn't win.  This meant
> that the right wing candidate was eliminated first, and all his votes
> transferred to the Centre candidate, who won.

> If this election had been conducted under the Condorcet method, I
> believe we could have "rigged" the election so the Left candidate won
> instead!  Here's how I represented this election in your demo:
>
> 1,Joan Left
> 2,Sam Middle
> 3,Max Right
> 19:1>2>3
> 16:2>1>3
> 5:3>2>1
>
>
> On these figures, the Centre candidate should clearly win.  He wins
> every pairwise contest.
>
> But if we in the Left decided to preference instead to our right-wing
> enemy, it goes like this:
>
> 19:1>3>2
> 16:2>1>3
> 5:3>2>1
>
> This results in a Condorcet tie-breaker, which is won by Joan Left,
> given that she had the most primary votes.  This means that in order to
> ensure victory, we should vote "against our principles", for the right
> wing extremist!
>
> This works for a fairly wide range of numbers - anywhere between 17 and
> 21 Left voters would make the same decision against 16 Centre and 5
> Right.
>
> And if you don't believe we're that cunning to work out the numbers like
> this beforehand, think again.  We stage manage elections as best we can,
> because the other side are doing the same thing. :)
>
> I'm interested to hear your defense of the Condorcet method in this
> case, given that it encourages people to vote en-bloc against their true
> preferences, in order to ensure victory for their favourite candidate.

The technique you refer to is "burying", and in small elections where a
high degree of private conspiracy can occur, [this is a problem]. The
larger the election, the less likely this is to work.

The risk that Joe Left supporters take is that the Sam Middle supporters
*also* practice order reversal (or truncation).  Suddenly, you may
find yourself in this situation:

19:1>3>2
16:2
5:3>2>1

The winner in this election is Max Right.

The plausible explanation for this is this: word gets out of a conspiracy
among Joan Left supporters to bury Sam Middle.  This outrages Sam Middle
supporters, to the point where they drop support for Joan Left off of
their ballot.  Perhaps even some Joan Left voters become alienated at the
sleezy technique.   It's a very risky strategy in an already close race
(in your first example, Joan Left is only a couple votes away from winning
legitimately)

It's admittedly a strategy that Condorcet is vulnerable to, but not
plausible in a large race.

The thing I'd like to float with the Election Methods list is whether or
not some of the newer methods (e.g. Condorcet SSD) mitigate this problem.

Rob




----
For more information about this list (subscribe, unsubscribe, FAQ, etc), 
please see http://www.eskimo.com/~robla/em