[EM] Consensus?: IRV vs. Primary w/Runoff

[Forest Simmons]

On Sat, 2 Feb 2002, Markus Schulze wrote:

> Dear Forest,
> 
> you wrote (1 Feb 2002):
> > Markus wrote (1 Feb 2002):
> > > Forest wrote (1 Feb 2002):
> > > > It seems to me that Primary with Runoff is virtually independent from
> > > > clones, too, because it would take at least a two thirds majority to
> > > > get two clones as finalists, unless the voters are asleep.
> > > 
> > > When primary with runoff is used then a party doesn't need a two thirds
> > > majority to manipulate the result of the elections by running clones.
> > > It is only necessary that the clones take so many votes away from the CW
> > > that he falls behind the candidate with the third largest number of first
> > > preferences.
> >
> > How can they do this if they have less than two thirds of voters on their
> > side? If more than one third of the voters don't want two clone candidates
> > to survive to the finals, they can agree to vote for the CW or some other
> > candidate that they prefer and prevent one of the clones from surviving to
> > the finals.
> 
> Let's say that candidate A gets 35%, candidate B gets 30%, candidate C
> gets 25% and candidate D gets 10% of the first preferences. Let's say that
> candidate B is the Condorcet winner. Let's say that candidate A would win
> the runoff against candidate C. Then to make candidate A win the elections,
> it is sufficient that clones of candidate B take so many first preferences
> away from candidate B that he gets less than 25% of the first preferences.

It seems to me that this kind of thing also happens with only three
candidates, so that IRV has the same problem.

It seems to me that election methods can potentially have two kinds of
problems with clones: (1) Some methods tend to give an advantage to
parties that runs lots of clones. (2) Other methods tend to penalize
parties that run lots of clones.

I think that the first kind of clone problem is worse than the second,
since the voters can strategically defend themselves from the latter kind
if like minded voters coordinate their voting.

Both Top Two Runoff and IRV suffer from the latter style clone problem.

> 
> ******
> 
> You wrote (1 Feb 2002):
> > Markus wrote (1 Feb 2002):
> > > Forest wrote (1 Feb 2002):
> > > > It (the instant version of Primary with Runoff) is summable, has
> > > > simpler strategy, is less likely to eliminate the sincere CW, and
> > > > (I believe) is less manipulable.
> > >
> > > It is questionable whether summability is desirable. E.g. Nurmi
> > > and Bartholdi consider summable methods to be more manipulable.
> >
> > They have a strange conception of manipulability.
> 
> Due to Nurmi and Bartholdi, the more information you need about the
> opinions of the voters resp. the more accurate this information must
> be to be able to calculate a strategy, the less vulnerable to strategies
> the used election method is. To my opinion, this argument by Nurmi and
> Bartholdi is plausible.
> 
> ******
> 
> You wrote (1 Feb 2002):
> > Markus wrote (1 Feb 2002):
> > > And your argument that primary with runoff "has simpler strategy"
> > > is rather an argument against primary with runoff.
> >
> > In what way?
> 
> When an election method has "simpler strategy" then, of course, this
> method is more vulnerable to strategies.
> 
> Markus Schulze
> 

Nurmi rates both Top Two Runoff and IRV at level IV (Roman Numeral Four)
in his classification.  Approval is rated at level II, which means he
considers Approval two levels more manipulable than IRV, because Approval
strategy is that much simpler.

But you are right that Nurmi should consider Top Two Runoff as more
manipulable by his criterion. Perhaps he does, but not enough to put it at
level V.

I will post the rest of my reply to this under the new heading of "Markus
and Manipulability" that has come up since I last looked at the EM list. 

Forest