[EM] ironclad pro-Condorcet argument?

[Adam Tarr]

Steve Eppley wrote:

>Adam T wrote:
> > Steve Eppley wrote:
>-snip-
> >> Right, we're defining Condorcet as a family of voting
> >> procedures that accept preference orders from the voters
> >> and elect the Condorcet winner, if there is one, given
> >> those votes.
> >
> > So... really, this is Condorcet.  "Condorcet" just means
> > a voting method that reliably satisfies the Condorcet
> > criterion.  Holding pairwise eliminations (in any order)
> > until only one candidate remains does satisfy Condorcet's
> > criteria.
>
>To be clear, that's not how I used the term.  I'm
>distinguishing between the Condorcet criterion
>and Condorcet methods.  As I'm using the term--
>and others too I think, which is why I'm using it
>this way--Condorcet methods are methods that both:
>
>    1) use as input the expressions of preference
>       orders submitted by the voters (and ignore
>       any additional information from the voters)
>
>    2) satisfy the Condorcet criterion (given
>       sincere votes)
>
>But if people here want to drop #1 from the definition,
>I don't mind.

I don't think it's necessary to drop #1.  Consider this method:

Randomly select pairs, use the preference orders to compare them pairwise, 
and eliminate one candidate.  Repeat until only one candidate remains.

That satisfies #1 and #2, and is consistent with what I wrote before.

>I don't know what Adam meant by "reliably."  Condorcet
>winners can lose even if the voting method is Condorcet-
>consistent, if some voters strategize and other voters
>don't counter the strategy.  The ease or difficulty of
>coordinating counter-strategies depends on the method
>and can be used to distinguish between the methods.
>(In other words, do they satisfy the non-drastic
>defense criterion?  The minimal defense criterion?
>The sincere defense criterion?)

I was using "reliably" in the way the academics tend to talk about methods 
- that is, we assume nothing aside from the preference ballots.  If the 
voted ballots contain a Condorcet winner, the method will find it.

Of course, that method (like any other) is vulnerable to strategy, although 
the random element makes it a bit more resistant to certain strategies, at 
the cost of being erratic.

> > Actually, this is a nice, simple way to describe Condorcet
> > methods.  Tell them it's like a single-elimination
> > tournament where candidates pair off against one another
> > and get eliminated.  That's not an accurate description
> > of Ranked Pairs or SSD/Beatpath, but it is an accurate
> > description of Condorcet.
>
>I expect there'll be disagreement over Adam's use
>of the adjective "accurate."  I agree with him that
>the example method, single-elimination pairings,
>is Condorcet-consistent and a useful analogy.

I mean that such a method complies with the Condorcet criteria, and as such 
is as much of a "Condorcet method" as Ranked Pairs, cSSD/Beatpath, or any 
of the others.  It is a weaker method in my opinion (primarily due to the 
lack of clone-independence) but it is still a Condorcet method.

-Adam