[EM] Re: "More general 5-candidate IRV failure example"

[Markus Schulze]

Dear Mike Ossipoff,

you wrote (23 Aug 2003):
> Markus said:
> > The fact that the binary methods are all vulnerable to
> > preference misrepresentation of an equal difficulty is
> > not surprising as the winning criterion of these methods
> > is a binary one by definition. Hence all relevant
> > information used by these methods is contained in the
> > pairwise comparison matrix.

I wrote (23 Aug 2003):
> Nope! That's a quotation from Hannu Nurmi and not a
> quotation from me.

You wrote (25 Aug 2003):
> What's the difference? When you copy your authors so loyally,
> is there any significant difference between your positions &
> theirs? I considered it to be understood that a position stated
> by you is a quotation, whether explicit or not.

It is possible to discuss a paper without having to agree fully
with the author. Isn't it?

******

I wrote (23 Aug 2003):
> According to Blake Cretney's terminology, a voter is
> "compromising" when he insincerely ranks a given candidate
> higher to make him win.
>
> Example:
>
>    40 ABC
>    35 BCA
>    25 CAB
>
>    Suppose the MinMax method is being used. Then the winner
>    is candidate A.
>
>    However, when the 35 BCA voters switch to CBA then the
>    MinMax winner is changed to candidate C. In so far as
>    the 35 BCA voters strictly prefer candidate C to candidate A,
>    insincerely voting CBA instead of BCA to change the winner
>    from candidate A to candidate C is a useful strategy for them.
>
> When I say that the MinMax method is vulnerable to compromising
> then this is simply an observation. Also your Nader-Democrat-
> Republican example for FPP is an example where the voter is
> compromising. I don't agree with you calling compromising
> voters "cheating," "shameful" and "dishonst" since there
> can be honourable reasons for a voter to compromise.
>
> Blake Cretney's terminology:
> http://www.condorcet.org/emr/defn.shtml

You wrote (25 Aug 2003):
> It isn't quite clear why you think Blake's terms are relevant
> here.
>
> Nurmi, and therefore, of course, you, consider compromising
> something that a method is "vulnerable to", which implies that
> you the compromiser to be thereby getting away with something
> that he shouldn't get away with, cheating the system somehow.
> Likewise in Plurality (FPP).
>
> IRV is also "vulnerable to compromising." Or, as the rest of us
> would say, IRV makes voters have to bury their favorite in order
> to keep their last choice from winning, as in Plurality.
>
> I doubt that said "shameful". But forget those names. If you have
> even a little honesty, you'll admit that you, copying your authors,
> believe that when someone compromises to protect majorilty rule,
> or to save the win of a CW, that they're attemptinlg something that
> they shouldn't be allowed to get away with. You call it "offensive
> strategy", and speak of vulnerability to it. Now, if you want t back
> down from your position, that's fine. In fact it would make more
> sense if you did. But that's no concern of mine.
>
> Look, you've stated your (Nurmi's) position on that, and I've
> commented on it. There's nothing more to add to that. Let's consider
> that topic concluded. If I don't reply to further unnecessary
> argument by you about it, that doesn't mean that you've said
> something irrefutable. It means only that I don't hjave time to
> reply to you on it forever. You did this on EM too. If you say
> something original and non-ridiculous, I might reply. Otherwise
> not.
>
> As I said, you call compromising, when it involves insincere voting
> that changes the outcome from the sincere result, an offensive
> strategy. When that stategy works, you say the method is "vulnerable"
> to it, with an implication that it would be better if a method weren't
> vulnerable to it. As I said, forget the name-words such as criminal, etc.
>
> You say that the Nader preferrer who votes Democrat to keep the Republican
> from winning is voting offensive strategy. Offensive against whom or what?
> But it gets funnier than that:
>
> Say it's Nader, the Democrat, & the Republican. Say you prefer Nader,
> and vote sincerely. No offensive strategizer you! But the Nader preferrers
> are numerous, and if they voted for the Democrat they could (offensively?)
> make him win instead of the Republican. You know that the Republican would
> win under sincere voting, but that some Nader people are going to vote
> Democrat. So you, a Nader preferrer, vote for the Republican--and you call
> that a defensive strategy :-)

Please read the above quotation from Nurmi again. Nurmi doesn't
say that compromising voters were "criminal," "cheating," "shameful"
or "dishonst." Neither did I say that compromising voters were
"criminal," "cheating," "shameful" or "dishonst."

You wrote: "When that stategy works, you say the method is 'vulnerable'
to it, with an implication that it would be better if a method weren't
vulnerable to it." When I say that compromising is a useful voting
behaviour e.g. under the MinMax method then this is primarily only
an observation. In my opinion, there can be quite honourable and
understandable reasons for a voter to compromise.

******

You wrote (23 Aug 2003):
> Condorcet's criterion doesn't depend on Condorcet's method. The
> Condorcet criterion is widely valued and used by people proposing
> methods other than Condorcet's method. Strategy is defensive if
> it's intended to protect the win of a Condorcet winner.

I wrote (23 Aug 2003):
> Of course, when you define "defensive" in this manner then the
> claim that non-Condorcet methods are lousy methods isn't one
> of your conclusions, it is one of your presumptions.

You wrote (25 Aug 2003):
> You're confusing "Condorcet's method" with "Pairwise-Count method".
> Condorcet's method is one of many pairwise-count methods, but most
> pairwise count methods are not Condorcet's method.

What makes you believe that I confuse "Condorcet's method" with
"Pairwise-Count method"?

******

I wrote (23 Aug 2003):
> A desired conclusion doesn't justify the argumentation.
> For example: In your 3 Aug 2003 mail you criticized IRV for
> violating the participation criterion. Then I mentioned
> that the Condorcet criterion and the participation criterion
> are incompatible and I posted a concrete example where adding
> a set of identical voters who strictly prefer candidate A to
> every other candidate and who strictly prefer every other
> candidate to candidate D changes the winner from candidate A
> to candidate D. Russ Paielli only replied (6 Aug 2003):
> "I can tell you that Mike knows that all Condorcet methods
> violate the Participation Criterion. I don't know why Mike
> even brought it up." So why did you bring it up?

You wrote (25 Aug 2003):
> You point out that Condorcet fails Participation and ask why
> I brought up Participation.
>
> A better question would be "Why shouldn't I?"
>
> At no time did I say that Condorcet passes the Participation
> Criterion. Why did I bring it up? Because it's a desirable
> criterion. That Condorcet fails Participation isn't a good
> thing. But Condorcet gives us important criterion coimpliances,
> such as SFC, GSFC, WDSC, SDSC, and those, in my opinion,
> outweigh its Participation failure. But IRV doesn't offer
> anything to outweigh its participation failure.

However, when you use IRV's violation of the participation
criterion as an argument against IRV and when you then suggest
Condorcet and Approval as possible solutions, then you should
at least mention that also Condorcet violates the participation
criterion. Otherwise, there is the danger that some readers get
the impression that you want them to mistakenly believe that
Condorcet passes the participation criterion.

Markus Schulze