[EM] Re: IRV-Approval hybrid

[Kevin Venzke]

Chris,

Just a few thoughts for the moment...

 --- Chris Benham <chrisbenham at bigpond.com> a écrit : 
> Kevin,
> I  have a couple of new ideas concerning  "Approval Elimination Runoff". 
>  The first concerns how to deal with multiple majorities.
> I now think that if  at any stage there are rival majorities, then there 
> should  a runoff between the biggest majority candidate and the most 
> approved majority candidate.  

In other words, look at the pairwise contest.  That would work.  My thoughts
would just be that 1) it's more complicated, and I'm a minimalist.  2) Doesn't
it beg the question of why we don't just look for a CW?  The appeal of AER,
and the like, in my view, is that we dodge pairwise contests altogether.

> My intuition tends to the view that 
>  compliance with Participation can survive one elimination runoff, but 
> not more.

I want to mention this below...

> My other idea is: if we can have IRV "without elimination"  (RWE), why 
> not  AER  also without elimination?  BTW, I  think we need a better word 
> for psuedo- eliminate  than just "eliminate".  I suggest the word 
> "shunt", as in one or some of the candidates is
> "shunted" on to a side-list where they may be processed in a different 
> way from candidates which remain on the "main" list.

I am trying to think why we couldn't have "AERWE," but nothing convincing
occurs to me.  (I'm asking myself why I didn't think of it.)  I really can't
think of why it would be worse than AER, unless perhaps people think the latter
is more intuitive.

"Shunt" sounds good, although I wasn't familiar with that word.

And from your other message:
>3.Those ballots which show as many preference-levels among the remaining 
>candidates as there are remaining candidates (N) are now converted to 
>ballots showing (N-1) preferences by disregarding the smallest gap in 
>the scores of adjacently ranked remaining candidates. If there is a tie 
>for "smallest gap" , then disregard the tied gap between the highest 
>adjacent preferences among the remaining candidates.
>
>4. Inferring rankings from these ballots (amended so that none show more 
>than N-1 preference-levels), elect the CW if there is one.  If not , 
>eliminate non-members of the Smith set, repeat step 3 and so on.

This is an interesting process.  What I wonder is whether we can really get
a better result (in whatever terms) by gradually compressing the ranks than
by going to Approval in one step.

I do think it's reasonable to think we CAN do it in one step, since all the
candidates are Schwartz set members.  It's not as though there are "sure losers" 
to throw off the automated cutoff placement.

>This is currently my favourite "high-resolution CR-ballot" single-winner 
>method. I think it should comply with Participation, and
>probably have fewer and/or less serious strategy problems than plain 
>ranked-ballot Condorcet.

About Participation...  It is so hard to meet, that I think it has to be
weakened it in order to have something both meaningful and realistic.
It seems it's been proven that Condorcet and Participation are incompatible.
I guess that's because it's impossible to break a cycle without making some
people regret their vote.  So overruling defeats, and eliminating candidates,
are both completely out, if we want Participation.  MCA doesn't do either and
it still fails.  What passes, honestly?  Approval, Plurality, and Borda, I
think: the single-step methods.

Even if we use a Participation-meeting method to resolve the Condorcet cycles,
we still won't meet Participation on the whole.  It may still be desirable
to prevent the cycle.

However, I don't really know how to weaken Participation.  If I ask, what
methods are "close enough"?, I don't know how to answer.  Which ones?  Why do
we think so?


Kevin Venzke
stepjak at yahoo.fr


___________________________________________________________
Do You Yahoo!? -- Une adresse @yahoo.fr gratuite et en français !
Yahoo! Mail : http://fr.mail.yahoo.com