[EM] Earlier-no-harm

[Kevin Venzke]

(This message has quotes from three people.)

James Green-Armytage said:
> Very much off the cuff: isn't this somewhat related to the favorite betrayal 
> criterion? 

I think it is, but a more general version of it.

Mike Ossipoff said:
> If IRV's Later-No-Harm "advantage" is the kind of benefit that we get from 
> Later-No-Harm, then it's questionable how important that criterion is.

I agree, but I am somewhat interested in what other methods can be defined
which meet Later-no-Harm.  For example, Woodall's DSC method does not use
traveling votes.  It does eliminate candidates, although I'm not able to
say at present whether it does so in a manner analogous to IRV.

Anthony Duff suggested that MinMax meets Later-no-harm.  I don't think
that's correct, though.


 --- Craig Carey <research at ijs.co.nz> a écrit : > 
> At 2003-12-23 23:47 +0100 Tuesday, Kevin Venzke wrote:
> >Thanks to everyone who responded to my last message.
> >
> >It's possible to define (a possibly limited version of) later-no-harm
> >as: "Adding strict preferences among candidates otherwise ranked last,
> >should not hurt the result of the election from the perspective of
> >this ballot."
> >
> >(By this definition, however, Approval passes, since approving an
> >additional candidate doesn't just involve adding strict preferences,
> >but also deleting others.)
> >
> >I don't believe I've ever heard anyone suggest the obvious counterpart,
> >"earlier-no-harm:" "Adding strict preferences among candidates otherwise
> >ranked FIRST, (etc...)."
> >
> >For example, if voting "A=B=C>D>E" gets me one of those first choices,
> >then voting "A>B>C>D>E" should not get D or E elected.
> >
> 
> Clearly Mr Venkze has not ruled out the 2 winner case. If he has one
> unstated restriction then it may take weeks or years before he finally
> managed to get them all into the mailing list and nicely requoted
> in each message.

"Earlier-no-harm" could work with any number of winners, although my
specific example assumed a single winner.  A zero-winner method, for
instance, could not fail Earlier-no-harm.

> 
> Suppose the votes are this (2 winner case)
> 
> Election 1:
>   10000 (D)
>   1 (A=B=C>D>E), D should win and does
> 
> Election 2:
>   10000 (D)
>   1 (A>B>C>D>E) , both D and E lose says Mr Venkze

But you can see that this example is not what I'm talking about, if you
look at the definition I wrote:
"Adding strict preferences among candidates otherwise ranked FIRST, 
should not hurt the result of the election from the perspective of this ballot."

So if the winners of election 1 are D and one of ABC, then in election 2, 
according to "Earlier-no-harm," the winners must not become D and E.

> 
> I assume that there is no need to get "A=B=C" precisely defined since
> unknown disputes over definitions might be bounded by factors 6 or 1/6.

It means that A, B, and C are ranked equal to each other at the top of
the ballot.

> >"Earlier-no-help" could also be defined.  If "A=B=C>D>E" elects D,
> >then "A>B>C>D>E" should not elect A.
> >
> >What do you think?
> 
> It looks like Mr Venkze won't accept one or both of these:
> 
>  * an axiom set from which a method is derived (a strange option given
>     the theme of Mr Venkze's whole message), or

I defined an axiom, not a method.

>  * the principle that the number of winners be correct (i.e. he sometimes
>     expects the number of winners be disagreeing with what was decided on
>     before the public filled in the voting papers, providing a public 
>     suitably exists).

I did that in my example, but not in my definition.

> I have never suspected that there is a good reason to use
>    "Later No Harm"
> instead of
>    "Truncation Resistance".

"Later-no-harm" is more specific.  Perhaps "Later-no-harm" together with
"Later-no-help" are equivalent to "Truncation Resistance."

> "Permuting preferences before" and "permuting preferences after" are a
> better wording of what is desired. 

This would be harder to satisfy than what I defined.  My definition supposes
that the ballot is changed only by turning some "="s into ">"s.

> I assume that if both are imposed
> then the method degenerates into FPTP or something similar.

No, with FPTP you cannot "permute preferences before" the winner without a
risk of changing the winner.  ...Perhaps Coombs satisfies this?

> So there may need to be a redesign of the papers (and presumably
> adding "=" won't produce insensitivity to permuting before). I would
> prefer no redesign (while preferring methods better than the
> second best AV n candidate 1 winner method) and instead allow elections
> where voters could change their mind.

The voter can't change his mind.  (Actually, I thought you didn't believe
in voters.)  Presenting two scenarios side-by-side doesn't mean the voter
is changing his mind.

The point of earlier-no-harm is so that the voter can't protest "I voted
A>B>C and got F, when voting A=B=C would have gotten C elected."

Kevin Venzke
stepjak at yahoo.fr


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