[EM] No evidence that IRV doesn't fail. Reasons why it must.

Paul Kislanko kislanko at airmail.net
Fri Jan 23 12:45:02 PST 2004

??? You said the same thing I did, but didn't recognize that's what you did.

>>Eric wrote:
>>>Consider the case of a polarizing issue, such as Abortion. To
>>>those on either side, their last place vote will matter just as
>>>much as their first place vote. Even their middle preferences
>>  >will matter greatly as it puts a buffer between the viewpoint
>>>they agree with and the viewpoint the simply hate.
>>There are a lot of implicit assumptions here about which voting method is
>>being used. You are assuming that fully ranked ballots are being used
>I have made no such assumption.

You make the same assumptions in this reply below.

>Pretend that Candidate A is pro-abortion. Candidate N is
>anti-abortion. The voter could easily rank:
>   A > B = ... = M > N
>If the voter sees no relative difference of candidates B - M, there
>is no need to fully rank them.

Exactly. So there's no reason to assume N is more "disliked" than M. In your
original post you didn't allow "=", but your example of a polarizing issue
would suggest
   A=B=C...>... X=Y=Z
and in your original post you suggested that that Z could be considered
"most disliked" in a ranked-ballot method, but with equal rankings allowed
your original suggestion is untenable.

>From your original post it was not possible to determine that your implicit
assumptions allowed equal rankings. When your reply used equal rankings in
an example, you actually pointed out the flaw in your original post. You not
only had implicit assumptions in your original post, to defend it you
CHANGED your assumptions to include a different method that allows equal

In no case did you address the questions raised about the original post.

You asserted that we could infer something from "Z" being ranked last on a
ballot, and your last reply said the same thing I did, you can't do that if
you require voters to rank all candidates unless you allow equal rankings.
So, as I said originally, you can't make the inferences you did unless you
assume a certain election method. Now you explicitly say that you assume one
that supports equal rankings on ranked ballots.

At least now one of the assumptions is explicit.

I don't disagree at all. I just think we need to be more precise about what
the assumptions are about what method is being used when we give examples
that include "results."
>Still, all preferences matter.
>>assuming that reading the ballots backwards implies a "dislike" function
>>same way that reading it forward implies a "like" function.
>This is simply what ranked ballots are. The alternatives are ranked
>by a voter from most to least liked.

See above. There is nothing you can "assume" about the voters' intent when
you require the voter to rank all alternatives. Once the voter is through
the ones he knows or cares about, the rest is mostly noise, except for the
single-issue voters who explicitly vote someone last because they know the
method will punish them for having last-place votes. I doubt very seriously
many single-issue voters are that smart, but they do follow instructions, so
any method that uses last-place-rankings for anything can be exploited by
the "noisy minority".

>>That would only be the case when every voter ranks all candidates, and
>>voter has been told that who they rank last matters.
>A full ranking is implied even when the voter truncates. For example,
>   A > B > C > D = ... = N
>Is the equivalent to:
>   A > B > C

Yes. But that wasn't what you were talking about when I replied.
Specifically, you said you infer something about  the relationship between
"D" and "N" from "A>B>C>D>N", and I merely noted that you can't UNLESS you
allow either the C>D=... or C>all others form.

The original point was you can't infer anything about the bottom of the
rankings unless you allow truncated ballots or equal rankings. To support
your position you used truncated ballots and equal rankings. So I guess we

>and we still have the function you describe.
>>No voter I know wants to go to that much trouble.
>There is no need to.
>>"Which do you like the most" gets to be a harder question after two or
>which is why one allows a voter to rank alternatives as being equivalent.

This view was not obvious in the post that started this, when someone
suggested that we could infer "most disliked" by whoever was ranked last on
ballot that was strictly monotonically-descending in preference.

And getting back to THAT point - now that we allow truncated ballots and
equal rankings, it is STILL true that you cannot infer from my ballot which
of the equally-ranked or not-listed candidates I like least, or most-want to
not be elected.

More information about the Election-Methods mailing list