[EM] An IRVing response

[Buddha Buck]

At 10:23 AM 08-03-2001 -0400, Douglas Greene wrote:
>Message: 2
>   Date: Fri, 03 Aug 2001 04:03:58 -0500
>   From: Randy Kunkee Subject: Re: The Problem(s) with Instant Runoff Voting
Doug:  Could you please forward my reply to Randy?

To reply to just one part of this...

>You have given no evidence to argue that the higher ranked candidate will
>lose unfairly, or that a lower ranked candidate will win unfairly.  It is no
>more difficult than regular elections to turn in counts that can be summed,
>and I have demonstrated that in other postings to this mailing list, and
>will gladly and easily demonstrate it to any who ask.  Briefly, all you have
>to do is turn in counts for each unique set of rankings.  Eg. there will be
>a certain number of L-D-R, L-R-D, R-L-D votes, etc.  So the actual number of
>counts turned in might go up a lot, but are certainly manageable.  Such
>counts could be published and anybody who wants to could apply the
>elimination process themselves to check the results.

There were 13 candidates on the Florida Presidential ballot in 
2000.  Assuming only full rankings, that means that there are 13! 
(6,227,020,800) different unique rankings.  Obviously, since there were 
less than 5 million votes cast in Florida, not all 6 billion different 
unique rankings will need to be tallied.  Perhaps only a few thousand 
different unique rankings will need to be tallied, because of people voting 
identically.

However, I find that I am not swayed by arguments that suggest we can 
simplify counting by tallying individual categories when the theoretical 
number of categories we have to keep track of exceeds the voting population 
by several orders of magnitude.

As for a lower-ranked candidate unfairly winning, Imagine this scenario:

There are three candidates, Larry (a liberal), Melissa (a moderate), and 
Chris (a conservative).  Because of the perceived one-dimensional nature of 
US politics, anyone who prefers Larry to Melissa is going to prefer Melissa 
to Chris, and vice versa.  However, some people who prefer Melissa to 
either of the others will prefer Larry to Chris, and vice versa.  That 
means there are four basic voting blocks:

Larry>Melissa>Chris  (The LMC block)
Chris>Melissa>Larry (The CML block)
Melissa>Chris>Larry (The MCL block)
Melissa>Larry>Chris (The MLC block).

The electorate is fairly polarized, so about 70% of the electorate prefers 
Larry or Chris over Melissa, and are about evenly split.  The voting blocks 
work out to be:

35% Larry>Melissa>Chris
35% Chris>Melissa>Larry
x% Melissa>Larry>Chris
y% Melissa>Chris>Larry

Looking solely at pairs of candidates, we note that 65% of the  voters 
prefer Melissa to Chris, and so if Larry wasn't involved, Melissa would 
win.  Likewise we note that 65% of the voters prefer Melissa to Larry, and 
so if Chris wasn't involved, Melissa would again win.  Both victories would 
be touted as "landslides".  Melissa is a better, more supported choice than 
-either- of the candidates, when the entire set of preferences are taken 
into account.

However, because Melissa only got 30% of the first-place votes, Melissa 
gets eliminated, and one of the other two wins.  This is going to upset 65% 
of the voters, who though Melissa was a better choice.  Only the original 
35% who supported the winner are happy with the choice.

This is a concrete example of this problem with IRV.


>I consider IRVs similarities to the current plurality system an advantage --
>in the people understand it and will accept it.  It also does not have to be
>the final solution, but will be an improvement over what we already have.
>
>Randy Kunkee
>The Instant Runoff Project
>www.instantrunoff.org