# [EM] Burlington VT reconsidering IRV 10 years after IRV failed to elect the Condorcet Winner

VoteFair electionmethods at votefair.org
Wed Dec 4 22:44:11 PST 2019

```On 12/4/2019 9:09 PM, robert bristow-johnson wrote:
> need to define the "pairwise-winning candidate".

Would this work?

In each round, if there is a not-yet-eliminated candidate who is ranked
higher on more than half the ballots when that candidate is compared to
each and every one of the other not-yet-eliminated candidates on a
pairwise (one-on-one) basis, then this candidate shall be identified as
the pairwise-winning candidate, and this pairwise-winning candidate
shall be elected.

> Even though for a pure Condorcet, equal ranking is allowed, this
> does not work for IRV ...

I concede that this issue of equal ranking is not worth your efforts in
this situation.

However, to participants in this forum I want to clarify that IRV can
use fractional votes.  Such IRV software would need to look through the
ballots after each elimination round to determine how much weight (such
as 0.25 if there are still 4 candidates ranked at the same preference
level) is given to each candidate.  The sum of the votes, including

BTW "fractional" votes does not refer to using fractions, but rather
refers to splitting a three-way equal ranking into decimal equivalents
such as 0.33 for one candidate and 0.33 for a second candidate and 0.33
for a third candidate. (Using actual fractions in software would be a
nightmare.)

I agree that fractional votes in STV (rather than IRV) would be a much
bigger problem -- but even then I believe there are ways to avoid
tossing out a ballot that uses equal rankings.

Yes, such software requires more work, but it only needs to be done once
by one person.

Credibility for this opinion is based on my experience writing IRV
counting code for use at VoteFair.org for comparison with VoteFair
ranking results. Yes, IRV calculations are very messy when they are done
right.  (Using a "look-ahead" approach even ties can be handled nicely.)
Yet IMO laziness is not a valid excuse when developing good software.

Robert, thank you for considering my wording suggestions.

Richard Fobes

On 12/4/2019 9:09 PM, robert bristow-johnson wrote:
>
>
>> On December 4, 2019 11:50 PM VoteFair <electionmethods at votefair.org> wrote:
>>
>>
>>
>> In each round, if there is a not-yet-eliminated candidate who is ranked
>> higher on more than half the ballots when that candidate is compared to
>> each and every one of the other not-yet-eliminated candidates on a
>> pairwise (one-on-one) basis, then that pairwise-winning candidate shall
>> be elected.
>
> need to define the "pairwise-winning candidate".
>
>> I'll again repeat my point that the wordings being proposed do not
>> specify what to do with ballots that rank more than one candidate at the
>> same preference level.
>
> Even though for a pure Condorcet, equal ranking is allowed, this does not work for IRV because, when a single candidate is promoted to a voter's effective (contingency) first choice, IRV would not know which of the equally-ranked candidates to promote.  That was clear in previous implementations of IRV.  I am not sure that this languages needs to be on the ballot where voters consider and approve (or reject) IRV.  But it must be on the Instructions to Voters on each ballot.
>
>> If voting will be done electronically then this
>> issue will only affect the few ballots submitted by mail (such as from
>> people overseas in military service), but otherwise this gap will allow
>> many, many ballots to be discarded as "spoiled" even though they are
>> marked in a reasonable way.
>
> I believe our paper-scan IRV software in 2009 was able to handle gaps in the ranking.  You could rank #1, #2, then #5 and if no other ranks are marked, the #5 rank is effectively #3.  All unranked candidates are tied for last place.
>
>> Again, good luck getting at least some improvements.
>
> Thank you.  We shall see.
>
> Just now I sent the working language I had to the city councilors for consideration.  Dunno if they'll listen.
>
> --
>
> r b-j                  rbj at audioimagination.com
>
> "Imagination is more important than knowledge."
>
```