[EM] RCV in SF Mayoral election
Brian Olson
bql at bolson.org
Sun Jun 10 07:46:31 PDT 2018
Ok, a few lines of Python poking the raw data shows I must have some bug in
my Condorcet implementation. Digging into that...
On Sun, Jun 10, 2018 at 10:08 AM, Greg Dennis <greg.dennis at voterchoicema.org
> wrote:
> Brian, how is it possible that those differ? Since all the other
> candidates are eliminated in the final round, shouldn't that necessarily be
> the same as the pairwise contest between those two?
>
>
> On Sun, Jun 10, 2018, 9:46 AM Brian Olson <bql at bolson.org> wrote:
>
>> I processed the latest data (2018-06-09) and posted the results of the SF
>> Mayor election using a few algorithms:
>> https://bolson.org/~bolson/2018/SF_Mayor_20180605.html
>>
>> The Condorcet win is now 97436 to 91740 for Leno over Breed.
>> The IRV final round is still just 94783 to 94393.
>>
>>
>> I'm using my software posted at https://github.com/brianolson/voteutil
>>
>> commands (needs maven installed for compiling Java, and needs Python3):
>>
>> curl -O http://www.sfelections.org/results/20180605/data/
>> 20180609/20180609_ballotimage.txt
>> curl -O http://www.sfelections.org/results/20180605/data/
>> 20180609/20180609_masterlookup.txt
>>
>> (mkdir -p ~/psrc && cd ~/psrc && git clone https://github.com/brianolson/
>> voteutil.git && cd ~/psrc/voteutil/java && mvn package)
>> python3 ~/psrc/voteutil/python/rcvToNameEq.py -m
>> 20180609_masterlookup.txt -b 20180609_ballotimage.txt -o 20180609_%s.nameq
>> java -jar ~/psrc/voteutil/java/target/voteutil-1.0.0.jar --rankings
>> --full-html --explain -i 20180609_Mayor.nameq >/tmp/a.html
>>
>>
>> On Sat, Jun 9, 2018 at 9:33 PM, Greg Dennis <
>> greg.dennis at voterchoicema.org> wrote:
>>
>>> I quickly looked at the vote data and saw that lots of ballots are
>>>> categorized as "Exhausted by Over Votes" and "Under Votes," but there is no
>>>> data indicating exactly how those ballots were marked, so we lack enough
>>>> information to be sure of final results.
>>>
>>>
>>> Actually, all the data you need is available from that page. The "Ballot
>>> Image" file will give you the full cast vote record of every individual
>>> ballot, and the "Master Lookup" is the legend that tells you what each
>>> number means. If you have trouble interpreting the numbers, just ping me!
>>>
>>>
>>> On Sat, Jun 9, 2018 at 4:36 PM, VoteFair <electionmethods at votefair.org>
>>> wrote:
>>>
>>>> On 6/9/2018 6:25 AM, Greg Dennis wrote:
>>>> > San Francisco always make the cast vote record public:
>>>> > https://sfelections.sfgov.org/june-5-2018-election-results-
>>>> detailed-reports
>>>>
>>>> I quickly looked at the vote data and saw that lots of ballots are
>>>> categorized as "Exhausted by Over Votes" and "Under Votes," but there is no
>>>> data indicating exactly how those ballots were marked, so we lack enough
>>>> information to be sure of final results.
>>>>
>>>> Converting instant-runoff counts into pairwise counts might be
>>>> (probably is?) possible, but I don't have time to do that analysis.
>>>>
>>>> > The probability of IRV not elected the Condorcet winner appears to be
>>>> > exceedingly low in practice. We're up to about ~200 IRV elections
>>>> > conducted nationwide since 2004 and Burlington 2009 is the only
>>>> > case so far.
>>>>
>>>> Yes, circular ambiguity -- in which there is no Condorcet winner -- is
>>>> rare when the number of ballots exceeds a few hundred.
>>>>
>>>> > On Sat, Jun 9, 2018 at 2:07 AM, robert bristow-johnson
>>>> > <rbj at audioimagination.com <mailto:rbj at audioimagination.com>> wrote:
>>>> > the limitation to only three levels of ranking is a problem. if
>>>> > someone ranked all three levels and none of the candidates ranked
>>>> > were either London Breed nor Mark Leno, that voter was effectively
>>>> > "disenfranchised" by being unable to weigh in on the final choice
>>>> of
>>>> > choosing the mayor.
>>>>
>>>> Based on a very quick guesstimate it looks like about 30 or so ballots
>>>> had this issue. Right?
>>>>
>>>> That's not a big number, but a fair counting method -- such as pairwise
>>>> counting -- would not have to discard any ballots.
>>>>
>>>> The bigger number is "under votes" and admittedly pairwise counting
>>>> cannot compensate for a voter saying "here is the only acceptable choice"
>>>> (or two choices in this case).
>>>>
>>>> It's great that these results are getting analyzed by people who do not
>>>> drink the FairVote kool-aid.
>>>>
>>>> In haste,
>>>> Richard Fobes
>>>> "The VoteFair guy"
>>>>
>>>>
>>>> On 6/9/2018 6:25 AM, Greg Dennis wrote:
>>>>
>>>>> San Francisco always make the cast vote record public:
>>>>> https://sfelections.sfgov.org/june-5-2018-election-results-
>>>>> detailed-reports
>>>>>
>>>>> Based on the most recent analysis of these numbers that I saw, Leno was
>>>>> indeed the Condorcet winner, and if Breed were to beat Leno in the
>>>>> final
>>>>> round, she would then necessarily be the Condorcet winner. The
>>>>> probability of IRV not elected the Condorcet winner appears to be
>>>>> exceedingly low in practice. We're up to about ~200 IRV elections
>>>>> conducted nationwide since 2004 and Burlington 2009 is the only case
>>>>> so far.
>>>>>
>>>>> On Sat, Jun 9, 2018 at 2:07 AM, robert bristow-johnson
>>>>> <rbj at audioimagination.com <mailto:rbj at audioimagination.com>> wrote:
>>>>>
>>>>>
>>>>> Richard, a few points:
>>>>>
>>>>> the limitation to only three levels of ranking is a problem. if
>>>>> someone ranked all three levels and none of the candidates ranked
>>>>> were either London Breed nor Mark Leno, that voter was effectively
>>>>> "disenfranchised" by being unable to weigh in on the final choice
>>>>> of
>>>>> choosing the mayor. however, i think the news media made it clear
>>>>> that the race was really gonna be between Leno, Breed, and Kim, so
>>>>> these fringe voters might have a chance to insincerely mark either
>>>>> Leno or Breed as their 3rd choice and betray their *true* third
>>>>> choice and, in doing so, have an effect in the final round.
>>>>>
>>>>> ignoring the problem of only 3 ranking levels, it is not possible
>>>>> that London Breed is the Condorcet Winner (a.k.a. "pairwise
>>>>> champion"). it might be the case that Mark Leno or Jane Kim is the
>>>>> Condorcet Winner and if the latter is the case, this is another
>>>>> real
>>>>> indictment against STV or IRV as a method of tallying RCV. and
>>>>> your
>>>>> reverse namesake, FairVote, is partially (or mostly) to blame.
>>>>>
>>>>> i wonder if the City of SF has a file of all of the cast and
>>>>> scanned
>>>>> ballots and the full ranking for each. if so, and if they release
>>>>> it to the public, we can investigate if there is a Condorcet Winner
>>>>> and if that CW is or is not Mark Leno. this would be interesting.
>>>>>
>>>>> L8r,
>>>>>
>>>>>
>>>>>
>>>>> r b-j
>>>>>
>>>>>
>>>>>
>>>>> > On 6/8/2018 6:24 PM, Christopher Colosi wrote:
>>>>> > > ... She stated “This is the system we are working with. That’s
>>>>> > > a discussion we can have at a later time. For now, we’re stuck
>>>>> with it.”
>>>>> > > - insinuating it is not fair. I was quite bothered to have a
>>>>> Dem
>>>>> in a
>>>>> > > progressive city insinuate that first past the post is more
>>>>> fair. ...
>>>>> >
>>>>> > This remark does not imply support for first past the post (FPTP,
>>>>> a.k.a
>>>>> > plurality counting).
>>>>> >
>>>>> > There are other ways to count the preference marks on
>>>>> "ranked-choice"
>>>>> > ballots. In particular, pairwise counting could be used instead
>>>>> of
>>>>> > instant-runoff counting, and that is fairer than FPTP.
>>>>> >
>>>>> > > 1. May not elect majority candidate
>>>>> > > ...
>>>>> > > Is this common? This is
>>>>> > > probably an abnormally close race. Thoughts?
>>>>> >
>>>>> > I doubt the voters would regard this as a close race if they had
>>>>> been
>>>>> > able to fully rank all the choices. The 3-choice limitation is
>>>>> > simplistic, and complicates the counting.
>>>>> >
>>>>> > Pairwise counting does not result in any exhausted ballots.
>>>>> Unmarked
>>>>> > choices are an indication that the choices are equally disliked.
>>>>> And
>>>>> > multiple candidates being marked at the same preference level is
>>>>> also no
>>>>> > problem.
>>>>> >
>>>>> > In other words, the ballots contain enough information that they
>>>>> can be
>>>>> > counted in other ways, besides instant-runoff counting. Those
>>>>> alternate
>>>>> > counting methods could reveal a clearer outcome.
>>>>> >
>>>>> > In haste,
>>>>> > Richard Fobes
>>>>> >
>>>>> >
>>>>> > On 6/8/2018 6:24 PM, Christopher Colosi wrote:
>>>>> >> Curious to hear people’s thoughts on some issues.
>>>>> >>
>>>>> >> 1. May not elect majority candidate
>>>>> >> In SF, we restrict to 3 choices to simplify the process. As the
>>>>> vote
>>>>> >> currently stands, 144 votes separate the top two candidates
>>>>> (<0.1%) and
>>>>> >> over 16,000 ballots have been exhausted (all 3 choices
>>>>> eliminated).
>>>>> >> About 9% of voters have been removed from the pool. It is very
>>>>> possible
>>>>> >> that the result would have shifted if they had the opportunity
>>>>> to
>>>>> rank a
>>>>> >> 4th candidate, and therefore, it is possible that we won’t
>>>>> elect the
>>>>> >> person who truly represents the majority. Is this common? This
>>>>> is
>>>>> >> probably an abnormally close race. Thoughts?
>>>>> >>
>>>>> >> 2. What are your thoughts on London Breed’s response to being
>>>>> asked if
>>>>> >> RCV is fair? She stated “This is the system we are working with.
>>>>> That’s
>>>>> >> a discussion we can have at a later time. For now, we’re stuck
>>>>> with it.”
>>>>> >> - insinuating it is not fair. I was quite bothered to have a
>>>>> Dem in a
>>>>> >> progressive city insinuate that first past the post is more
>>>>> fair. It
>>>>> >> also felt divisive. If Leno wins, will her supporters feel that
>>>>> >> democracy prevailed, or that the election was stolen? She also
>>>>> presents
>>>>> >> herself as a minority candidate and it is my understanding that
>>>>> RCV
>>>>> >> gives minority candidates better chances and causes all
>>>>> candidates to be
>>>>> >> more likely to campaign to minority communities. Am I mistaken?
>>>>> Are
>>>>> >> there any legitimate arguments that FPTP can be more fair?
>>>>> Thoughts?
>>>>> >>
>>>>> >> Regards,
>>>>> >> —Chris
>>>>> >>
>>>>> >>
>>>>> >>
>>>>> >> ----
>>>>> >> Election-Methods mailing list - see http://electorama.com/em
>>>>> for
>>>>> list info
>>>>> >>
>>>>> > ----
>>>>> > Election-Methods mailing list - see http://electorama.com/em for
>>>>> list info
>>>>> >
>>>>>
>>>>> --
>>>>>
>>>>> r b-j rbj at audioimagination.com
>>>>> <mailto:rbj at audioimagination.com>
>>>>>
>>>>> "Imagination is more important than knowledge."
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> ----
>>>>> Election-Methods mailing list - see http://electorama.com/em for
>>>>> list info
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> *Greg Dennis, Ph.D. :: Policy Director*
>>>>> Voter Choice Massachusetts
>>>>>
>>>>> e :: greg.dennis at voterchoicema.org <mailto:greg.dennis@
>>>>> voterchoicema.org>
>>>>> p :: 617.863.0746 <tel:617.863.0746>
>>>>> w :: voterchoicema.org <http://voterchoicema.org/>
>>>>>
>>>>> :: Follow us on Facebook
>>>>> <https://www.facebook.com/voterchoicema> and Twitter
>>>>> <https://twitter.com/voterchoicema> ::
>>>>>
>>>>>
>>>>> ----
>>>>> Election-Methods mailing list - see http://electorama.com/em for list
>>>>> info
>>>>>
>>>>> ----
>>>> Election-Methods mailing list - see http://electorama.com/em for list
>>>> info
>>>>
>>>
>>> ----
>>> Election-Methods mailing list - see http://electorama.com/em for list
>>> info
>>>
>>>
>> ----
>> Election-Methods mailing list - see http://electorama.com/em for list
>> info
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20180610/a2e2355d/attachment-0001.html>
More information about the Election-Methods
mailing list