[EM] One pager about what happened in Vermont

Rob Lanphier robla at robla.net
Sat Mar 6 23:49:02 PST 2010

Hi Robert,

Thanks for the detailed feedback.  I've got a slightly updated version
available here:

More inline...

On Sat, Mar 6, 2010 at 4:16 PM, robert bristow-johnson <
rbj at audioimagination.com> wrote:

> On Mar 6, 2010, at 4:18 PM, Rob Lanphier wrote:
>  The key insight I had was to break the electorate up by their first
>> preference, and then do Schulze-wv tallies on each subset of ballots.  That
>> yielded a reasonably realistic way to lay out the candidates in a 2D spatial
>> representation.
> Schulze?  For the viewer who might have trouble visualizing IRV (despite
> your quite good layout, i really like what you're doing)?

Well, there's nothing Schulze-specific in the tallies...that's just the
Condorcet-variant that's my default starting point (most used in the wild,
and most tested in Electowidget, but I just as easily could have used any
other variant in this case).   The only point of that part of the exercise
was to get a sense of what each of the constituencies thought of the other
candidates, which was a pretty quick exercise once I had all of the data

That said, since it was only a one line change to this page, I changed from
Schulze-wv to Copeland, because Copeland is the most accessible of the
Condorcet methods:

listen the Condorcet result of the 2009 election is unambiguous.  it gets
> ordered [M]ontroll 1st, [K]iss, [W]right, [S]mith, and [H]Simpson last.
>  every candidate ordered higher beats any candidate ordered lower in a
> head-to-head race. i think there are 10 head-to-head pairings.
>  M 4064
>  K 3477
>  M 4597     K 4313
>  W 3664     W 4061
>  M 4570     K 3944     W 3971
>  S 2997     S 3576     S 3793
>  M 6263     K 5515     W 5270     S 5570
>  H  591     H  844     H 1310     H  721
> i like a triangle instead of a defeat matrix, at least if it can be ordered
> so nicely.

I've got some ideas I'm going to play around with.  The Debian folks do a
pretty good job displaying the results of their elections:


I think that type of a diagram could possibly be dolled up and fleshed out a
little bit.  However, before I go down that road, I'm planning to play
around with a few new ideas of my own to see if I can hit upon something
that really tells the story.

> "H" is for "Homer".  notice how well he does against Wright.  what does
> that tell us about Wright's "negatives"?  and Wright *barely* beast Smith
> (and Independent with no party backing).  i just can't believe the New North
> Enders, who were primarily behind the repeal movement, think that Wright was
> the guy who was robbed, because he had the plurality of first-choice votes.

I'm certain of it.  It also doesn't help that their least liked candidate is
someone that they generally preferred *two other viable candidates* to.  The
whole *point* of egghead solutions like IRV is that this sort of thing isn't
supposed to happen.  I can fully imagine them assuming this whole thing was
the conspiracy all along in order to let the hippie communists like Kiss run
things (just to be clear...I'm merely channeling an imaginary Wright voter.
 I have no idea whether Kiss is a "hippie" or a "communist", or if there's
even a Wright voter out there who might even think that.  Whew....that was a
lot of work for a joke that wasn't even very good.  Though I'll be there is
a least one Wright voter who thinks that Kiss is a hippie and a communist.
 Just sayin')

That's my longwinded way of saying that while Wright voters' only have a
legitimate beef about one of the candidates not winning (Montroll), it's not
hard to imagine why they likely felt triply screwed over in this election.

my numbers come from the files from the clerk's office.  do check out
> Warren's page at http://rangevoting.org/Burlington.html .  there are small
> discrepancies, but no more than 5 votes.

I'll double check the numbers.  I pulled mine from a zipfile that was linked
to from the Wikipedia page.  I *think* they were the official results, but I
might be wrong.

Note that what may have happened was that they updated the "final" results a
month or two after the election.  If you pulled your results very soon after
the original election, it's quite possible that you have something different
from the same source.  I don't know if there were any official recounts, but
it wouldn't surprise me at all given the hubbub about this particular race.

> i mean, if i wanted to see, in a nutshell, what happened, what went wrong,
> what could have gone wronger with IRV.
I think we're on the same page.  This particular one pager was actually
meant to be "one of x", where x>1.  I don't think I'm going to try to
illustrate "wronger" with this particular case (wronger is illustrated
pretty well with the Tennessee example[1] and with Debian 2003[2]), but I do
want the mechanics of IRV in this particular election to be crystal clear,
as well as showing how a Condorcet-compliant method would have been much,
much better.  This result was bad enough and consequential enough that I
think it's important there's a more general understanding of it.

One of the few thigns I like about IRV is that we get really great data from
IRV elections (at least, when we get *full* results).  While it's
theoretically possible to game IRV elections, I doubt there's much strategic
voting going on in any place other than Australia perhaps....people here
just don't know how to do it yet.  So, the results are about as purely
sincere as you're probably going to get in a real election.

> this looks very cool, Rob.


Links from above:
[1]  Tennessee example:
[2] Debian 2003:
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20100306/78063f37/attachment-0004.htm>

More information about the Election-Methods mailing list