Where the best Condorcet methods (Schulze, Ranked Pairs, River etc.) differ (Toby Pereira)
On Wed, Apr 14, 2021 at 12:54 PM robert bristow-johnson
rbj@audioimagination.com wrote:
But the voters of the City of Burlington did pass the Charter Change
returning RCV to Burlington after 11 years and uncorrected. Same old Hare
RCV that FairVote sells everyone else. [...] When this Charter Change becomes
a bill and is introduced to the legislature, it will be directed to the
Government Operations Committee. [...] As soon as I find out more about
when this Charter Change comes up, I will contact y'all and we can discuss
what we can do. But feel free to contact me or even post here on the list
what you might be thinking about this and how best to advocate.
Thanks for the background. It seems to me that the strategy with the
best odds of working would be to ask the Government Operations
Committee to send the bill back to the Burlington City Council for
modification. We make sure that the Government Operations Committee
is aware of the 2009 election, and suggest the smallest possible tweak
to the language that was approved in the Charter Change.
The tweak that I would propose: when the winner of the Hare RCV
election is NOT the undefeated pairwise winner, then the candidate who
WAS the undefeated pairwise winner has the legal right to issue a
"pairwise challenge". A pairwise challenge would be a runoff election
between the pairwise winner and the Hare RCV winner. So (for example)
in 2009, Andy Montroll would have had the right to file for holding a
pairwise challenge election to Bob Kiss. Importantly, Kurt Wright
would NOT have had the right to hold a pairwise challenge, since (in
2009) Wright had pairwise losses to both Montroll and to Kiss.
My proposed tweak would probably not deal with the case where the
Smith Set has three or more candidates in a pairwise loop. That seems
okay to me; I don't think we should let the perfect be the enemy of
the good. I'm not aware of any real-world election that had three or
more members in the Smith Set. However, I am aware of at least two
real-world elections (Burlington 2009 and Debian 2003) where Hare RCV
and a CWC-compliant method resulted in a different winner. (Note:
"CWC-compliant" is "Condorcet-winner-criterion compliant"). Moreover,
it seems very likely that there will be many others if we keep
gambling with Hare RCV.
Robert, if you're able to give me a link to the text of the Charter
Change, then I'll take a crack at crafting some legalese to describe
my proposed tweak. The goal of crafting the legalese would be to have
something to present to the Burlington City Council, should the State
of Vermont Government Operations Committee send the bill back to the
Burlington City Council. If nothing else, I'd personally feel more
comfortable testifying if I had the text of the Charter Change (and a
minimal proposed change) in my back pocket.
Sound reasonable?
Rob
p.s. we need a less-nerdy way to describe the "Condorcet-winner criterion"
I'm not suggesting re-naming the academic name for "Bottom Two Runoff
IRV," yet I am suggesting that for political discussion purposes you
come up with something that is easier to say, and easier for a
non-academic person to remember.
So, here are some off-the-top-of-my-head name ideas for BTR-IRV:
Ranked Choice With Bottom Two Runoff (RCWBTR)
If IRV is "Ranked Choice Hare" then:
Ranked Choice Beyond Hare (RCBH)
Ranked Choice Better Than Hare (RCBTH)
Admittedly some of these suggestions (and others that come to mind) are
half-silly. Yet I'm sure other folks here can suggest better names.
As a reminder, the person who was hired to promote STAR voting under its
original name said she would only do the job if they came up with a
better name, and that's when it became STAR voting. I don't recall the
original name.
FYI, STAR advocates sometimes refer to STAR voting as "Ranked Choice 2.0".
Names do make a difference.
BTW, I avoid the word "Condorcet" when discussing Ranked Choice
Including Pairwise Elimination (RCIPE) because the FairVote folks have
poisoned that word. Also, it refers to Condorcet winners whereas RCIPE
eliminates Condorcet losers, and does not always yield a Condorcet winner.
I do use the word "pairwise" which is self-descriptive. And when first
referring to pairwise vote counting, I may mention "one-on-one matches"
along with a soccer team analogy.
I and others are also working toward an animated video that demonstrates
RCIPE in a way that does not say or show any numbers. It will just show
stacks of ballots, with the different heights visually indicating the
number of ballots. As they say, a picture is worth a thousand words. I
suppose that means a video is worth ten thousand words.
Richard Fobes
On 4/15/2021 10:56 AM, robert bristow-johnson wrote:
On 04/15/2021 12:29 PM Kristofer Munsterhjelm km_elmet@t-online.de wrote:
On 15/04/2021 05.45, VoteFair wrote:
I looked at your proposed wording change and I like the fact that it's a
simple way to improve IRV. (Anything helps.)
As I understand it, there's a pairwise comparison between the two
candidates who have the fewest transferred votes, and the loser in that
pair is eliminated.
Perhaps you should give it a name. Perhaps one that begins with the
words "Ranked Choice ...". But not Ranked Choice Including Pairwise
Elimination, aka RCIPE, because that's already in use.
It's called BTR-IRV in this post:
https://election-methods.electorama.narkive.com/LKfc52OI/an-example-of-btr-stv#post4
(Electowiki also has a page on it:
https://electowiki.org/wiki/Bottom-Two-Runoff_IRV)
That's Bottom Two Runoff (but also "better").
The term "ranked choice voting" is too ambiguous for me to tell whether
you could just move it right over and call the method BTR-RCV. To me,
"ranked choice" sounds like just another way of saying "ranked", but of
course, FairVote thinks differently.
-km
actually at the top of the thread it was called "BTR-STV". I include a screenshot, but i dunno if the list will pass attachments. i considered BTR-RCV but i didn't like two "R"s. i sent Rob LeGrand an email longer ago and i don't think he objected.
because FairVote has ditched "IRV" (because it lost cache) and dishonestly replaced it with the more general term "RCV", I want to associate "IRV" with only the Hare (or Ware) method. Not with a reform that makes in Condorcet compliant.
I credit Hare more than Ware for this method that is in wide use today. Hare's innovation is the Single Transferrable Vote which gave this method some legal meat on the bones. Hare kept it generalized for multi-winner elections. Ware's only contribution was "Hay, let's use this Hare method for single-winner elections."
Because of FairVote's appropriation of "RCV" for the Hare method that they are selling, I want to always differentiate the Hare method from all the other RCV methods (Borda, Bucklin, and the several Condorcet methods).
I believe "Hare RCV" is the best label I can toss in to make sure people know what I am talking about in relation to the now common term "RCV".
My personal favorite Condorcet method is not BTR-STV because it is still sequential rounds and does not allow equal ranking of candidates. But as I said on this mailing list at least a couple times: Some Condorcet RCV is better than any non-Condorcet RCV. Especially since I am already arguing about a corner case (when Hare RCV did not elect the Condorcet winner). The difference between the Condorcet methods is the corner of the corner case. Already FairVote is arguing that what happened in Burlington is likely to never happen again and happened only once out of something like 300 RCV elections.
But (with apologies to Markus) what is important is to get a decent Condorcet method into some simple and completely self-contained legal language understandable in the common tongue. Somewhere on this list someone pointed out BTR-IRV to me (since I was not here in 2006, I only joined after the Burlington 2009 election and it was my current opponent, Terry Bouricius, that told me about this list in 2009). BTW, he called this my "personal pet project" when he was describing the Charter Change question on the CCTV. I can find a YouTube link if you're interested.
Election-Methods mailing list - see https://electorama.com/em for list info
On 15/04/2021 19.56, robert bristow-johnson wrote:
On 04/15/2021 12:29 PM Kristofer Munsterhjelm km_elmet@t-online.de wrote:
It's called BTR-IRV in this post:
https://election-methods.electorama.narkive.com/LKfc52OI/an-example-of-btr-stv#post4
(Electowiki also has a page on it:
https://electowiki.org/wiki/Bottom-Two-Runoff_IRV)
That's Bottom Two Runoff (but also "better").
The term "ranked choice voting" is too ambiguous for me to tell whether
you could just move it right over and call the method BTR-RCV. To me,
"ranked choice" sounds like just another way of saying "ranked", but of
course, FairVote thinks differently.
-km
actually at the top of the thread it was called "BTR-STV". I
include a screenshot, but i dunno if the list will pass attachments. i
considered BTR-RCV but i didn't like two "R"s. i sent Rob LeGrand an
email longer ago and i don't think he objected.
I think BTR-STV is the multiwinner generalization. It's also called
STV-ME (e.g.
http://lists.electorama.com/pipermail/election-methods-electorama.com/2006-June/083621.html).
My personal favorite Condorcet method is not BTR-STV because it is
still sequential rounds and does not allow equal ranking of candidates.
But as I said on this mailing list at least a couple times: Some
Condorcet RCV is better than any non-Condorcet RCV.
It's algorithmically pretty easy to do equal-rank with BTR-IRV: Just
count the first preferences as fractions of a vote when there's a tie.
(There are other options that may be better still, but they're more
complex.)
This is not as egregrious as fractional Plurality because as soon as all
but one of the equal-ranked candidates have been eliminated, the ballot
counts as full strength for the remaining candidate.
It would be more complex on the vote-counting end, though, so it might
not be worth it.
Already FairVote is arguing that what happened in Burlington is
likely to never happen again and happened only once out of something
like 300 RCV elections.
Steve Eppley's response is probably the best one there: what elections
you get (and how many of them are simple enough for the method to
resolve) depends on what that method is.
okay, more about terminology and semantics and strategy...
Again, I only want to use the term "IRV" for Hare RCV (a.k.a. Hare STV) which is the common RCV you see in Australia, Ireland, Maine, several U.S. cities including Burlington from 2005 to 2010. I do not want to use that term for Rob LeGrand's adaptation of IRV to be Condorcet compliant.
I will make sure that the audience (this Gov Ops committee and anyone else in the legislature) will understand that "RCV" means any ranked ballot and will allude to how FairVote appropriated that term for two disingenuous reasons: 1. "IRV" lost cache and they realized that and 2. they want to insinuate that any time "RCV" is used, it means the product they are selling. They're like "Of course Hare RCV is the only way to implement RCV in government elections." But, in discussion, I will always used "Hare RCV" or "Hare STV" to mean the FairVote method.
I will make clear that Condorcet preceded Hare by a half century and considered Hare's method before Hare did. Hare came up with the term "Single Transferable Vote" I believe and really was the main innovator for that method. Ware contributed nothing, really, but is sometimes credited with creating IRV. STV is a useful legal instrument, but I would prefer a flat method such as Ranked Pairs or Schulze. But I will not promote either methods. It's important to make this reform self-contained (all of the legal language exist in the law without external reference) and as short and as familiar as possible.
BTR-STV (and that is the label I will use) is the smallest incremental change to the Hare STV method that will make it Condorcet compliant. I only want to say "Condorcet method" or "Condorcet-compliant" (as an adjective) for any method that satisfies the Condorcet criterion. I think that "CWC-compliant" or "Condorcet-winner-criterion compliant" is more verbiage than necessary.
Now I would like to know more about this Debian 2003 thing. I thought it was a city, but was it about this: https://www.debian.org/News/2003/20031010 ? I would like solid information about any other time (other than Burlington 2009) that Hare RCV did not elect the Condorcet Winner. FairVote claims there is only one case out of more than 300. I am skeptical that they were able to get ballot date for 300+ elections but if anyone knows a solid case of another IRV failure, I would appreciate knowing it. (I haven't done the math, but if the 3rd-place candidate is low enough relative to the top two, I think you can prove mathematically that the CW was elected with Hare-STV even if you don't have the individual ballot data. The failure of IRV happens when it's a close 3-way race and the spoiler candidate has enough support to get into the final round and loses.)
About a good name for the Condorcet Winner (sometimes "Condorcet candidate"), there was a discussion on this list a while back. "Beats-all winner" sucks. It sounds aggressive and tacky. In the 2004 Scientific American article https://drive.google.com/file/d/1m6qn6Y7PAQldKNeIH2Tal6AizF7XY2U4/view?usp=sharing the coined "True Majority Rule" and I will point to that but not use it.
"Pairwise Champion" was mentioned and isn't so bad, but I coined the term "Consistent Majority Candidate" because I will get into the claim that IRV "guarantees a majority winner". I will ask them how to justify that claim and eventually they will have to boil it down to two candidates. I will get specific about Burlington 2009 and they will claim that Bob Kiss was the majority candidate because of his 4313 to 4061 majority over Kurt Wright. Between two candidates there is always a simple majority, unless they tie. But I will ask, "Why not identify Kurt Wright as the majority candidate over Dan Smith (3971 over 3793)?" and if they say "Because the voters preferred Kiss over Wright so Wright's majority over Smith is not relevant", then I will pounce.
What I need are more knowledgeable people that are willing to say that Hare RCV is flawed and why. My position is that their three big selling points: 1. guarantees majority winner, 2. eliminates spoiler effect, 3. removes tactical burden from voters that might want to vote 3rd party or independent. It failed all three in 2009. I will also point out that when you fail to elect the true majority candidate (the CW), that besides spoiling the election, you have valued the votes of the fewer voters preferring the IRV non-majority candidate over the greater number of voters preferring the true majority candidate. In my opinion, One-person-one-vote is sacrosanct and violating it in a democracy is profane.
And what I need are as many solid examples of IRV failure as I can get. So far, all I have is Burlington 2009 (other than the other towns that have repealed IRV because they just didn't like the ranked ballot).
So showing up and cheering will help. Perhaps answering directed questions. And other examples of "bad IRV" or where Condorcet (likely to be Schulze) is used and is good. There appears to be one city in Portugal that has used Schulze in government. But I will have to point to organizations.
I will also talk about corner cases. Of all RCV elections in government, perhaps 99.999% have a Condorcet Winner (no cycles). And perhaps in about 99%, Hare RCV had elected the CW. So I will be arguing about the 1% corner case which has happened at least once (in my very own city in 2009).
My point will be that even though this is a corner case, it is worth correcting, much like correcting the few cases where the Electoral College did not elect the popular vote winner. And now in Vermont is the point, the optimal time and place, to make that correction rather than jumping on the FairVote train.
If anyone brings up Arrow and the 0.001% corner case, I will acknowledge it, illustrate it with a simple Rock-Paper-Scissors example and show that BTR-STV will be more consistent than Hare-STV. BTR will always elect the largest vote getter of the three (let's label that candidate "Rock") and Hare will sometimes elect Paper, sometimes Rock.
But that is the minutest corner of the corner case and, as Arrow points out, it can be that it is just not possible to elect a candidate that has no one else that voters prefer more and that is an unavoidable flaw. However failing to elect the CW when such exists (and the CW almost always exists) is an avoidable flaw. And democracy is worth it to preemptively correct a flaw resulting in even a rare avoidable failure.
"That's my story and I'm sticking to it."
--
r b-j rbj@audioimagination.com
"Imagination is more important than knowledge."
Hi Robert,
Other parts of your email also deserve reply, but I just wanted to
follow up on the "Debian 2003" reference
On Fri, Apr 16, 2021 at 3:25 PM robert bristow-johnson
rbj@audioimagination.com wrote:
Now I would like to know more about this Debian 2003 thing. I thought it was
a city, but was it about this: https://www.debian.org/News/2003/20031010 ?
Sorry for the confusion; it was about a different election, but you
found the right organization. "Debian" has long been a very popular
distribution of Linux, which serves as the foundation for other
popular Linux distributions (e.g. Ubuntu Linux, Mint Linux,
elementaryOS, PopOS, and many others). When I refer to "Debian 2003",
I mean the "Debian Project Leader" (DPL) annual election for 2003,
which is usually that project's most important election of the year.
The Debian project has been electing their project leader using
Schulze/Condorcet since the very early days of the project, thanks to
several of us (led by Mike Ossipoff and Markus Schulze, among others)
convincing their leaders that the "Concorde method" they were using
was a mildly-flawed implementation of a Condorcet method.
Debian has published their annual election results for the DPL
elections, and 2003 was interesting:
https://www.debian.org/vote/2003/vote_0001
My interest was for similar reasons to yours (as you wrote):
I would like solid information about any other time (other than Burlington 2009) that
Hare RCV did not elect the Condorcet Winner.
Before Burlington 2009, that 2003 DPL election ("Debian 2003") was my
standard example of a real-world Condorcet<->IRV discrepancy. Shortly
after the Debian 2003 election was held, I analyzed the results, and
wrote up my findings, publishing an essay on Electorama. I recently
republished those findings on Electowiki:
https://electowiki.org/wiki/User:RobLa/Debian2003
The DPL election wasn't a municipal election, so it's not as
interesting as government-administered elections. Still, it's
interesting because the people voting in the election were not voting
nerds, but rather, were keenly interested in the results of the
election for the sake of their project. It appeared to be a close
election in 2003.
FairVote claims there is only one case out of more than 300 [of a Condorcet <-> IRV discrepancy].
I don't think I would book a flight out of any airport that claims
that 299 out of every 300 airplanes makes it off the runway, and then
tries to dismiss every 300th airplane as rare flukes that aren't
worrying about ("just look at the other 299 flights!"). I share a bit
of your skepticism about the ballot data, but it wouldn't entirely
surprise me if FairVote's analysis is accurate. It seems relatively
rare for the Condorcet winner to be different from the IRV winner.
It may be that we just haven't looked hard enough (where "we" is the
group of people who prefer Condorcet winners prevail in ranked ballot
elections), or it could be that there are very examples involving
real-world election outcomes. Like you, I'm really interested in
other election data which either confirms or disproves the rarity of
Condorcet<->IRV discrepancies, since my hunch+fear is that they'll be
more common than IRV advocates suggest.
Rob
On 17/04/2021 11.03, Rob Lanphier wrote:
Before Burlington 2009, that 2003 DPL election ("Debian 2003") was my
standard example of a real-world Condorcet<->IRV discrepancy. Shortly
after the Debian 2003 election was held, I analyzed the results, and
wrote up my findings, publishing an essay on Electorama. I recently
republished those findings on Electowiki:
https://electowiki.org/wiki/User:RobLa/Debian2003
The DPL election wasn't a municipal election, so it's not as
interesting as government-administered elections. Still, it's
interesting because the people voting in the election were not voting
nerds, but rather, were keenly interested in the results of the
election for the sake of their project. It appeared to be a close
election in 2003.
That's a good example of IRV's chaos problem - that a slight
modification to the ballots can unpredictably change who the winner is.
Every Condorcet-compliant variant of IRV (naturally) cuts the chaos from
the regions where there's a CW.
-km