RCV Challenge
Hi Robert,
Le samedi 25 décembre 2021, 00:29:30 UTC−6, robert bristow-johnson rbj@audioimagination.com a écrit :
I'm just a Condorcet guy. Everyone's vote counts the same. If a simple majority
of voters agree that A is preferred to B, then when at all possible, B is not
elected. After that add whatever contingency you think best for the case of a
cycle. I, personally, believe that cycles will be very rare and perhaps the
simplest process in lieu of a Condorcet winner would be the plurality winner of
first-preference votes. I know everyone will say that's the worst, most gameable
Condorcet "completion process", but I think in a government election that the
rules should be well known and well understood.
In the three-candidate case Condorcet//FPP should be the same thing as BTR-IRV,
except with greater clarity, as you point out, about how the method will play out
in general. I value that a lot.
It won't be my own favorite method, just because my main interest is in reducing
compromise incentive (or: minimizing the appearance that some candidate was a
spoiler).
The question of how gameable it is, vs. certain other methods, is interesting.
I'll leave it for now. But it didn't occur to me to think it was the worst or
most gameable rule.
Kevin
On 24.12.2021 01:08, Forest Simmons wrote:
Despite our best efforts, I'm not sure that we've yet seen or heard the
best possible deterministic, Ranked Choice Voting proposals.
In my next message I will submit the best public proposal that I can
think of in that category (the category of Universal Domain ... i.e.
based purely on Ranked Choice/Preference information ... equal rankings
and truncations allowed). Of course, anybody can easily improve on any
such method by coloring outside of the UD lines ... for example by use
of explicit approval cutoffs, scores, grades, judgments, virtual
candidates, and other devices for stratifying rank relations by relative
importance/strength, as well as probabilities, random ballot drawings, etc.
But let's temporarily put aside all of these power tools and see what we
can accomplish with screwdriver, pliers, etc.
The challenge is to make the method as simple as possible while
complying with clone independence, monotonicity, and the other most
basic criteria like Pareto, anonymity, neutrality, majority, etc.
Simplicity is in the eye of the beholder ... hard to pin down, but you
know it when you see it.... definitely not just a bunch of ad hoc rules
cobbled together to patch up an out moded second rate method from
yesteryear. The fewer seams, the better.
How about this for minmax?
A candidate clears a threshold if he beats every other candidate by at
least the threshold's margin, one-on-one.
Elect the candidate who clears the highest threshold.
(I.e. the candidate who gets closest to a landslide against any opponent.)
It's not cloneproof, though.
-km
Succinct and seamless enough to amply compensate for marginal clone
dependence!
El sáb., 25 de dic. de 2021 1:53 a. m., Kristofer Munsterhjelm <
km_elmet@t-online.de> escribió:
On 24.12.2021 01:08, Forest Simmons wrote:
Despite our best efforts, I'm not sure that we've yet seen or heard the
best possible deterministic, Ranked Choice Voting proposals.
In my next message I will submit the best public proposal that I can
think of in that category (the category of Universal Domain ... i.e.
based purely on Ranked Choice/Preference information ... equal rankings
and truncations allowed). Of course, anybody can easily improve on any
such method by coloring outside of the UD lines ... for example by use
of explicit approval cutoffs, scores, grades, judgments, virtual
candidates, and other devices for stratifying rank relations by relative
importance/strength, as well as probabilities, random ballot drawings,
etc.
But let's temporarily put aside all of these power tools and see what we
can accomplish with screwdriver, pliers, etc.
The challenge is to make the method as simple as possible while
complying with clone independence, monotonicity, and the other most
basic criteria like Pareto, anonymity, neutrality, majority, etc.
Simplicity is in the eye of the beholder ... hard to pin down, but you
know it when you see it.... definitely not just a bunch of ad hoc rules
cobbled together to patch up an out moded second rate method from
yesteryear. The fewer seams, the better.
How about this for minmax?
A candidate clears a threshold if he beats every other candidate by at
least the threshold's margin, one-on-one.
Elect the candidate who clears the highest threshold.
(I.e. the candidate who gets closest to a landslide against any opponent.)
It's not cloneproof, though.
-km
On 12/25/2021 3:05 AM Kevin Venzke stepjak@yahoo.fr wrote:
Hi Robert,
Le samedi 25 décembre 2021, 00:29:30 UTC−6, robert bristow-johnson rbj@audioimagination.com a écrit :
I'm just a Condorcet guy. Everyone's vote counts the same. If a simple majority
of voters agree that A is preferred to B, then when at all possible, B is not
elected. After that add whatever contingency you think best for the case of a
cycle. I, personally, believe that cycles will be very rare and perhaps the
simplest process in lieu of a Condorcet winner would be the plurality winner of
first-preference votes. I know everyone will say that's the worst, most gameable
Condorcet "completion process", but I think in a government election that the
rules should be well known and well understood.
In the three-candidate case Condorcet//FPP should be the same thing as BTR-IRV,
except with greater clarity, as you point out, about how the method will play out
in general. I value that a lot.
It won't be my own favorite method, just because my main interest is in reducing
compromise incentive (or: minimizing the appearance that some candidate was a
spoiler).
The question of how gameable it is, vs. certain other methods, is interesting.
I'll leave it for now. But it didn't occur to me to think it was the worst or
most gameable rule.
Hi Kevin,
If there are conditions that can make a candidate a spoiler, can't that be used in some way to swing the election?
I mean, it depends on what you call "gaming" an election. In Burlington 2009, if 371 or more Wright voters that disliked Kiss the most had conspired to compromise their vote and rank Montroll over Wright and snatch the election away from Kiss. Would that be "gaming the system"?
I just thought that any vulnerability to a spoiler pathology can be used to some side's advantage.
--
r b-j . _ . _ . _ . _ rbj@audioimagination.com
"Imagination is more important than knowledge."
.
.
.
Hi Robert,
Le samedi 25 décembre 2021, 21:06:28 UTC−6, robert bristow-johnson rbj@audioimagination.com a écrit :
In the three-candidate case Condorcet//FPP should be the same thing as BTR-IRV,
except with greater clarity, as you point out, about how the method will play out
in general. I value that a lot.
It won't be my own favorite method, just because my main interest is in reducing
compromise incentive (or: minimizing the appearance that some candidate was a
spoiler).
The question of how gameable it is, vs. certain other methods, is interesting.
I'll leave it for now. But it didn't occur to me to think it was the worst or
most gameable rule.
Hi Kevin,
If there are conditions that can make a candidate a spoiler, can't that be used in
some way to swing the election?
Yes, I think by definition.
I mean, it depends on what you call "gaming" an election. In Burlington 2009, if
371 or more Wright voters that disliked Kiss the most had conspired to compromise
their vote and rank Montroll over Wright and snatch the election away from Kiss.
Would that be "gaming the system"?
I just thought that any vulnerability to a spoiler pathology can be used to some
side's advantage.
I don't have a personal definition for "gaming." It could certainly include any
instance where voters vote insincerely in order to get a better result. However,
people tend to have a more negative opinion of strategies like burial, where the
voter completely fabricates a preference.
With compromise strategy, attitudes are usually different. Picture under FPP the
Nader voter (to use an old example) who can vote sincerely for Nader or
strategically for Gore. If he votes for Nader, his least favorite candidate Bush
wins. If he votes for Gore, at least his second choice Gore will win. A common
feeling is that the Nader voter shouldn't have to make this decision. That, if
Nader is not a viable candidate then the method itself should, effectively,
execute the strategy on the Nader voter's behalf. (Or, equivalently, if the
method can tell that Nader isn't viable, then it should drop him out of the race
as part of the algorithm, such that the result is undisturbed whether he has
entered the race or not.)
So the voter tells the method what he wants, and the method should represent his
interests. In theory the voter then doesn't need to misrepresent preferences by
prematurely compromising. And also on Nader, the candidate, there is less
pressure to drop out of the race if he isn't going to win it. There should be
potentially more choices available on the ballot, and more voters feeling safe
to indicate preferences for those choices.
Burlington looks much the same to me. If some Republicans abandoned their first
preference candidate to stop the Progressive from winning, you could call it
"gaming" on their part, but any Condorcet advocate will sooner argue that they
shouldn't need to do this, that the result they gain from strategizing in this
IRV election is actually what they are entitled to naturally.
Kevin
Hi Forest,
Does IACC give different results from TACC? Off the top of my head it seems that
in a three-candidate cycle both methods will elect whichever candidate beats the
approval loser.
For myself, here's a short list of ideas.
- As you initially described this challenge (with UD, monotonicity, clone
independence) the bar seemed very high, making me think only a few methods would
be worth mentioning.
One of them is River. If I'm not mistaken River's resolution process can be
explained and executed very easily.
Say that each candidate "has" a set of candidates, which initially contains only
the candidate himself.
Evaluate the propositions from strongest to weakest.
When you consider "A beats B," ask simply whether B is still in his original
set. If so, then move every candidate in B's set into whichever set A currently
occupies. Otherwise do nothing.
(If A is in B's original set, it's the same result to move or to not move.)
After considering all the propositions, elect from among the candidates who
remain in their original set.
If I deviate from the criteria:
-
Condorcet//Approval with implicit approval is probably the simplest Condorcet
method that I like. I think with three slots it's best, though. It's very simple
to understand, and burial strategy has high risk. -
I am still intrigued by the "BTP" method from a year ago, in which voters
basically acquiesce to every candidate who "Beats or Ties" each candidate
"Preferred" to them on that ballot. I'm still unsure what pitfalls this method
may have.
It is a strange quirk to see, for example, Gore prevail over Bush with
acquiescence interpreted to come not from the Nader voters, but from the Bush
voters.
-
Not quite a Condorcet method, but I like Approval-Elimination Runoff in which
we eliminate the least-approved candidates until there is a majority favorite. -
My "King of the Hill" method. Between the first preference winner and the
candidate with the most first preferences who is involved in a majority-strength
pairwise contest with the FP winner, elect the winner of that contest, if there is
such a contest.
This is based on the intuition that a high first preference count indicates
viability, and that respecting a full majority among such candidates should at
least ensure the winner comes from the "correct" side of the spectrum. This
method also admits no burial strategy.
Kevin
Kevin,
Simplest River formulation ever! I think that's my new favorite.
And thanks for keeping all of these valuable insights alive ... very
stimulating!
And yes, IACC is just another formulation of TACC.
In that regard I wonder if it would be better to use the implicit approval
winner's recommended/published order rather than the implicit approval
order itself. It seems like that would strongly discourage the likely IA
winner from burying the sincere CW.
El dom., 26 de dic. de 2021 1:17 p. m., Kevin Venzke stepjak@yahoo.fr
escribió:
Hi Forest,
Does IACC give different results from TACC? Off the top of my head it
seems that
in a three-candidate cycle both methods will elect whichever candidate
beats the
approval loser.
For myself, here's a short list of ideas.
- As you initially described this challenge (with UD, monotonicity, clone
independence) the bar seemed very high, making me think only a few methods
would
be worth mentioning.
One of them is River. If I'm not mistaken River's resolution process can be
explained and executed very easily.
Say that each candidate "has" a set of candidates, which initially
contains only
the candidate himself.
Evaluate the propositions from strongest to weakest.
When you consider "A beats B," ask simply whether B is still in his
original
set. If so, then move every candidate in B's set into whichever set A
currently
occupies. Otherwise do nothing.
(If A is in B's original set, it's the same result to move or to not move.)
After considering all the propositions, elect from among the candidates who
remain in their original set.
If I deviate from the criteria:
-
Condorcet//Approval with implicit approval is probably the simplest
Condorcet
method that I like. I think with three slots it's best, though. It's very
simple
to understand, and burial strategy has high risk. -
I am still intrigued by the "BTP" method from a year ago, in which
voters
basically acquiesce to every candidate who "Beats or Ties" each candidate
"Preferred" to them on that ballot. I'm still unsure what pitfalls this
method
may have.
It is a strange quirk to see, for example, Gore prevail over Bush with
acquiescence interpreted to come not from the Nader voters, but from the
Bush
voters.
-
Not quite a Condorcet method, but I like Approval-Elimination Runoff in
which
we eliminate the least-approved candidates until there is a majority
favorite. -
My "King of the Hill" method. Between the first preference winner and
the
candidate with the most first preferences who is involved in a
majority-strength
pairwise contest with the FP winner, elect the winner of that contest, if
there is
such a contest.
This is based on the intuition that a high first preference count indicates
viability, and that respecting a full majority among such candidates
should at
least ensure the winner comes from the "correct" side of the spectrum. This
method also admits no burial strategy.
Kevin
Here's an idea for a burial resistant Condorcet efficient DSV Approval
flavored method:
Ballot B approves candidate X iff for each candidate Y that is unranked on
ballot B, candidate X both pairwise defeats Y and is ranked above Y on
ballot B.
[Pretty simple!]
Note that if X is a Condorcet candidate, then by this rule, X will be
approved on all ballots except those on which it is unranked, and none of
those ballots will contribute any approval to any candidate.
In summary, no candidate can have more approval than the CC when there is
one.
And the rest of this message it an effort to convince you that this method
is highly protective of sincere CC's.
Now suppose that burying the Condorcet Winner X on some ballot B puts X
into a Condorcet cycle. How will this burial affect the approval of a
typical candidate Y?
After the burial both X and any candidate Y still beaten pairwise by X will
lose the approval had before the burial.
This includes any candidate Y ranked above X before the burial.
So why would the B ballot faction want to make a move that would surely
lower approval for the set of candidates they ranked better than or equal
to X with a risky move having only a slight chance of improving some score
for some candidate they liked less than X?
I find this quite convincing .. but time will tell as we try it out on
difficult cases and random simulations.
-FWS
El jue., 23 de dic. de 2021 4:08 p. m., Forest Simmons <
forest.simmons21@gmail.com> escribió:
Despite our best efforts, I'm not sure that we've yet seen or heard the
best possible deterministic, Ranked Choice Voting proposals.
In my next message I will submit the best public proposal that I can think
of in that category (the category of Universal Domain ... i.e. based purely
on Ranked Choice/Preference information ... equal rankings and truncations
allowed). Of course, anybody can easily improve on any such method by
coloring outside of the UD lines ... for example by use of explicit
approval cutoffs, scores, grades, judgments, virtual candidates, and other
devices for stratifying rank relations by relative importance/strength, as
well as probabilities, random ballot drawings, etc.
But let's temporarily put aside all of these power tools and see what we
can accomplish with screwdriver, pliers, etc.
The challenge is to make the method as simple as possible while complying
with clone independence, monotonicity, and the other most basic criteria
like Pareto, anonymity, neutrality, majority, etc.
Simplicity is in the eye of the beholder ... hard to pin down, but you
know it when you see it.... definitely not just a bunch of ad hoc rules
cobbled together to patch up an out moded second rate method from
yesteryear. The fewer seams, the better.
Simplicity includes simplicity of data summary, simplicity of computation,
simplicity of formulation/description, etc.
One antonym of simplicity is complexity ... complexity of the basic
idea/heuristic, logical complexity, computational complexity, etc.
I look forward to seeing some of your favorite methods ... original or
not. And don't worry if they do not completely comply with the ideal
criteria I outlined above ... a really good, intuitively appealing, simple
idea can be forgiven a small transgression or two .... and could become the
germ for an even better method.
I put simplicity ahead of familiarity because a simple idea can easily
become familiar, so lack of familiarity is a temporary problem caused by a
history of poor attention to civics education.
This challenge is an opportunity for you to take one small step to help
remedy that educational deficiency!
Thanks!
-Forest
On 12/26/2021 1:38 PM Kevin Venzke stepjak@yahoo.fr wrote:
Le samedi 25 décembre 2021, 21:06:28 UTC−6, robert bristow-johnson rbj@audioimagination.com a écrit :
...
If there are conditions that can make a candidate a spoiler, can't that be used in
some way to swing the election?
Yes, I think by definition.
I mean, it depends on what you call "gaming" an election.
...
So the voter tells the method what he wants, and the method should represent his
interests. In theory the voter then doesn't need to misrepresent preferences by
prematurely compromising.
...
Burlington looks much the same to me. ... you could call it
"gaming" on their part, but any Condorcet advocate will sooner argue that they
shouldn't need to do this, that the result they gain from strategizing in this
IRV election is actually what they are entitled to naturally.
Kevin, I totally agree.
I'm just a Condorcet guy. How cycles get resolved is less motivating to me than insuring that the Condorcet winner is always elected. (And the simplest legislative language that accomplishes that goal is usually what I advocate.)
BTW, my little RCV paper about Burlington 2009 has been getting more developed. Now with figures in color sorta spelling out the Center Squeeze effect.
https://drive.google.com/file/d/1jIhFQfEoxSdyRz5SqEjZotbVDx4xshwM/view
I hope some of you guys might take a look at it, if you hadn't before.
--
r b-j . _ . _ . _ . _ rbj@audioimagination.com
"Imagination is more important than knowledge."
.
.
.
Robert,
You wrote ...
"I'm just a Condorcet guy. How cycles get resolved is less motivating to
me than insuring that the Condorcet winner is always elected"
But "how cycles get resolved" has a big influence on whether or not they
come into existence. And when these cycles are created, the sincere
Condorcet Winner goes out the window ... no longer showing up as a
"beats-all" candidate on the ballot set. Can the cycle resolution method
reconstruct the lost CW? All bets are off.
The easier it is to game the cycle resolution method, the more incentive
for the gamers to create the cycles.
It seems that most Condorcet cycles are artificially created.
For example, one way to create a cycle is by an insincere rank reversal
technique called "burial".
If the cycle resolution method has no built in disincentive/negative
feedback for this (or any other) kind of cycle creation, over time the
gamers find out that they can manipulate elections to their advantage with
impunity, so that artificially created cycles become more and more common,
giving the false impression that cycles are a normal fact of nature.
Naive voters are easy prey for sophisticated gamers unless the method
designers engineer manipulation resistant safe-guards into the method.
And that is really the only essential thing that distinguishes one
Condorcet method from another ... what to do if there is a cycle, and how
to make sure that opportunistic creators of cycles are not rewarded for
their perfidy.
So cycle resolution has to be the main concern of Condorcet method
engineers, but it should be no concern at all for the voters. That's why
seamless incorporation of the cycle resolution is so important.
Take the following RCV method for example:
Elect the candidate which on the most ballots beats, both in rank and
pairwise, every candidate unranked by that ballot.
This succinct 20 word method definition makes no mention of cycles or
Condorcet winners or majorities whatsoever, yet it completely resolves all
cycles automatically, disincentivizes manipulators, and reliably elects the
resulting vouch safed Condorcet Winner.
The voter can drive the car so-to-speak, without worrying about the
machinery under the hood.
-FWS
El lun., 27 de dic. de 2021 7:35 p. m., robert bristow-johnson <
rbj@audioimagination.com> escribió:
On 12/26/2021 1:38 PM Kevin Venzke stepjak@yahoo.fr wrote:
Le samedi 25 décembre 2021, 21:06:28 UTC−6, robert bristow-johnson <
rbj@audioimagination.com> a écrit :
...
If there are conditions that can make a candidate a spoiler, can't
that be used in
some way to swing the election?
Yes, I think by definition.
I mean, it depends on what you call "gaming" an election.
...
So the voter tells the method what he wants, and the method should
represent his
interests. In theory the voter then doesn't need to misrepresent
preferences by
prematurely compromising.
...
Burlington looks much the same to me. ... you could call it
"gaming" on their part, but any Condorcet advocate will sooner argue
that they
shouldn't need to do this, that the result they gain from strategizing
in this
IRV election is actually what they are entitled to naturally.
Kevin, I totally agree.
I'm just a Condorcet guy. How cycles get resolved is less motivating to
me than insuring that the Condorcet winner is always elected. (And the
simplest legislative language that accomplishes that goal is usually what I
advocate.)
BTW, my little RCV paper about Burlington 2009 has been getting more
developed. Now with figures in color sorta spelling out the Center Squeeze
effect.
https://drive.google.com/file/d/1jIhFQfEoxSdyRz5SqEjZotbVDx4xshwM/view
I hope some of you guys might take a look at it, if you hadn't before.
--
r b-j . _ . _ . _ . _ rbj@audioimagination.com
"Imagination is more important than knowledge."
.
.
.
Election-Methods mailing list - see https://electorama.com/em for list
info