Fwd: (3) MJ -- The easiest method to 'tolerate'
3: A
20: A>B
25: B>A
3: B
20: C>B
29: C
C > B 49-48, C > A 49-48, B > A 48-23.
Top Ratings scores: C 49, B 28, A 23
Approval scores: B 68, C 49, A 48.
On 9/7/2016 7:09 AM, Jameson Quinn wrote:
IBIFA elects C. U/P elects B. It's clear to me that B is better. The 3
A-only voters are likely attempting a chicken strategy.
C: C is the clearly voted Condorcet winner and by far the most top-rated
candidate. The suggestion that "B is better" because of
some guess that 3 truncators are "likely" attempting some strategy is
absurd.
The best we can reasonably do about chicken strategy is just prevent (or
reduce the likelihood of) it from succeeding. If the 3 A voters' were
attempting a
"chicken" (defection) strategy against B, then electing C means it failed.
BTW, if we weaken all the candidates by 3 plumping votes each, thus:
20: A>B
25: B>A
20: C>B
26: C
then the U/P winner would change from B to C.
Chris Benham
On 9/7/2016 7:09 AM, Jameson Quinn wrote:
2016-09-06 16:07 GMT-04:00 C.Benham <cbenham@adam.com.au
mailto:cbenham@adam.com.au>:
On 9/7/2016 1:21 AM, Jameson Quinn wrote:
40: D>...>A,B
35: A>B>D
25: B>...>A,D
...
But then IBIFA elects B.
You're right. So, let's find a situation where they actually differ:
5: C>B>A,D
5: C>A>B,D
30: D>...>A,B,C
35: A>B>D,C
25: B>...>A,D,C
In this scenario, IBIFA elects A, and U/P (my new name for DA, as I
stated separately) elects B. I'd still argue that B is the right
answer... though I think this whole scenario (and any other I can
think of where the two systems differ) is getting pretty implausible,
so it's hardly worth arguing.
No candidate X's Top-Rating score is higher than any candidate Y's
approval (i.e. above-bottom rating) score on ballots that don't
top-rate X, so
IBIFA elects the most approved candidate, B.
Obviously, one could tweak these numbers to try to make one or
the other system look better or worse.
C: Well then I am very curious to see any example where IBIFA
"looks worse" than DA.
In your format:
3: A
20: A>B
25: B>A
3: B
20: C>B
29: C
IBIFA elects C. U/P elects B. It's clear to me that B is better. The 3
A-only voters are likely attempting a chicken strategy.
This is an artificial scenario designed to make IBIFA look bad. You
could do the same to make it look good. I don't think this proves
anything besides the fact that IBIFA is not strictly dominant over U/P.
No virus found in this message.
Checked by AVG - www.avg.com http://www.avg.com
Version: 2016.0.7752 / Virus Database: 4649/12956 - Release Date: 09/06/16
2016-09-07 8:17 GMT-04:00 C.Benham cbenham@adam.com.au:
3: A
20: A>B
25: B>A
3: B
20: C>B
29: C
C > B 49-48, C > A 49-48, B > A 48-23.
Top Ratings scores: C 49, B 28, A 23
Approval scores: B 68, C 49, A 48.
On 9/7/2016 7:09 AM, Jameson Quinn wrote:
IBIFA elects C. U/P elects B. It's clear to me that B is better. The 3
A-only voters are likely attempting a chicken strategy.
C: C is the clearly voted Condorcet winner
But not the majority Condorcet winner. I regard majority Condorcet as
something any good voting system should tend to respect, but non-majority
Condorcet as only a weak signal.
and by far the most top-rated candidate.
Whoops... I'm sorry. Those were the averages over a large number of polls,
but then a week before the election, A had a heart attack. The actual
scores, with 3 ballots still to count from A's former stronghold precinct,
are:
48: B
49: C
I guess C wasn't so dominant in top-ratings after all. Once we count those
last three ballots, we'll have a clear winner here, and there's every
chance it could be B.
The suggestion that "B is better" because of
some guess that 3 truncators are "likely" attempting some strategy is
absurd.
No, B is better because B has majority approval where C is rejected by a
majority. The argument about likely truncation is merely how IBIFA went
wrong, not how U/P got it right.
...Again, I don't regard this as a knockdown argument against IBIFA. It's
only one scenario; a roughly plausible one, but not particularly likely. I
think we should probably drop this. But I do think it demonstrates pretty
conclusively that IBIFA at least isn't strictly dominant. After all, as
long as we're making up arbitrary scenarios, we can arbitrarily assume that
the 3 A voters are in fact attempting strategy. Punishing strategy might
not be the worst thing in the world, but rather than punishing 48 voters
for the strategy of 3, it would be better to ignore the 3.
I'll write a broader response in a minute.
On 9/7/2016 10:15 PM, Jameson Quinn wrote:
C: C is the clearly voted Condorcet winner
But not the majority Condorcet winner. I regard majority Condorcet as
something any good voting system should tend to respect, but
non-majority Condorcet as only a weak signal.
C: The Majority Condorcet winner is always the Condorcet winner. It is
ridiculous to suggest that failing to elect a Condorcet winner is a step
in the
direction of "tending to respect the majority Condorcet winner".
The "Majority Condorcet" concept is vulnerable to Irrelevant Ballots,
and to put it forward as some especially desirable criterion is
nonsensical.
In our example, would removing 3 plumping ballots from each of the
candidates really turn C into a stronger candidate? Because the effect
would
be to change C from a mere Condorcet winner (who you say should lose to
the approval winner) into a fabulous Majority Condorcet winner (who you
say the method should "tend" to respect).
(The only point of the criterion is that some methods happen to be able
to meet it, but not the normal Condorcet criterion.)
20: A>B
25: B>A
20: C>B
26: C
91 ballots. Majority threshold = 46.
C > B 46-45, C > A 46-45.
Or of course the question can be posed the other way: does giving each
candidate 3 extra plumping ballots really make C a weaker candidate?
Whoops... I'm sorry. Those were the averages over a large number of
polls, but then a week before the election, A had a heart attack. The
actual scores, with 3 ballots still to count from A's former
stronghold precinct, are:
48: B
49: C
I guess C wasn't so dominant in top-ratings after all. Once we count
those last three ballots, we'll have a clear winner here, and there's
every chance it could be B.
C:I take it this is Jameson not being serious (and perhaps trying to
shift goal-posts).
One of IBIFA's obvious advantages over "U/P" and MJ and the other
Bucklin methods is that it has a
much weaker truncation incentive. In the example (with 100 ballots)
U/P penalised the 20 C>B voters
for not truncating. Had they done so C would have won. (Unsurprising,
since that would have made C positionally
dominant as well the Condorcet winner).
It seems that you want a quite complicated method with a nearly as
strong as possible truncation incentive and that nearly always
simply elects the Approval winner (while somehow "tending to respect"
majority Condorcet without actually meeting the Majority
Condorcet criterion).
BTW, I'd be mildly interested in seeing one or two of your realistic
examples where U/P doesn't elect the Approval winner.
Given from what I can make of your method goals, why don't you simply
propose Majority Condorcet//Approval ?
Chris Benham
2016-09-07 8:17 GMT-04:00 C.Benham <cbenham@adam.com.au
mailto:cbenham@adam.com.au>:
3: A
20: A>B
25: B>A
3: B
20: C>B
29: C
C > B 49-48, C > A 49-48, B > A 48-23.
Top Ratings scores: C 49, B 28, A 23
Approval scores: B 68, C 49, A 48.
On 9/7/2016 7:09 AM, Jameson Quinn wrote:
IBIFA elects C. U/P elects B. It's clear to me that B is better.
The 3 A-only voters are likely attempting a chicken strategy.
C: C is the clearly voted Condorcet winner
But not the majority Condorcet winner. I regard majority Condorcet as
something any good voting system should tend to respect, but
non-majority Condorcet as only a weak signal.
and by far the most top-rated candidate.
Whoops... I'm sorry. Those were the averages over a large number of
polls, but then a week before the election, A had a heart attack. The
actual scores, with 3 ballots still to count from A's former
stronghold precinct, are:
48: B
49: C
I guess C wasn't so dominant in top-ratings after all. Once we count
those last three ballots, we'll have a clear winner here, and there's
every chance it could be B.
The suggestion that "B is better" because of
some guess that 3 truncators are "likely" attempting some strategy
is absurd.
No, B is better because B has majority approval where C is rejected by
a majority. The argument about likely truncation is merely how IBIFA
went wrong, not how U/P got it right.
...Again, I don't regard this as a knockdown argument against IBIFA.
It's only one scenario; a roughly plausible one, but not particularly
likely. I think we should probably drop this. But I do think it
demonstrates pretty conclusively that IBIFA at least isn't strictly
dominant. After all, as long as we're making up arbitrary scenarios,
we can arbitrarily assume that the 3 A voters are in fact attempting
strategy. Punishing strategy might not be the worst thing in the
world, but rather than punishing 48 voters for the strategy of 3, it
would be better to ignore the 3.
I'll write a broader response in a minute.
No virus found in this message.
Checked by AVG - www.avg.com http://www.avg.com
Version: 2016.0.7752 / Virus Database: 4649/12961 - Release Date: 09/07/16
Once again, I'm going to stand behind my view that U/P is a better
practical reform proposal than IBIFA, and that there are cases where U/P
gets the right answer and IBIFA doesn't. But once again, I'm going to
preface it by saying I think IBIFA is a great method overall.
So. Remaining debates.
Condorcet versus Majority Condorcet: I do not believe in the concept of
"irrelevant ballots", or that we should "simplify" a scenario by removing a
balanced set of ballots.
Dominant in top-ratings: taking this as a strong argument for who the
correct winner is, is inherently subject to clone effects. My "unserious"
example was a (serious) attempt to show this.
"It seems that you want a quite complicated method with a nearly as strong
as possible truncation incentive and that nearly always
simply elects the Approval winner (while somehow "tending to respect"
majority Condorcet without actually meeting the Majority
Condorcet criterion)."
I am completely baffled by the idea that IBIFA is less complicated than
U/P. If you want to, we can each post a 3-minute video where we explain our
method to a random "person on the street" and then ask the person to
explain it back to us. I'll bet that this would show U/P as far simpler to
understand. If you agree, I'll let you set any reasonable procedure for
each choosing our subject/explainee.
I also don't agree that the U/P has a "nearly as strong as possible
truncation incentive" or "nearly always simply elects the approval winner".
On the first point: I think that in a chicken dilemma situation, U/P would
lead to less truncation than many other methods, including things like MCA
or MJ, because in U/P, chicken truncation is only helpful precisely insofar
as it's risky. On the second: I think that there is a not-insignificant
subset of voters who strongly want to cast a ballot that's expressive
enough to distinguish their favorite from their preferred frontrunner. In
approval, such people would bullet vote; in U/P, they could still rate the
compromise frontrunner as acceptable. This could easily lead to different
(and better) outcomes.
2016-09-07 15:01 GMT-04:00 C.Benham cbenham@adam.com.au:
On 9/7/2016 10:15 PM, Jameson Quinn wrote:
C: C is the clearly voted Condorcet winner
But not the majority Condorcet winner. I regard majority Condorcet as
something any good voting system should tend to respect, but non-majority
Condorcet as only a weak signal.
C: The Majority Condorcet winner is always the Condorcet winner. It is
ridiculous to suggest that failing to elect a Condorcet winner is a step in
the
direction of "tending to respect the majority Condorcet winner".
The "Majority Condorcet" concept is vulnerable to Irrelevant Ballots, and
to put it forward as some especially desirable criterion is nonsensical.
In our example, would removing 3 plumping ballots from each of the
candidates really turn C into a stronger candidate? Because the effect
would
be to change C from a mere Condorcet winner (who you say should lose to
the approval winner) into a fabulous Majority Condorcet winner (who you
say the method should "tend" to respect).
(The only point of the criterion is that some methods happen to be able to
meet it, but not the normal Condorcet criterion.)
20: A>B
25: B>A
20: C>B
26: C
91 ballots. Majority threshold = 46.
C > B 46-45, C > A 46-45.
Or of course the question can be posed the other way: does giving each
candidate 3 extra plumping ballots really make C a weaker candidate?
Whoops... I'm sorry. Those were the averages over a large number of polls,
but then a week before the election, A had a heart attack. The actual
scores, with 3 ballots still to count from A's former stronghold precinct,
are:
48: B
49: C
I guess C wasn't so dominant in top-ratings after all. Once we count those
last three ballots, we'll have a clear winner here, and there's every
chance it could be B.
C:I take it this is Jameson not being serious (and perhaps trying to shift
goal-posts).
One of IBIFA's obvious advantages over "U/P" and MJ and the other
Bucklin methods is that it has a
much weaker truncation incentive. In the example (with 100 ballots) U/P
penalised the 20 C>B voters
for not truncating. Had they done so C would have won. (Unsurprising,
since that would have made C positionally
dominant as well the Condorcet winner).
It seems that you want a quite complicated method with a nearly as strong
as possible truncation incentive and that nearly always
simply elects the Approval winner (while somehow "tending to respect"
majority Condorcet without actually meeting the Majority
Condorcet criterion).
BTW, I'd be mildly interested in seeing one or two of your realistic
examples where U/P doesn't elect the Approval winner.
Given from what I can make of your method goals, why don't you simply
propose Majority Condorcet//Approval ?
Chris Benham
2016-09-07 8:17 GMT-04:00 C.Benham cbenham@adam.com.au:
3: A
20: A>B
25: B>A
3: B
20: C>B
29: C
C > B 49-48, C > A 49-48, B > A 48-23.
Top Ratings scores: C 49, B 28, A 23
Approval scores: B 68, C 49, A 48.
On 9/7/2016 7:09 AM, Jameson Quinn wrote:
IBIFA elects C. U/P elects B. It's clear to me that B is better. The 3
A-only voters are likely attempting a chicken strategy.
C: C is the clearly voted Condorcet winner
But not the majority Condorcet winner. I regard majority Condorcet as
something any good voting system should tend to respect, but non-majority
Condorcet as only a weak signal.
and by far the most top-rated candidate.
Whoops... I'm sorry. Those were the averages over a large number of polls,
but then a week before the election, A had a heart attack. The actual
scores, with 3 ballots still to count from A's former stronghold precinct,
are:
48: B
49: C
I guess C wasn't so dominant in top-ratings after all. Once we count those
last three ballots, we'll have a clear winner here, and there's every
chance it could be B.
The suggestion that "B is better" because of
some guess that 3 truncators are "likely" attempting some strategy is
absurd.
No, B is better because B has majority approval where C is rejected by a
majority. The argument about likely truncation is merely how IBIFA went
wrong, not how U/P got it right.
...Again, I don't regard this as a knockdown argument against IBIFA. It's
only one scenario; a roughly plausible one, but not particularly likely. I
think we should probably drop this. But I do think it demonstrates pretty
conclusively that IBIFA at least isn't strictly dominant. After all, as
long as we're making up arbitrary scenarios, we can arbitrarily assume that
the 3 A voters are in fact attempting strategy. Punishing strategy might
not be the worst thing in the world, but rather than punishing 48 voters
for the strategy of 3, it would be better to ignore the 3.
I'll write a broader response in a minute.
No virus found in this message.
Checked by AVG - www.avg.com
Version: 2016.0.7752 / Virus Database: 4649/12961 - Release Date: 09/07/16
On 9/8/2016 5:01 AM, Jameson Quinn wrote:
I am completely baffled by the idea that IBIFA is less complicated
than U/P.
C: I never said that it was. I just said that U/P is "quite
complicated", meaning that it is a lot more complicated than Approval
(and more complicated than MTA). Its extra complexity over
those methods doesn't seem to buy much.
if you include in U/P 's explanation all the palaver about the default
ratings it becomes quite long and complex. 3-slot IBIFA can be defined
fairly succinctly, but I'm
not sure how well the "man in the street" would understand it.
I also don't agree that .. U/P "nearly always simply elects the
Approval winner".
C: Again, I'd be interested in seeing a plausible example of when it
doesn't.
Condorcet versus Majority Condorcet: I do not believe in the concept
of "irrelevant ballots", or that we should "simplify" a scenario by
removing a balanced set of ballots.
C: Why not?
Dominant in top-ratings: taking this as a strong argument for who the
correct winner is, is inherently subject to clone effects
C: I never used the phrase "dominant in top-ratings". These are
supposed to be ratings (instead of simply rankings). In that case clones
are technically supposed to be rated the same.
At one point I said that C would be "positionally dominant" if all C's
supporters truncated, meaning C would have both the highest Top Rating
score and the highest Approval score.
After all, as long as we're making up arbitrary scenarios, we can
arbitrarily assume that the 3 A voters are in fact attempting strategy.
C: No, you have to be able to justify the result on the assumption that
all the votes are sincere, or at least are all equally likely to be
sincere. Otherwise you'd be able to argue
that in this or that scenario the most ridiculous method "gave the best
result".
Chris Benham
Once again, I'm going to stand behind my view that U/P is a better
practical reform proposal than IBIFA, and that there are cases where
U/P gets the right answer and IBIFA doesn't. But once again, I'm going
to preface it by saying I think IBIFA is a great method overall.
So. Remaining debates.
Condorcet versus Majority Condorcet: I do not believe in the concept
of "irrelevant ballots", or that we should "simplify" a scenario by
removing a balanced set of ballots.
Dominant in top-ratings: taking this as a strong argument for who the
correct winner is, is inherently subject to clone effects. My
"unserious" example was a (serious) attempt to show this.
"It seems that you want a quite complicated method with a nearly as
strong as possible truncation incentive and that nearly always
simply elects the Approval winner (while somehow "tending to respect"
majority Condorcet without actually meeting the Majority
Condorcet criterion)."
I am completely baffled by the idea that IBIFA is less complicated
than U/P. If you want to, we can each post a 3-minute video where we
explain our method to a random "person on the street" and then ask the
person to explain it back to us. I'll bet that this would show U/P as
far simpler to understand. If you agree, I'll let you set any
reasonable procedure for each choosing our subject/explainee.
I also don't agree that the U/P has a "nearly as strong as possible
truncation incentive" or "nearly always simply elects the approval
winner".
On the first point: I think that in a chicken dilemma situation, U/P
would lead to less truncation than many other methods, including
things like MCA or MJ, because in U/P, chicken truncation is only
helpful precisely insofar as it's risky. On the second: I think that
there is a not-insignificant subset of voters who strongly want to
cast a ballot that's expressive enough to distinguish their favorite
from their preferred frontrunner. In approval, such people would
bullet vote; in U/P, they could still rate the compromise frontrunner
as acceptable. This could easily lead to different (and better) outcomes.
2016-09-07 15:01 GMT-04:00 C.Benham <cbenham@adam.com.au
mailto:cbenham@adam.com.au>:
On 9/7/2016 10:15 PM, Jameson Quinn wrote:
C: C is the clearly voted Condorcet winner
But not the majority Condorcet winner. I regard majority
Condorcet as something any good voting system should tend to
respect, but non-majority Condorcet as only a weak signal.
C: The Majority Condorcet winner is always the Condorcet winner.
It is ridiculous to suggest that failing to elect a Condorcet
winner is a step in the
direction of "tending to respect the majority Condorcet winner".
The "Majority Condorcet" concept is vulnerable to Irrelevant
Ballots, and to put it forward as some especially desirable
criterion is nonsensical.
In our example, would removing 3 plumping ballots from each of the
candidates really turn C into a stronger candidate? Because the
effect would
be to change C from a mere Condorcet winner (who you say should
lose to the approval winner) into a fabulous Majority Condorcet
winner (who you
say the method should "tend" to respect).
(The only point of the criterion is that some methods happen to be
able to meet it, but not the normal Condorcet criterion.)
20: A>B
25: B>A
20: C>B
26: C
91 ballots. Majority threshold = 46.
C > B 46-45, C > A 46-45.
Or of course the question can be posed the other way: does giving
each candidate 3 extra plumping ballots really make C a weaker
candidate?
Whoops... I'm sorry. Those were the averages over a large number
of polls, but then a week before the election, A had a heart
attack. The actual scores, with 3 ballots still to count from A's
former stronghold precinct, are:
48: B
49: C
I guess C wasn't so dominant in top-ratings after all. Once we
count those last three ballots, we'll have a clear winner here,
and there's every chance it could be B.
C:I take it this is Jameson not being serious (and perhaps trying
to shift goal-posts).
One of IBIFA's obvious advantages over "U/P" and MJ and the
other Bucklin methods is that it has a
much weaker truncation incentive. In the example (with 100
ballots) U/P penalised the 20 C>B voters
for not truncating. Had they done so C would have won.
(Unsurprising, since that would have made C positionally
dominant as well the Condorcet winner).
It seems that you want a quite complicated method with a nearly as
strong as possible truncation incentive and that nearly always
simply elects the Approval winner (while somehow "tending to
respect" majority Condorcet without actually meeting the Majority
Condorcet criterion).
BTW, I'd be mildly interested in seeing one or two of your
realistic examples where U/P doesn't elect the Approval winner.
Given from what I can make of your method goals, why don't you
simply propose Majority Condorcet//Approval ?
Chris Benham
2016-09-07 8:17 GMT-04:00 C.Benham <cbenham@adam.com.au
<mailto:cbenham@adam.com.au>>:
3: A
20: A>B
25: B>A
3: B
20: C>B
29: C
C > B 49-48, C > A 49-48, B > A 48-23.
Top Ratings scores: C 49, B 28, A 23
Approval scores: B 68, C 49, A 48.
On 9/7/2016 7:09 AM, Jameson Quinn wrote:
IBIFA elects C. U/P elects B. It's clear to me that B is
better. The 3 A-only voters are likely attempting a chicken
strategy.
C: C is the clearly voted Condorcet winner
But not the majority Condorcet winner. I regard majority
Condorcet as something any good voting system should tend to
respect, but non-majority Condorcet as only a weak signal.
and by far the most top-rated candidate.
Whoops... I'm sorry. Those were the averages over a large number
of polls, but then a week before the election, A had a heart
attack. The actual scores, with 3 ballots still to count from A's
former stronghold precinct, are:
48: B
49: C
I guess C wasn't so dominant in top-ratings after all. Once we
count those last three ballots, we'll have a clear winner here,
and there's every chance it could be B.
The suggestion that "B is better" because of
some guess that 3 truncators are "likely" attempting some
strategy is absurd.
No, B is better because B has majority approval where C is
rejected by a majority. The argument about likely truncation is
merely how IBIFA went wrong, not how U/P got it right.
...Again, I don't regard this as a knockdown argument against
IBIFA. It's only one scenario; a roughly plausible one, but not
particularly likely. I think we should probably drop this. But I
do think it demonstrates pretty conclusively that IBIFA at least
isn't strictly dominant. After all, as long as we're making up
arbitrary scenarios, we can arbitrarily assume that the 3 A
voters are in fact attempting strategy. Punishing strategy might
not be the worst thing in the world, but rather than punishing 48
voters for the strategy of 3, it would be better to ignore the 3.
I'll write a broader response in a minute.
No virus found in this message.
Checked by AVG - www.avg.com <http://www.avg.com>
Version: 2016.0.7752 / Virus Database: 4649/12961 - Release Date:
09/07/16
No virus found in this message.
Checked by AVG - www.avg.com http://www.avg.com
Version: 2016.0.7752 / Virus Database: 4649/12966 - Release Date: 09/07/16