Re: [EM] MAM vs Schulze
(Replying farther down)
On Oct 6, 2016 2:14 AM, "Toby Pereira" tdp201b@yahoo.co.uk wrote:
I agree. I don't find it compelling at all. For any deterministic
Condorcet method, I could devise another one where the winner pairwise
beats the winner of that one more often than vice versa. Someone could have
a method they call BEST METHOD. Then all I have to do is say under my new
method, elect the Condorcet winner if there is one. If there isn't, elect a
candidate that pairwise beats the winner using BEST METHOD, if there is one
(pick at random if there's more than one). Otherwise just pick the same
winner as BEST METHOD.
(endquote)
Sorry, no good.
MAM's winner doesn't beat Schulze's winner in that contrived manner.
The MAM winner beats the Schulze winner for a simple, obvious reason:
MAM doesn't disregard a defeat unnecessarily or without obvious, compelling
justification. Schulze does.
Look at the brief, simple, natural & obvious MAM definition that I posted.
Michael Ossipoff
From: C.Benham cbenham@adam.com.au
To: election-methods@lists.electorama.com
Sent: Thursday, 6 October 2016, 4:26
Subject: Re: [EM] MAM vs Schulze
Marcus,
That chance of that happening in a real public election is close enough
to zero, so therefore "MAM versus Shulze"
strikes me as pointless.
And if it didn't I wouldn't find the argument that one's winner pairwise
beats the other's a small proportion of times more
than vice versa very compelling.
Chris Benham
Election-Methods mailing list - see http://electorama.com/em for list info
On 10/9/2016 6:25 AM, Kristofer Munsterhjelm wrote:
If you had to pick a method passing Minimal Defense but not using
Approval information, which would it be? (That is, do you have any
favorites in that category, and if so, which?)
Kristopher,
Since I am happy to interpret above-bottom ranking (or non-truncation)
as "approval",
I don't see any practical or theoretical advantage in "not using
Approval information" so
that isn't one of my categories.
Minimal Defense is incompatible with Chicken Dilemma. The choice between
those two
corresponds with the "aesthetic" choice between focussing on tops of the
ballots versus
the bottoms.
I presently can't think of any method in your suggested category apart
from Winning Votes,
but some ok alternative could perhaps be invented.
A pure pairwise criterion that implies Minimal Defense could be:
If X's pairwise score versus Y is larger than Y's largest pairwise
score, then Y can't win.
Chris Benham
On 10/08/2016 03:06 PM, C.Benham wrote:
Mike,
As far as I can tell, for all intents and purposes MAM, Schulze, River
and Smith//MinMax (wv) are all just different wordings
of the same method.
If you think that MAM is better than Shulze, then what criterion (that
we might care about) is met by MAM and not Shulze?
Or perhaps you have some example in mind where you think the MAM winner
is much prettier than the Schulze winner?
I'm not Mike (and I don't see his posts), but I would probably say LIIA
for MAM and IPDA/ISDA for River. Those criteria probably only become
relevant (that is, failed by Schulze) when the Smith set size is large,
however, so whether you consider them important would depend in part on
whether you think large Smith sets may occur in real elections. On the
one hand, they haven't so far; on the other, adoption of a Condorcet
method might lead to a richer political landscape where they could.
MAM and River may also be simpler to explain than Schulze, since a brief
description of beatpaths require recursion, and recursion is a tricky
concept to get if you haven't already been exposed to it.
MAM's brief definition just says:
A defeat is affirmed if it isn't the weakest defeat in a cycle whose
other defeats are affirmed.
How about, for MAM:
Sort defeats in order from strongest to weakest. Going from strongest to
weakest, affirm a defeat unless it forms a cycle with earlier affirmed
defeats. The candidate who isn't beaten in any affirmed defeat wins.
Schulze:
A has a beatpath to B if A beats B or A beats someone who has a beatpath
to B. The strength of a beatpath is defined by the strength of its
weakest defeat. The candidate who, for every other candidate has a
stronger beatpath to that candidate than that candidate has to him, wins.
Neither definition goes into the subtlety of tiebreakers (random voter
hierarchy for MAM) nor margins vs wv.
If you want a method that (like WV) meets Minimal Defense then I prefer
Forest's "Max Covered Approval" (which would nearly always be equivalent
to Smith//Approval, which I also like.)
If you had to pick a method passing Minimal Defense but not using
Approval information, which would it be? (That is, do you have any
favorites in that category, and if so, which?)
Election-Methods mailing list - see http://electorama.com/em for list info
No virus found in this message.
Checked by AVG - www.avg.com
Version: 2016.0.7797 / Virus Database: 4664/13168 - Release Date: 10/07/16
Mike,
When it comes to Condorcet methods, I think that (up to a point) some
truncation incentive (and so some "vulnerability to
truncation") is somewhere between a good thing and a relatively small
necessary evil ("necessary" to avoid greater evil).
The strategy situation of the methods you listed isn't as good as Bucklin.
Surely the purpose of a pairwise-count method is to improve on Bucklin.
C: The methods I listed all meet Smith and Bucklin doesn't. Under
Bucklin all voters have such a strong truncation incentive that
the method is more-or-less strategically equivalent to Approval.
Under Benham and LV(erw) SME informed strategists have a much weaker
truncation incentive and in the zero-info. case there
is no truncation incentive.
Given that burial vulnerability is unavoidable in Condorcet methods, I
think that is more democratic if (in this respect) larger factions
have the advantage over smaller factions.
43: A
03: A>B
44: B>C (sincere is B or B>A)
10: C
C>A 54-46, A>B 46-44, B>C 47-10.
Here A is the sincere CW and supported by the largest of the three
factions of voters, but Winning Votes rewards the buriers by electing B.
Benham and LV(erw)SME easily elect A. Smith//Approval elects C.
Smith//Approval shares the great vulnerability to truncation & burial.
C: Obviously the supporters of the sincere CW have much less truncation
incentive under Smith//Approval than they do under Bucklin,
so I wonder what example you have in mind.
40: A>B
35: B
25: C
The Condorcet winner is A, but under Bucklin the A supporters' failure
to truncate gives the win to B.
Chris Benham
On 10/9/2016 7:50 AM, Michael Ossipoff wrote:
On Oct 8, 2016 6:06 AM, "C.Benham" <cbenham@adam.com.au
mailto:cbenham@adam.com.au> wrote:
Mike,
As far as I can tell, for all intents and purposes MAM, Schulze,
River and Smith//MinMax (wv) are all just different wordings
of the same method.
No. They sometimes choose different winners.
If you think that MAM is better than Shulze, then what criterion
(that we might care about) is met by MAM and not Shulze?
Sometimes they choose the same, sometimes they don't.
When they don't, the MAM winner is publicly preferred to the Schulze
winner several times more often than vice-versa.
(It seems to me that it might have been something like 4 to 1, or 5 to
- Steve Eppley would be the one to ask.)
So: Choose in keeping with public preference, or contrary to it. Your
choice
MAM's brief, natural & obvious definition is the opposite of the
arbitrary definition of Schulze or CSSD.
MAM's definition clearly is the one that doesn't unnecessarily
disregard a defeat.
It disregards a defeat only it's the weakest in a cycle with defeats
for which there isn't
justification to disregard them.
And, as I said, it's no surprise when unnecessarily disregarding
defeats results in a winner to whom the public prefer the MAM winner.
Or perhaps you have some example in mind where you think the MAM
winner is much prettier than the Schulze winner?
Publicly-preferred is prettier.
Minimally disregarding defeats only with obviousl, strong
justification, never unnecessarily disregarding a defeat--That's prettier.
MAM's brief definition just says:
A defeat is affirmed if it isn't the weakest defeat in a cycle
whose other defeats are affirmed.
C: Is that definition fully adequate?
Yes.
You wrote:
It doesn't tell you where to start.
It isn't a procedural definition or a count-instruction. It's a brief
recursive definition.
Given a set of rankings, it fully and definitely specifies a set of
affirmed defeats, & a set of not-affirmed defeats.
...and fully specifies the winner.
For a procedure:
Write down the strongest defeat.
Below it, write down the next strongest defeat.
Below that, write down the next strongest defeat, if it doesn't cycle
with defeats already written down.
Repeat the paragraph before this one, until all the defeats have been
considered as described in that paragraph.
A candidate wins if s/he has no written-down defeats.
(end of count instruction)
So, if it will be rare for them to differ, does that mean that we
should propose the more complicatedly-worded, elaborately- worded one?
...the less obviously, naturally and clearly motivated & justified one?
C: Recently you accepted that Winning Votes is at best "maybe a
bit questionable", so why do you think that we should "propose" either?
(endquote)
I said they might be iffy or questionable. I didn't say they're ruled out.
For that questionable-ness, you get a chance for much better strategy.
...at the cost of the possibility of the strategic mess of the
perpetual-burial fiasco.
I'd say that MAM, Smith//MMPO, & plain MMPO are worth a try.
They should be included in a proposal that lists a number of suggested
methods.
In particular, for an unlimited-rankings method, Plain MMPO offers the
most, for current conditions.
If you want a Condorcet method that meets Chicken Dilemma then I
prefer both "Benham" and Losing Votes (erw) Sorted Margins Elimination.
(endquote)
They're far too vulnerable to truncation & burial.
In WV & MMPO, truncation just doesn't work. The CWs still wins.
The strategy situation of the methods you listed isn't as good as Bucklin.
Surely the purpose of a pairwise-count method is to improve on Bucklin.
In Bucklin & Approval, the CWs's preferrers can protect hir win by
plumping.
In Benham, Woodall, & Margins-Sorted LV Elimination, the best they can
do is:
Say it's Worst (W), Middle (M), & Favorite (F).
M is the middle CWs.
The M voters could estimate or look up the expected sizes of the W & F
factions. ...& rank one over the other probabilistically, so that
each one's probability of pair-beating the other is 50%.
...so that burial has a 50% chance of backfiring.
In Bucklin, WV or MMPO, they need merely to plump.
Or the F voters could rank M alone in 1st place. That would work in
IRV too (nothing else would).
As I said, surely the purpose of a pairwise-count method is to
improve on Bucklin.
If you want a method that (like WV) meets Minimal Defense then I
prefer Forest's "Max Covered Approval" (which would nearly always be
equivalent
to Smith//Approval, which I also like.)
Smith//Approval shares the great vulnerability to truncation & burial.
Michael Ossipoff
Chris Benham
What do you mean by unnecessarily disregarding a defeat?
From: Michael Ossipoff <email9648742@gmail.com>
To: Toby Pereira tdp201b@yahoo.co.uk; election-methods@electorama.com
Sent: Saturday, 8 October 2016, 23:15
Subject: Re: [EM] MAM vs Schulze
(Replying farther down)On Oct 6, 2016 2:14 AM, "Toby Pereira" tdp201b@yahoo.co.uk wrote:
I agree. I don't find it compelling at all. For any deterministic Condorcet method, I could devise another one where the winner pairwise beats the winner of that one more often than vice versa. Someone could have a method they call BEST METHOD. Then all I have to do is say under my new method, elect the Condorcet winner if there is one. If there isn't, elect a candidate that pairwise beats the winner using BEST METHOD, if there is one (pick at random if there's more than one). Otherwise just pick the same winner as BEST METHOD.(endquote)Sorry, no good.MAM's winner doesn't beat Schulze's winner in that contrived manner.The MAM winner beats the Schulze winner for a simple, obvious reason:MAM doesn't disregard a defeat unnecessarily or without obvious, compelling justification. Schulze does.Look at the brief, simple, natural & obvious MAM definition that I posted.Michael Ossipoff
I've never been able to work out why local independence of irrelevant alternatives (LIIA) is even something we should care about. It seems to be a criterion for the sake of a criterion.
From: Kristofer Munsterhjelm <km_elmet@t-online.de>
To: election-methods@electorama.com
Sent: Saturday, 8 October 2016, 20:55
Subject: Re: [EM] MAM vs Schulze
On 10/08/2016 03:06 PM, C.Benham wrote:
Mike,
As far as I can tell, for all intents and purposes MAM, Schulze, River
and Smith//MinMax (wv) are all just different wordings
of the same method.
If you think that MAM is better than Shulze, then what criterion (that
we might care about) is met by MAM and not Shulze?
Or perhaps you have some example in mind where you think the MAM winner
is much prettier than the Schulze winner?
I'm not Mike (and I don't see his posts), but I would probably say LIIA
for MAM and IPDA/ISDA for River.
On Oct 9, 2016 11:01 AM, "Toby Pereira" tdp201b@yahoo.co.uk wrote:
What do you mean by unnecessarily disregarding a defeat?
I discuss that in my most recent reply to Markus.
Michael Ossipoff ________________________________
From: Michael Ossipoff email9648742@gmail.com
To: Toby Pereira tdp201b@yahoo.co.uk; election-methods@electorama.com
Sent: Saturday, 8 October 2016, 23:15
Subject: Re: [EM] MAM vs Schulze
(Replying farther down)
On Oct 6, 2016 2:14 AM, "Toby Pereira" tdp201b@yahoo.co.uk wrote:
I agree. I don't find it compelling at all. For any deterministic
Condorcet method, I could devise another one where the winner pairwise
beats the winner of that one more often than vice versa. Someone could have
a method they call BEST METHOD. Then all I have to do is say under my new
method, elect the Condorcet winner if there is one. If there isn't, elect a
candidate that pairwise beats the winner using BEST METHOD, if there is one
(pick at random if there's more than one). Otherwise just pick the same
winner as BEST METHOD.
(endquote)
Sorry, no good.
MAM's winner doesn't beat Schulze's winner in that contrived manner.
The MAM winner beats the Schulze winner for a simple, obvious reason:
MAM doesn't disregard a defeat unnecessarily or without obvious,
compelling justification. Schulze does.
Look at the brief, simple, natural & obvious MAM definition that I
posted.
Michael Ossipoff
(Replying farther down)
On Oct 9, 2016 5:47 AM, "C.Benham" cbenham@adam.com.au wrote:
Mike,
When it comes to Condorcet methods, I think that (up to a point) some
truncation incentive (and so some "vulnerability to
truncation") is somewhere between a good thing and a relatively small
necessary evil ("necessary" to avoid greater evil).
The strategy situation of the methods you listed isn't as good as
Bucklin.
Surely the purpose of a pairwise-count method is to improve on Bucklin.
C: The methods I listed all meet Smith and Bucklin doesn't. Under
Bucklin all voters have such a strong truncation incentive that
the method is more-or-less strategically equivalent to Approval.
(endquote)
Yes. ...unless you're at least fairly sure that you're majority-favored, in
which case you benefit from Bucklin's MMC.
Bucklin isn't fancy, complex or deluxe. It's just solid & reliable, like
Approval, on which it's based.
But yes, I'd rather just have Approval. But there may be many voters who
need it want rankings or MMC.
But it might often be difficult to know if you're majority-favored (& thus
in a position to benefit from MMC).
Because not everyone is MF (majority favored), I'd just prefer Approval.
Regarding that truncation incentive to which you refer:
That's how the CWs's voters can protect the CWs's win.
But it doesn't help in Benham, Woodall, or Margins-Sorted LV Elimination.
With those methods, the only way for the CWs's preferrers to protect
hir from burial or truncation, us the probabilistic strategy that I
described.
That's why I said that those methods' strategy situation is worse than that
of Bucklin.
Under Benham and LV(erw) SME informed strategists have a much weaker
truncation incentive and in the zero-info. case there
is no truncation incentive.
With 0-info, use Approval, & approve your top-set. In fact, do so anyway,
because predictive information is unreliable.
Given that burial vulnerability is unavoidable in Condorcet methods, I
think that is more democratic if (in this respect) larger factions
have the advantage over smaller factions.
I'll check out your example.
43: A
03: A>B
44: B>C (sincere is B or B>A)
10: C
C>A 54-46, A>B 46-44, B>C 47-10.
Here A is the sincere CW and supported by the largest of the three
factions of voters, but Winning Votes rewards the buriers by electing B.
Benham and LV(erw)SME easily elect A. Smith//Approval elects C.
Smith//Approval shares the great vulnerability to truncation & burial.
C: Obviously the supporters of the sincere CW have much less truncation
incentive under Smith//Approval than they do under Bucklin,
so I wonder what example you have in mind.
Any example in which the CWs is truncated from one side.
40: A>B
35: B
25: C
The Condorcet winner is A, but under Bucklin the A supporters' failure
to truncate gives the win to B.
Yes, Bucklin is a simple, solid Approval method, whose strategy involves
truncation.
WV & MMPO try for better, by making the truncation not always necessary.
Maybe those methods won't have the perpetual burial fiasco.
The methods you describe need a lot more than defensive truncation.
Michael Ossipoff
Chris Benham
On 10/9/2016 7:50 AM, Michael Ossipoff wrote:
On Oct 8, 2016 6:06 AM, "C.Benham" cbenham@adam.com.au wrote:
Mike,
As far as I can tell, for all intents and purposes MAM, Schulze,
River and Smith//MinMax (wv) are all just different wordings
of the same method.
No. They sometimes choose different winners.
If you think that MAM is better than Shulze, then what criterion
(that we might care about) is met by MAM and not Shulze?
Sometimes they choose the same, sometimes they don't.
When they don't, the MAM winner is publicly preferred to the Schulze
winner several times more often than vice-versa.
(It seems to me that it might have been something like 4 to 1, or 5 to
- Steve Eppley would be the one to ask.)
So: Choose in keeping with public preference, or contrary to it. Your
choice
MAM's brief, natural & obvious definition is the opposite of the
arbitrary definition of Schulze or CSSD.
MAM's definition clearly is the one that doesn't unnecessarily disregard
a defeat.
It disregards a defeat only it's the weakest in a cycle with defeats for
which there isn't
justification to disregard them.
And, as I said, it's no surprise when unnecessarily disregarding defeats
results in a winner to whom the public prefer the MAM winner.
Or perhaps you have some example in mind where you think the MAM
winner is much prettier than the Schulze winner?
Publicly-preferred is prettier.
Minimally disregarding defeats only with obviousl, strong justification,
never unnecessarily disregarding a defeat--That's prettier.
MAM's brief definition just says:
A defeat is affirmed if it isn't the weakest defeat in a cycle whose
other defeats are affirmed.
C: Is that definition fully adequate?
Yes.
You wrote:
It doesn't tell you where to start.
It isn't a procedural definition or a count-instruction. It's a brief
recursive definition.
Given a set of rankings, it fully and definitely specifies a set of
affirmed defeats, & a set of not-affirmed defeats.
...and fully specifies the winner.
For a procedure:
Write down the strongest defeat.
Below it, write down the next strongest defeat.
Below that, write down the next strongest defeat, if it doesn't cycle
with defeats already written down.
Repeat the paragraph before this one, until all the defeats have been
considered as described in that paragraph.
A candidate wins if s/he has no written-down defeats.
(end of count instruction)
So, if it will be rare for them to differ, does that mean that we
should propose the more complicatedly-worded, elaborately- worded one?
...the less obviously, naturally and clearly motivated & justified
one?
C: Recently you accepted that Winning Votes is at best "maybe a bit
questionable", so why do you think that we should "propose" either?
(endquote)
I said they might be iffy or questionable. I didn't say they're ruled
out.
For that questionable-ness, you get a chance for much better strategy.
...at the cost of the possibility of the strategic mess of the
perpetual-burial fiasco.
I'd say that MAM, Smith//MMPO, & plain MMPO are worth a try.
They should be included in a proposal that lists a number of suggested
methods.
In particular, for an unlimited-rankings method, Plain MMPO offers the
most, for current conditions.
If you want a Condorcet method that meets Chicken Dilemma then I
prefer both "Benham" and Losing Votes (erw) Sorted Margins Elimination.
(endquote)
They're far too vulnerable to truncation & burial.
In WV & MMPO, truncation just doesn't work. The CWs still wins.
The strategy situation of the methods you listed isn't as good as
Bucklin.
Surely the purpose of a pairwise-count method is to improve on Bucklin.
In Bucklin & Approval, the CWs's preferrers can protect hir win by
plumping.
In Benham, Woodall, & Margins-Sorted LV Elimination, the best they can
do is:
Say it's Worst (W), Middle (M), & Favorite (F).
M is the middle CWs.
The M voters could estimate or look up the expected sizes of the W & F
factions. ...& rank one over the other probabilistically, so that each
one's probability of pair-beating the other is 50%.
...so that burial has a 50% chance of backfiring.
In Bucklin, WV or MMPO, they need merely to plump.
Or the F voters could rank M alone in 1st place. That would work in IRV
too (nothing else would).
As I said, surely the purpose of a pairwise-count method is to improve
on Bucklin.
If you want a method that (like WV) meets Minimal Defense then I
prefer Forest's "Max Covered Approval" (which would nearly always be
equivalent
to Smith//Approval, which I also like.)
Smith//Approval shares the great vulnerability to truncation & burial.
Michael Ossipoff
Chris Benham