[EM] Voting Matters, Issue 17
Markus Schulze
markus.schulze at alumni.tu-berlin.de
Tue Dec 16 16:22:31 PST 2003
Hallo,
I forward a conversation between Craig Carey and me about
issue 17 of Voting Matters.
**********************************************************
> To: single-transferable-vote at yahoogroups.com
> From: Craig Carey <research at ijs.co.nz>
> Date: Mon, 15 Dec 2003 23:09:23 +1300
> Subject: Voting Matters, Issue 17, of McDougall Trust
>
> http://lists.electorama.com/pipermail/election-methods-electorama.com/2003-December/011387.html
>
> http://groups.yahoo.com/group/election-methods-list/message/12838
>
> --->
>
> _______________________________________________________________________
>
> >[EM] Voting Matters, Issue 17
> >Markus Schulze markus.schulze at alumni.tu-berlin.de
> >Sun Dec 14 07:30:01 2003
> >
> >...
> >I have uploaded issue 17 (October 2003) of "Voting Matters" (208 kB):
> >http://groups.yahoo.com/group/election-methods-list/files/issue17.pdf
> >
> >This issue contains the following papers:
> >
> >1. Douglas R. Woodall, "QPQ, a quota-preferential STV-like
> > election rule," page 1-7,
> >
> >2. Joe Otten, "Fuller Disclosure than Intended," page 8,
> >
> >3. Markus Schulze, "A New Monotonic and Clone-Independent
> > Single-Winner Election Method," page 9-19,
> >
> >4. James Gilmour, "Calculation of Transfer Values -- Proposal
> > for STV-PR Rules for Local Government Elections in Scotland,"
> > page 20-24.
> >
> >Markus Schulze
> _______________________________________________________________________
>
>
>
> Just in case members needed a review here is one
>
>
> (4) ---------------- [from below]
>
> I agree with the with the view of Mr Gilmour that more than 2
> significant digits is needed when computing STV.
>
> * using only 2 decimal digits is arbitrary. I was assuming
> that arbitrariness is a way a line of consideration that
> can lead to rejection of a preferential voting method.
>
> Possibly a superior idea to
> * getting the correct leaders elected (i.e. "accuracy")
> is:
> * avoiding the arbitrariness that occurs when the winners
> were wrong but there are other ways to get the winners
> wrong. So the reasoning was inadequate.
>
> Once replacing need-for-accuracy with a lack-of=-arbitrariness,
> then to that can be added this idea:
> * politicians might be right to reject a preferential voting
> if it is arbitrary
>
> Minimal arbitrariness can allow the least popular candidate to
> be the winner. Politicians can't use the word accuracy
> when talking about winners if not knowing who the correct
> winners are. They just won't do the maths of casting
> n-D shadows which is the best possible so not arbitrary.
>
>
>
> (1) ----------------
>
> (a) Mr Otten says that STV papers be kept secret. That article
> fails to be very clear on when. Governments could have their
> statistian produce hard to believe stroies on how it would be
> expensive to anonymize the ballot paper preferences. It could
> cost between 100 and 200 dollars.
>
> (b) Surely before the election the papers are already secret.
> Mr Wichmann missed these unclear parts of the paper:
> (b)(1) before vs after ?: when is secrecy needed ?
> The text of Mr Wichmann at the top suggests that secrecy
> maybe would occur long after the election is finished:
>
> "The paper of Joe Otten arose from a
> resolution put to the ERS AGM requesting
> that the full election data of the
> preferences specified should be available
> for STV elections. (Such disclosure was
> available for the three Irish constituencies
> for which electronic voting was em-ployed in
> the June 2002 elections.) The paper explains
> a potential danger from full disclosure with
> a proposed resolution."
>
> * who is the wrong doer that would be thwarted by
> secrecy over what is on STV ballot papers?.
>
>
> (c)
> STV allows advantage to be gained by saying how to vote.
> The gain could be huge.
> But offender is the algorithm: it would divert votes from a
> possible winner to a 3rd candidate while simultaneously
> actually keeping the power under control. Elections are one
> process where the prime aim of people is to manipulate the
> results. So what is wrong with a candidate manipulating ?.
> Similarly the targetting of enemies is allowed. Suppose a
> Tory leader tripped and was zoomed in on by TV cameras after
> a Labour backbencher winked and whistled.
>
> Further suppose the Labour MP was not a corrupt person
> and thus not barred to national politics. It he is not
> guilty of corruptly tripping the man up, then it was
> excessively manipulative. But the aim in elections is for
> every voter to be maximally manipulative.
>
> A good reason from withholding information would be to
> stop wrongdoing by an individual. Maybe society gets harmed
> by the manipulation that STV could allow. But perhaps no
> court or judge has restarted an election because it was
> snowing on election day or etc.. A solution could be to
> have a Green say
> "You need and now want STV. Wrong results will follow
> from the use of dumb preferential voting algorithms".
>
>
>
> (2) ----------------
>
> I have not looked at Mr Woodall's method.
>
> It could be tested numerically.
> E.g. this way:
> (1) for both 3 candidates and 4 candidates, it is better
> in the 2 cases, or worse. Then maybe a 5 candidate
> test could be put off.
> (2) 5 candidate testing is done and some more strict
> rules may be needed. That would involve days of
> programming and more.
>
> I have already defined the fairness polytope and its
> dual for the 1 winner 4 candidate election. Porta
> can compute dual polytopes. That would be extended
> for 2w4c,1w5c,2w5c,3w5c,4w5c cases, Then hand
> steering of 2 points finds the longest line segment
> that violates a constraint. There may be too many
> papers for manual navigating when 5 candidates and
> 2 winners, for the truncating can only be done after
> the 2nd/last of papers naming both candidates.
> So an optimizer is needed and one that has a chance
> of finding an answer is one that locates the slope
> and position of the Boolean boundary. I assume that
> it still might not be written in 1 year's time. I
> don't know.
>
> The method of D R Woodall uses one by one eliminating
> which I assume is the source of the problems and so
> I don't suspect it would turn out to be a long
> lasting method.
>
> Both numerically and symbolically, a researcher has
> to develop general purpose solving tools that are
> likely to be useful outside of preferential voting.
> The non-linear optimizer is missing.
>
>
> (3) ----------------
>
> Mr Schulze's article.
>
> Before I get to that I quote this text of the editor, Mr Wichmann:
>
> "It has been decided that the Editor should
> ensure that the main points of such papers
> are intelligible to non-mathematical readers
> by placing an appropriate summary here."
>
> The whole article of Mr Schulze is unbelievable.
>
> A defective spot in Mr Schulze's article is this:
>
> "4 Implementation
> ...
> It cannot be said frequently that the order
> of the indices in the triple-loop of the
> Floyd algorithm is not irrelevant."
>
> The principle of having the winner unaffected by the lettering
> might be failed.
>
> E.g. this could happen. (The only change is a reordering
> of ALL the for loops inside of the algorithm):
>
> Harry (=#1) Winner
> Horace (=#2) Loser
> Horseradish (=#3) Loser
>
> Harry (=#2) Loser
> Horace (=#1) Winner
> Horseradish (=#3) Loser
>
> Mr Schulze's "neutrality" principle prevents the problem (if
> any). The text "neutral" is absent from the PDF article.
> An error at the top of the algorithm passes into the rest of
> the code and make analysis more complex.
>
> There ought be a program that tests and that writes out numbers.
**********************************************************
> To: single-transferable-vote at yahoogroups.com
> From: Markus Schulze <markus.schulze at alumni.tu-berlin.de>
> Date: Mon, 15 Dec 2003 15:40:28 +0100
> Subject: Re: Voting Matters, Issue 17, of McDougall Trust
>
> Dear Craig,
>
> you wrote (15 Dec 2003):
>
> > A defective spot in Mr Schulze's article is this:
> >
> > "4 Implementation
> > ...
> > It cannot be said frequently that the order
> > of the indices in the triple-loop of the
> > Floyd algorithm is not irrelevant."
> >
> > The principle of having the winner unaffected by the lettering
> > might be failed.
> >
> > E.g. this could happen. (The only change is a reordering
> > of ALL the for loops inside of the algorithm):
> >
> > Harry (=#1) Winner
> > Horace (=#2) Loser
> > Horseradish (=#3) Loser
> >
> > Harry (=#2) Loser
> > Horace (=#1) Winner
> > Horseradish (=#3) Loser
>
> The Floyd algorithm is a shortest-path algorithm. When I said that
> "the order of the indices in the triple-loop of the Floyd algorithm
> is not irrelevant" then I meant that you have to consider the
> possible short cuts in the mentioned order. Otherwise the Floyd
> algorithm does not work (i.e. does not give the correct result).
>
> I wrote:
>
> > for i : = 1 to N do
> > for j : = 1 to N do
> > if ( i <> j ) then
> > for k : = 1 to N do
> > if ( i <> k ) then
> > if ( j <> k ) then
> > {
> > s : = min { p[j,i], p[i,k] } ;
> > if ( p[j,k] < s ) then
> > p[j,k] : = s ;
> > }
>
> However, Mike Ossipoff always describes the Floyd algorithm as
> follows:
>
> > for i in range(N)
> > for j in range(N)
> > for k in range(N)
> > low=min(B[A(i,j)],B[A(j,k)]
> > if low>B[A(i,k)]
> > B[A(i,k)]=low
>
> This does not work (i.e. this does not find the correct values
> for the shortest paths).
>
> Markus Schulze
**********************************************************
> To: single-transferable-vote at yahoogroups.com
> From: Craig Carey <research at ijs.co.nz>
> Date: Tue, 16 Dec 2003 15:35:22 +1300
> Subject: Re: Voting Matters, Issue 17, of McDougall Trust
>
> At 2003-12-15 15:40 +0100 Monday, Markus Schulze wrote:
> >Dear Craig,
> >
> >you wrote (15 Dec 2003):
> >
> >> A defective spot in Mr Schulze's article is this:
> >>
> >> "4 Implementation
> >> ...
> >> It cannot be said frequently that the order
> >> of the indices in the triple-loop of the
> >> Floyd algorithm is not irrelevant."
> >>
>
> That is ambiguous and means: *all* 'for loops', or *some* of them
>
> >> The principle of having the winner unaffected by the lettering
> >> might be failed.
> >>
> >> E.g. this could happen. (The only change is a reordering
> >> of ALL the for loops inside of the algorithm):
> >>
> >> Harry (=#1) Winner
> >> Horace (=#2) Loser
> >> Horseradish (=#3) Loser
> >>
> >> Harry (=#2) Loser
> >> Horace (=#1) Winner
> >> Horseradish (=#3) Loser
> >
>
> I.e. secret internal relettering (or renaming or renumbering) inside
> of the algorithm affects who the winner is.
>
> You could answer these this question:
>
> 1. Does your method fail the test.
>
> You could write a Monte-Carlo routine.
>
> It seems the McDougall Trust operation is prepared to leave no
> stone unturned in order to exclude me when decisions to reject
> (all or parts of) submitted articles are made. I don't know if I
> can get things into the journal.
>
>
> >The Floyd algorithm is a shortest-path algorithm. When I said that
> >"the order of the indices in the triple-loop of the Floyd algorithm
> >is not irrelevant" then I meant that you have to consider the
> >possible short cuts in the mentioned order. Otherwise the Floyd
> >algorithm does not work (i.e. does not give the correct result).
>
> I am expecting a rejection of the method, and not any clarification
> of previous wordings about the method.
>
> And as you know, there is other allegation that would make your method
> too unlike ST to ever promote here, which is that the specifiying or
> not specifying the very last preference affects the winners when in
> neither case there was a tie. Your algorithm creates the mistake in
> the first lines and then sends it into the cyclic washing machine of
> the for loops. It is plain when the algebra is done. I suppose the
> purpose is that you have no idea of what you are doing when designing
> and then have an idea that you should not discard the result.
>
> Since you prefer wording up ideas, how would you word uop the idea of
> not discarding an unsatisfactory method that obviously no one else
> would get serious over using. To fix the mistakes would lead to less
> complexity. So readers need not look at your arguments.
>
> It does not matter whether your method is monotonic: it should be
> rejected while people presume my allegations are correct.
>
>
> >I wrote:
> >
> >> for i : = 1 to N do
> >> for j : = 1 to N do
> >> if ( i <> j ) then
> >> for k : = 1 to N do
> >> if ( i <> k ) then
> >> if ( j <> k ) then
> >> {
> >> s : = min { p[j,i], p[i,k] } ;
> >> if ( p[j,k] < s ) then
> >> p[j,k] : = s ;
> >> }
> >
>
> That is incomprehensible to me.
>
>
> >However, Mike Ossipoff always describes the Floyd algorithm as
> >follows:
> >
> >> for i in range(N)
> >> for j in range(N)
> >> for k in range(N)
> >> low=min(B[A(i,j)],B[A(j,k)]
> >> if low>B[A(i,k)]
> >> B[A(i,k)]=low
> >
> >This does not work (i.e. this does not find the correct values
> >for the shortest paths).
> >
>
> That is a dud argument or no argument.
>
> Your method is in the trash can for failing 2 tests. Can you get
> it out ? (rather than being creating natural language statements),
>
> I would not use the word correct but say "as intended". You seem
> to be in private communication with Mr Ossipoff.
>
> The main idea is that tests are ordered in a sequence and once a
> failure occurs, then there is no need to read about your untrusted
> arguments indicating that the method is failed under this particular
> check.
>
> --------------------------------------------------------------------
>
> Regarding my comment on Mr Woodall's method, I missed a way to
> test the method and compare it with Meek STV. I could use REDLOG
> to derive a 2 winner 2 preference 4 candidate method and then simply
> see which of the 2 STV-like methods is most similar. REDLOG was
> making mistakes and I could have tried hard to debug the output
> of the German software. Due to the grand plan of eliminating the
> German university's QE solver, I am now unable to compare Meek STV
> with the D R Woodall method.
>
> Testing using a sequence of:
> * 1 preference methods, then
> * 2 preference methods, then
> * 3 preference methods, ...;
> seems to be a bit plausible.
>
> Craig Carey
**********************************************************
> To: single-transferable-vote at yahoogroups.com
> From: Markus Schulze <markus.schulze at alumni.tu-berlin.de>
> Date: Tue, 16 Dec 2003 10:14:17 +0100
> Subject: Re: Voting Matters, Issue 17, of McDougall Trust
>
> Dear Craig,
>
> I wrote (15 Dec 2003):
> > The Floyd algorithm is a shortest-path algorithm. When I said that
> > "the order of the indices in the triple-loop of the Floyd algorithm
> > is not irrelevant" then I meant that you have to consider the
> > possible short cuts in the mentioned order. Otherwise the Floyd
> > algorithm does not work (i.e. does not give the correct result).
>
> You wrote (16 Dec 2003):
> > That is incomprehensible to me.
>
> You can find Floyd's shortest-path algorithm in many books on graph
> theory or combinatorial optimization. When you cannot find any good
> book on Floyd's shortest-path algorithm then I suggest that you
> should implement Dijkstra's shortest-path algorithm. Also Dijkstra's
> shortest-path algorithm has a runtime of O(N^3), where N is the
> number of candidates; the main reason why I use Floyd's algorithm
> and not Dijkstra's algorithm in my paper is that the source code
> of Dijkstra's algorithm is significantly longer. However, Dijkstra's
> algorithm is significantly less difficult to understand. ("Dijkstra"
> is also known as "Dykstra".)
>
> By the way: The proofs that my method is well defined and satisfies
> Pareto, monotonicity, resolvability, independence from clones,
> reversal symmetry, and Woodall's plurality criterion do not
> presume that I use Floyd's algorithm.
>
> You wrote (16 Dec 2003):
> > Since you prefer wording up ideas, how would you word uop the idea
> > of not discarding an unsatisfactory method that obviously no one
> > else would get serious over using.
>
> My method is already used by the "Software in the Public Interest"
> (SPI) project and the DEBIAN project.
>
> Markus Schulze
**********************************************************
> Date: Wed, 17 Dec 2003 08:25:39 +1300
> To: Schulze <markus.schulze at alumni.tu-berlin.de>
> From: Craig Carey <research at ijs.co.nz>
> Subject: Single-Transferable-Vote: I moderated your addres
>
> To: Theoretician Schulze <markus.schulze at alumni.tu-berlin.de>
> Subject: Re: Voting Matters, Issue 17, of McDougall Trust
>
>
> I am moderating your e-mail address : markus.schulze at alumni.tu-berlin.de
>
> I shall delay sending reasoning and I may not even send reasoning.
> But I would e-mail reasoning if you request that information.
>
> Remember that we know that you believe during these days when you
> try to present yourself as never having heard the word. The aspect
> of withholding leads me to moderate your future messages.
>
> Recall that you wrote this:
>
> __________________________>__
> At 1998-04-24 17:07 +0200 Friday, Markus Schulze wrote:
> >Dear Mike, dear David,
> >
> ...
> > To my opinion, it is not possible, that B is elected, because
> > this would be a violation of the Neutrality Criterion (i.e.,
> > every candidate is treated equally), the Anonymity Criterion
> > (i.e., every voter is treated equally) or the Weak Monotonicity
> ...
> >
> >Markus
> __________________________<__
>
> The words "not possible" seem to rule out that you were tentative and
> unsure, on the so called "neutrality" rule.
>
> Is it your opinion that the method in the PDF file passes a neutrality
> rule ?. If so, then there is the other rule about the last preference
> being without an improper influence.
>
> ---
>
> You seemed to be in the wrong in 1988 as well, in that there is no
> need for a neutrality rule unless the set of principles is wrong.
>
> Re=lettering a, neutrality failing, preferential voting method would
> make it be multivalued in general.
>
> The right rules and excluding neutrality, don't lead to multivalued
> winner sets.
>
> So neutrality is implied by other right rules.
>
> Omitting one rule may fail to delete the rule.
>
>
> At 2003-12-16 10:14 +0100 Tuesday, Markus Schulze wrote:
> >Dear Craig,
> >
> >I wrote (15 Dec 2003):
> >> The Floyd algorithm is a shortest-path algorithm. When I said that
> >> "the order of the indices in the triple-loop of the Floyd algorithm
> >> is not irrelevant" then I meant that you have to consider the
> >> possible short cuts in the mentioned order. Otherwise the Floyd
> >> algorithm does not work (i.e. does not give the correct result).
> >
> >You wrote (16 Dec 2003):
> >> That is incomprehensible to me.
> >
> >You can find Floyd's shortest-path algorithm in many books on graph
> >theory or combinatorial optimization. When you cannot find any good
> >book on Floyd's shortest-path algorithm then I suggest that you
> ...
>
>
>
> You failed to reply to what I wrote.
>
> I maintain the worlds best mailing list on preferential voting.
>
> I am blocking you because your omitted information.
>
> A best defence for you could be to say that you would be writing
> two messages and the 2nd message would answer the questions (so
> you have to write software too: I guess you already did that but
> are keeping the results and the existence of the software, secret).
> Also the defence would say that the first message should have
> mentioned that another was coming.
>
> I I wouldn't alter it too much since it is currently able to
> persuade.
>
> Like the McDougall Trust said to me, we shall see how it goes, ie.
> where you go from here. It seems "not possible" that a follower of
> pairwise comparing would want to make a method be fair to
> the smallest minorities.
>
> Craig Carey
> Auckland
More information about the Election-Methods
mailing list