[EM] Steph, re: CW example

[Elisabeth Varin/Stephane Rouillon]

MIKE OSSIPOFF a écrit :

> Steph--
>
> you wrote:
>
> Approval definitively favors centrists candidates.
>
> Are you saying that's bad? As i've said, Approval quickly homes in
> on the voter median candidate, and stays there. you're right about that.
> Also, when voters in Approval err because the all believe an incorrect
> poll, Approval errs by picking someone in the middle, unlike
> irv and margins methods, which can err by electing an extreme
> candidate.
>
> if everyone believes mistakenly that Favorite will outpoll Worst, so
> that Favorite voters don't vote for Middle, then the Worst voters,
> using those polls, will vote for Middle.
>
> With everyone believing the same poll, Approval can err if Favorite
> has a 1st choice majority, but people believe that Worst will outpoll
> Favorite. Then the Favorite voters will vote for Middle, and give the
> election away to Middle. But that failure is one which elects Middle,
> not an extreme.

>From a representaion point of view, yes I think favouring Middle candidates as
you call them is not good. Not because I am an extremist. First I think extreme
and middle candidates definition are relative to further sincere preferences. So
a condorcet ranking (typically Ranked Pairs (use the criteria you want)) is the
best tool to identify who is middle and who is extreme. Sometime middle is good,
other times extreme is good.
I am affraid, that with approval, all parties would adapt their ideas to enlarge
them
so to reach the consensual median ideology, typically the actual statu quo. So
the feedback effect of approval would be that parties, in order to get elected,
would all defend the actual system. I do think the actual system is mainly good,
we would have needed to change it if it was not fitting previous needs. But new
ideas and new discoveries occur, sometimes changing the world (TV is a great
example). Sometimes,
we need some change and the sooner the experience the sooner we now it is good
or bad. Society is like a child to me, it needs to try in order to learn,
sometimes from mistakes. If you never change, you never try, and you can learn
to late.

Sorry for the philosophical course. In other words, the method should consider
middle and extreme both potentially good.


> you continued:
>
> As a voter I should be free to decide if I want to support one or more
> candidates without the electing method affecting my winning probabilities.
>
> i'm not sure what you mean by that. if you're saying that Approval
> requires more strategy than Condorcet(wv), no one will argue with
> you on that.

I agree. I will try to include a simple approval model when I will e-mail
gain esperance for different methods. In another mail.

> So let us check another subject:
> relative margins (rm) vs winning-votes (wv)
>
> MIKE OSSIPOFF a écrit :
>
> >A margins advocate could say that he doesn't consider the majority
> >defensive strategy criteria important, but they measure for the
> >standards of majority rule and getting rid of the lesser-of-2-evils
> >problem. The wv methods are the ones that do well by the majority
> >defensive strategy criteria. The wv versions of Ranked-Pairs and
> >BeatpathWinner/CSSD meet all of them.
>
> you wrote:
>
> None of these criteria has shown to me that they can protect a Condorcet
> Winner. It seems they show an enhancement of the probability of being able
> to
> protect a Condorcet Winner and I like that. By how much are they efficient?
>
> i reply:
>
> Again, i'm not sure what you mean. if you're saying that you want a
> method to always elect the CW, no matter what, then i admit that i
> can't offer you that. What method can?

None yet.

> But it might be that you're confusing the defensive strategy criteria
> with some others, since the defensive strategy criteria say nothing
> about probability. if you want to write a probability criteria, then
> by all means do so. Then, after you've done that, that would be the
> best time to talk about probability in criteria.
>
> >Aside from whether you like those particular criteria, it can't really be
> >denied that margins is the big
> >violator of majority rule, and is the method that makes greater strategic
> >demands on voters who want to protect majority rule, or who want to protect
> >the win of a CW.
>
> you said:
>
> Please check
> http://groups.yahoo.com/group/election-methods-list/message/10107
> for an example where Winning Votes harms a Condorcet Winner.
> So both criteria can harm a Condorcet Winner.
>
> i reply:
>
> That's the old poorly-supported-CW example. i've discussed it many
> times. i've answered that objection many times. yes, if there's no
> majority who care to vote the CW over candidate B, then the B voters
> can steal the election by truncation. i never claimed that a
> poorly supported CW can't lose because of truncation. But so what if
> he can? in margins, and in your relative margins, a majority-supported
> CW can lose to voters who steal the election by truncation. Which is
> worse?
>
> As i said, SFC and GSFC don't mention truncation, but they tell what
> i claim about wv when no one order-reverses.
>
> you said:
>
> The question is: how big is the
> majority rule violation in both cases?
>
> i reply:
>
> in the example that you cite above, there's no majority rule violation
> when A loses to B. With 11 voters, A beats B 4 to 3. With 4 voters
> voting A over B, out of 13 voters, there's no majority rule violation
> if B wins instead of A.
>
> Give me numbers from an A, B, C, AB, BC,
> AC, BA, CB, CA distribution. I will do it myself, but for a small number of
> voters. I sound like M. Carey, working with 3 candidates and 4 voters...
>
> i've given you examples in which your method defeats a CW in violation
> of majority rule under conditions where that needn't happen. you've
> shown an example in which a truncation defeats a poorly supported CW
> with no majority rule violation. i've shown that wv meets SFC and
> GSFC. i'm not sure what you're asking for above.
>
> >It was shown here that, with margins methods, there are plausible
> >ordinary situations in which the only Nash equilibria are ones in which
> >defensive order-reversal is used.
> >
> >We probably agree that it's undesirable to have to drop a defeat,
> >since each pairwise defeat represents a public vote about the
> >relative merit of 2 candidates. Dropping a defeat overrules all the
> >voters who won that public vote. So we'd like to at least overrule
> >as few as possible.
>
> you said:
>
> Agreed. I even rallied to M. Tidemann's method consequence. It minimizes the
> largest overruled pairwise comparison, so it minimizes the largest overruled
> opinion created by different people.

All previous stuff will be adressed in some kind of probabilistic criteria....

> i reply:
>
> if you're not going to read my messages, it's difficult for us to
> talk. See below:
>
> >Margins advocates would answer that not dropping a defeat overrules
> >the voters who voted agains that defeat. Wrong. Those voters were
> >overruled by the result of the public vote between those candidates.
> >Not dropping a defeat that the public enacted doesn't overrule anyone.
>
> The same is valid for winning votes, no?

I do read. I do not always understand.

> No. winning votes doesn't count losing voters as being overruled by
> the count rule, because losing voters, opponents of a publicly-voted
> defeat, have already been overruled when the voters voted between
> those 2 candidates. X beats Y. margins advocates want to say that
> the y>x voters are overruled by the count rule if we keep that defeat.
> No, they've already been overruled when they lost that pairwise
> comparison in the public vote. But if we drop that defeat, then
> we're ignoring a collectively-voted public preference, then it is we
> who are overruling the voters who voted for that defeat.
>
> Do you see the difference? it's margins, not wv, that tries to claim
> that we're overruling voters who actually have already lost and who
> have no claim to getting their way.

I read. I do not see the difference. For me both methods overrul local pairwise
results in order to obtain the best global result they can.

> >For that reason winning-votes (wv) is the more democratic class of
> >methods, when it comes to honoring pairwise votes and expressed
> >public wishes.
>
> you said:
>
> Because I have not yet been able to quantify (wv) and (rm) Condorcets
> Winners
> stilling probability, I cannot use this criteria to select the "most
> democratic
> criteria".
>
> i reply:
>
> i admit that i don't quite know what you mean there.
>
> you said:
>
> I prefer to use a quantifiable objective, the minimization of the
> probability of overuling the wrong pairwise comparison, which is equivalent
> in
> my sense to minimize the largest overuled relative margin.
>
> i reply:
>
> So you're not so much interested in overruled voters, but you want
> to judge which proposition is right and which is wrong. First you have
> to provide an operational definition of a right or wrong proposition.
>
> That's a definition that contains a test by which we can determine whether
> or not a proposition is right or wrong according to that
> definition. Without that, it's meaningless to talk about right or
> wrong propositions.
>
> And don't say that your definition of a right proposition is one with
> a large margin or relative margin. you see the circularity of saying
> that relative margins favors right propositions then, right?
>
> your definition of a right or wrong propositon should be one that
> in some way concides with what we expect right or wrong to mean.
> how would you test a propostion for rightness or wrongness in a way
> that makes sense according to the usual meaning of those words?
>
> mike ossipoff

I agree and I have heard some people saying the same about you.
Actually I do not think it is a misleading attempt, it is just sometimes we do
not have any other way to express what we think.
I will try to present things differently, in a non-circular way.

I have to go, I will continue soon.

Steph.


----
For more information about this list (subscribe, unsubscribe, FAQ, etc), 
please see http://www.eskimo.com/~robla/em