[EM] Steph: 3-level voting

[Elisabeth Varin/Stephane Rouillon]

Just to finish my previous society evolution theme:
I think society must change and adapt to fit evolution.
We now have to face cloning dilemas for example.
If the election method encourages society to stall,
when one day people finally choose an extreme to
face the reality, it can lead to a sudden revolution
instead of a serie of reforms... I prefer soft transition
when needed and possible. So keep the advantages when
proven, but fight inertia when it is just what has been...
So middle or extreme? It depends. Both can be good.

MIKE OSSIPOFF a écrit :

> Steph--
>
> you said:
>
> Why I do not like 3-level voting.
>
> First from an informatician point of view, there is no reason to
> artificially restrain the ranking input to 3-level. Why not let
> the voter use as many levels (s)he wishes.
>
> i reply:
>
> i told why in my message: y/n uses the same ballots that are now
> used in elections that allow voters to vote yes or no on several things.
>
> its' simple. The voter doesn't have to choose which point score, in
> a large range, to give someone. Give him a negative, a positive, or
> nothing.
>
> i have no objection to larger rating-ranges, but they'll require new
> balloting equipment, and they may have an unnecessary degree of complexity
> compared to the simpler -1,0,1 method that i described.
> if people preferred 0-10, that would be fine with me.

Weigthing methods introduce too much strategy. Ranking is far better. I do not
think
excessive simplicity is mandatory compared to getting the exact wishes of the
people.

> you continued:
>
> It could be one, two, three
> or more as fits. So a (possibly truncated) preferential ballot is ideal.
> It is used in Ireland, so I do not think the rest of the planet is too
> stupid to use it. If voting means expressing my preferences, let
> me express all those I can and I want...
>
> i reply:
>
> Sure, i too prefer the best rank methods to CR, in its various versions.
> But that's beside the point. There are infinitely many ways  to count
> ranked ballots. That makes it difficult or impossible to get one of
> the few good ones adopted. That's one reason to propse Approval or
> -1,0,1 instead of a rank method.

IRV can be counted by hand with some minimal phone or computer relays.
People of any language usually know numbers... I am not sure which of
Approval result or IRV result would usually be nearer to Ranked pairs result.
For IRV changes are minimal too.
Ballot would change or not depending in which country you live.
American punchers that make holes would not fit, but in Canada
it would fit. Instead of putting a cross in front or your most preferred
candidate, we would just write the ranks in the circle.
Where do you live?

>
>
> MIKE OSSIPOFF wrote:
>
> >
> >An advantage of Approval over CR is that Approval doesn't require a
> >ballot different from the existing one. An advantage of CR over
> >Approval is that CR is familiar to people, whereas some mistakenly
> >feel that Approval violates 1-person-1-vote.
>
> you replied:
>
> I am one of these persons. It is OBVIOUS that approval gives more power to
> voters who like several candidates.
>
> i reply:
>
> Then this is a case in which something is obvious to you but incorrect.
>
> The person who votes for one candidate only is the person whose
> favorite probably has a win. The person who votes for several candidates
> is someone who needs to compromise. Someone who needs to compromise
> isn't more powerful.
>
> if voting for more candidates makes you more powerful, then how
> powerful would you be if you voted for all the candidates?
>

You are more then right. I should have precised my position. I think voting for
half
of the candidates is the most "powerful" strategy. Still I do not treat power
and
fairness the same. I think approval is fair from a voter perspective, it simply
does not give the same power to each voter.

> it's been shown several times on this list that plurality causes
> voters' voting power to differ by a many times greater factor than
> Approval does. Approval gives more uniform voting power. if you
> disagree with those posted demonstrations, then tell us which part
> you disagree with.
>
> you continue:
>
> It favours a concensual voter (who would
> accept to rally with other candidates) over a one-minded voter.
>
> i reply:
>
> no it doesn't necessarily. if your favorite appears to have a win,
> Approval doesn't favor you if you vote for several candidates.
>
> And, in the comparison between Smith and jones, it doesn't matter how
> many candidates you've voted for--you only have one vote for determining
> whether Smith outpolls jones, or jones outpolls Smith.
>
> you continued:
>
> So approval
> voting forces a voter to choose between approving or not his (her) second
> preference and next
>
> i reply:
>
> That was hardly a secret.
>
> you continued:
>
> ...without benefiting from other voter preferences
> information.
>
> i reply:
>
> if you're saying that Approval requires more strategy than
> Condorcet(wv), no one denies that.
>
> you continued:
>
> Rallying methods force you to make the same choice, but at the
> round it occurs, the decision is more guided because some candidates have
> already been identified as loosers.
>
> i reply:
>
> Every method determines who loses and who wins. irv decides who
> loses based on incomplete information, just the top entry in each
> ranking. make an important irrevocable decision based on a fraction
> of the available information?

I totally agree about this being IRV's flaw.

> you continue:
>
> Despite this, I do not think approval is
> unfair to voters. Voters have an equal opportunity. They are just encouraged
> to become multiple-approving voters.
>
> i reply:
>
> Wrong. if your favorite appears to have a win, then Approval doesn't
> encourage you to become a multiple-approving voter. likewise, if
> your favorite is the only candidate who is at all acceptable to you,
> Approval doesn't encourage you to be a multiple-approving voter.
>
> Some will vote for 1, some for 2, some for more, depending on their
> utilities and their estimates of who will be the frontrunners.
>
> i suggest that you check the Approval Strategy articles at:
>
> http://www.barnsdle.demon.co.uk/vote/sing.html
>
> The first article of those will be rewritten for better clarity, but
> check the articles now anyway.

I could in fact be wrong. I should look at this more carefully. It is another
problem I wished I had the time to look at with a quantifiable criteria.
I will put it higher on my list.

> you continued:
>
> Thus approval IS UNFAIR to candidates.
> Even a Condorcet Winner can loose with approval because the winner can be
> the 7th choice of all voters, the last acceptable candidate.
>
> i reply:
>
> no comprendo

An example should do the job. Sincere preferences:
2: ABCDEFGH
1: BCDEFGHA
2: ACDEFBGH
2: ABDEFCGH
2: ABCEFDGH
2: ABCDFEGH
2: ABCDEFGH
7: H

A is clearly the winner!
When you ask these people for an approval vote, all of them could be very scared

that H could win because he's a political extremist that rally several
one-minded fans.
So just to be safe against that new potential dictator, they put their approval
cut-off after G.
Anybody but H. Result: G is elected. Very unfair to A...

> you continued:
>
> Finally an other reason that proves approval does not give the same power to
> every voter is that the candidates final weights do not represent
> a proportional result with an equal weight for each voter. This is a basic
> monotonicity feature.
>
> no comprendo.  By final weights, do you mean vote totals? Approval
> isn't proportional representation, but you seem to be objecting that
> Approval isn't proportional in some not-quite-specified way.
>
> you want the candidates' vote totals to be proportion to exactly what?
> And why?

I mean residual approval weights. In order to match the wish of the electorate,
the result should represent an equal contribution of each voter. So if any voter

changes is mind it can affect the result an equal amount the result as anyone
else.
You can later use this result to only elect one person or more or interpret it
the way you want. But the intermediate result should show the one person, one
equal impact on the result principle. From an equity point of view this is
mandatory.
Neither the electoral method nor the number or type of candidate should affect
this property. Or it means the weight of your contribution depends upon
how it is counted, or who runs or how many people run. Such cases are clearly
unfair to at least some voters, the one having less impact on the result.
Sum of the final residual approval weights should be 100%. Remove any
voter, you should remove the same amount of representation. This is impossible
with approval. This is why I say approval gives more power to some voters.

> >But someone wrote to me suggesting a method that has both of those
> >advantages: Yes/No voting:
> >
> >Each voter may give a candidate "Yes", "No", or nothing. A candidate's
> >No votes are subtracted from his Yes votes, and the result is his
> >score. The candidate with highest score wins.
> >
> >That's equivalent to CR, with -1, 0, 1. Which is equivalent to
> >0, 1, 2 CR. Which of course is strategically equivalent to Approval.
>
> you replied:
>
> I definitely disagree with Mike on this. Triplets cannot be strategically
> equivalent to Approval.
>
> i reply:
>
> you're certainly free to disagree with anything. But Approval's strategic
> equivalence to CR has been demonstrated on EM. if you've demonstrated,
> in the message to which i'm replying, that Approval isn't strategically
> equivalent to CR, i must have missed that part of your message. Would
> you repeat it for me?

I think we have a nomenclature problem. I will try to rephrase.
Approval and CR use the same way to determine their optimal strategy.
Maybe this is what you define as strategically equivalent.
The same strategy or algorithm cannot be used for both methods,
because one needs to define one cut-off between 0 and 1, and the other needs to
define two cut-off. They should be of the same family, using different utility
neighbourhood,
but it cannot be the same algorithm so this is what I mean by not strategically
equivalent.
Triplets (0, 1, 2) is strategically equivalent to triplet (-1, 0 , 1) using my
vocabulary.
I think we agree.

> you continued:
>
> Again these methods do not ensure an equal weight to
> every voters when some of them truncate their ranking
>
> i reply:
>
> Are you talkinga about rank methods or Approval?

Weighting methods. Ranking methods can be compiled to fit the residual
approval weights property.Weighting methods could fit if people are forced
to give full total weight to their ballot.

> you continued:
>
> , thus it violates the
> 1-person, 1-vote concept (1-"full vote").
>
> i reply:
>
> But it isn't clear what kind of 'weight' you're referring to, or
> why it should be equal.  Any voter has the power to cancel out any
> other voter. Each voter has exactly one vote on whether or not Smith
> will outpoll jones. Approval, compared to plurality, gives voters'
> voting power that differs by a many times smaller factor, where voting
> power is measured by a voter's power to improve his expectation with
> his ballot. That's been demonstrated several times on EM. if you're
> going to disagree with it, then tell us which part of the demonstrations
> you disagree with, or post your own demonstration that it isn't true.
>
> But we should thank you for re-affirming the claim that some people
> are going to insist that Approval violates 1-person-1-vote. it's something
> that we run into periodically when talking to the public
> about Approval. Thank you for demonstrating it here for us.
>
> >Every county already has balloting equipment that supports voting
> >yes or no on a list of things. People are familar with it and won't
> >call it a violation of 1-person-1-vote. Moreover, many people would
> >very much like to give negative votes.
>
> you replied;
>
> And it would really be the start for unrepresentation. An election is a
> representation exercise, not an ultimate wrestling contest where only one
> winner has to get out of the ring.
>
> i reply:
>
> So now you're saying that only proportional representation is right.
> pr merely postpones single-winner choices to parliament. Single-winner
> social choice methods are the fundamental implementation issue of
> democracy. proportional representation has shown itself to be unwinnable
> in the u.s., in these times. The reason is obvious: pr is a new concept
> of representation. Voters, suspicious of their representatives, fear
> a completely new represation concept, believing that it will worsen
> their representation, giving more power to some groups that they fear.
>
> single-winner reform is just a fairer way of doing what we already do
> in the u.s., Canada, England, France, and other single-winner countries.
>
> minimal change is more winnable than entirely new concepts, and better
> single-winner methods would produce ultimate results similar to pr
> anyway.

Yes I think proportional representation is good and France uses a proportional
representation of lists for municipal elections. For parliament and presidential

elections, two round single-winner methods as you said.
No, this part of my discussion is not to defend PR. It is the same logic
although that leads to a direct link between each voter and its equal
fractional part of the residual approval weights.
Finally you should not mix feasibility of methods with
identifying their qualities. If it is the best method with the best qualities,
the goal is clear, the tactic to get it in place is a different problem.
We should all try to focus on the goal first. Next look at the best tactic.
Looking at tactics is useless if we do not agree on the goal first.

> you continued:
>
> Nobody should ever see its vote being
> cancelled by somebody else. It is fundamental that proportions be respected.
> A 5011 to 4999 win is VERY different than a 12-0 win. The first is almost a
> tie, the second expresses a clear preference. When generalized to obtain a
> fully proportional chamber, those results have to produce different results.
>
> i reply:
>
> maybe your method would be good for pr.

I will post soon a copy of the multiple-winner proposition I will defend in
February
infront of the Assemblee Nationale du Quebec. An english draft version is
available at:
http://www.fairvotecanada.org/phpBB/viewtopic.php?topic=8&forum=1&3
(fourth post)

> you continue:
>
> Do not destroy the information from the start!
>
> i reply:
>
> Good point. margins and relative margins start out by destroying
> information about majorities, explaining why your method is such
> a violator of majority rule.

I do not concur. Please explain. I found that winning votes destroys the
relevant information about the probability of a pairwise comparison
being sure or shaky...

> you continued:
>
> The very same logic is
> applied to compare pairwise victories and it leads to relative margin as
> being the fairest comparison criteria. A 54 republicans and 48 democrats
> senate is not the same than a 6 republicans senate.
>
> i reply;
>
> you're confusing single-winner elections with pr. i've told you
> several reasons why wv is better than margins or relative margins
> in terms of majority rule, lesser-of-2-evils problem, and the goal
> of minimizing overruled voters, and you reply that relative margins
> is better as a pr method. Then use it for pr.

I am not confusing them. I use the same logic, applied to single-winner
fair representation instead of multiple-winner fair representation. You
might not find it relevant to elect one person, I do. It shows what method
are fair between all voters according to the same principles.

> you continued:
>
> Do not dismiss the
> opposition, it counts.
>
> i reply:
>
> Everyone's vote counts, for each pairwise comparison. But a pairwise
> vote for a proposition that has already lost no longer counts the
> same as a pairwise vote for a proposition that has won. We needn't speak of
> overruling the former, because it was overruled when the
> public voted.
>
> you continued:
>
> NO to negative votes.
>
> i reply:
>
> Wait a minute--what does that have to do with negative votes. Are
> you saying 'no to the CR that has -1,0,1' ?  it makes no difference if
> we call it -1,0,1  or 0,1,2.  in one sense the former is simpler, in
> another sense the latter is simpler. either would be fine with me,
> as would 0,1.   i prefer 0,1.
>
> By the way, if you feel that Approval violates 1-person-1-vote, then
> how about this points system?: Each voter may give to each candidate
> either 1 or 0 points. Does that violate 1-person-1-vote?
>
> if i give 1 to more candidates than you do, i high-rate more candidates
> than you do, but you low-rate more candidates than i do. who says that
> my high-ratings carry more power than your low-ratings. Would you
> really have more power if you high-rated lots of candidates whom you
> dislike, against your own interest?
>
> mike ossipoff

This discussion has some point in common with the margins, relative margins
and winning votes debate. It is all about how we perceive and treat the opposant

about a result. Should we muzzle them, shifting the result as soon loosers
become winners?
Should we annihilate them by taking a winner down with each looser? Should we
consider them as a fraction of opponents, being able to contribute in a relative

part of their number? I pick the last, choose your own... I find the first case
(wv)
too unstable as if we only had the choice between hearing any extreme. I find
the second
(margins) not making the difference between a small but decisive victory and an
almost
tie result where everybody wanted to participate.
Which criteria carries the most relevant information?
I might try to illustrate how I see this. Imagine we have an UNO representative
from your country having to vote on a couple of issues.What information would
mean most
her or him? margin: "1% more of the people were in favor than the people that
were against"
Was it a hot debate or only the 1% activists filled the poll? How to know?
relative margin: "25% more of the persons who took a position were in favor"
How many were they? Almost none or all of them?
winning votes: "2,5% of the people expressed a vote in favor, beating their
opponent".
2,5%, is it a lot? Was there no votes against or nearly 2,5% ?
The same kind of data with only one criteria for each issue.
What kind of criteria would you use to identify among several issues, which one
to
compromise over? Results for the previous case were: 2,5% in favor and 1,5%
against.
Now I ask again: which criteria carries the most relevant information?

The same logic still applies here. Negative vote transforms a 32-28 victory as a
4-0 victory.
Margin is not good when it comes to estimate the opposition fraction. Negative
vote destroys
the information. So I dismiss any counting method that would sum up points like
borda suggests.
None leads to an equal residual approval weigth for each voter. To verify this
principle,
1 and 0 votes should be treated as splitted shares of a ballot. Thus if you
approve two
candidates they would receive half a vote each as do cumulative single-seat
(Donald Davison)
or normalized ratings (Tom Ruen). They do not violate 1-person 1-vote as I
define it.

For your last question about power of a vote with approval, I will have to do my
homeworks...

Steph.
PS: Adam illustrated to me examples that a sincere CW can be poorly supported or
very supported.
I understood even without a mathematical definition.

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