[EM] SARA voting: easier-to-describe MAS

[Michael Ossipoff]

(Replying farther down, after &&&&&&&&&&&&&&...)

Chris quotes Jameson & me:

On Oct 25, 2016 9:40 AM, "C.Benham" <cbenham at adam.com.au> wrote:
>
> On 10/25/2016 4:47 AM, Michael Ossipoff wrote:
>
>> When sincerely-voted Approval chooses differently from Condorcet, it's
choosing socially better.
>>
>> (Jameson wrote):
>>
>> But even if they do win under honesty, they will probably lose under
"semi-honest offensive strategy" (SHOS) in which the stronger wing
truncates the CW to equal-bottom.
>>
>> (endquote)
>>
>> In rank methods, we've been calling that offensive truncation.
>>
>> In Approval, I don't regard it as an offensive strategy at all. In
Approval, the middle CWs doesn't need support from both sides, as s/he does
in general pairwise-count methods.
>>
>> The larger wing isn't the one that needs the CWs, if you're after the
best particular candidate you can get. But you're talking about a strategy
problem that shouldn't concern voters in Approval.
>
>

&&&&&&&&&&&&&&&&

(You wrote):

> Mike,
>
> Given my selection of quotes above, I find your negativity on
Smith//Approval  (very similar to Max Covered Approval) a bit inconsistent.

(endquote)

Good point. Because Approval is a good method, then how can it be bad to
choose the Approval winner among the Smith-set?

Fine, but by sincere Approval's standard, is Smith//Approval   as good as
just using Approval? Obviously not.

Approval, not Smith//Approval, elects the candidate in the most people's
top-set, when people vote the optimal strategy of approving (only) their
top-set.

Smith//Approval of course elects the CWv if there is one, instead of the
Approval winner.

Anyway, of course, if a ranking method is chosen, it's because the choice
is based on different goals or considerations:

Maybe people want to choose _among_ their top-set, maybe assuming
(incorrectly) that that won't increase the probability of electing from
their bottom-set.

Maybe, due to overcompromising tendency, or rivalry among similar parties,
voters need rankings, to avoid or mitigate their resulting voting-errors.

...justifying a rank method.

If a method meets MMC, and if a voter is majority-favored (MF), then s/he
can rank sincerely, & if hir mutual majority do so too, then the winner
will be in hir top-set.

In other words, s/he can choose among hir top-set in complete safety.

Smith//Approval & IRV meet MMC, and so they both offer that guarantee to a
MF voter.

So far, so good.

But wait...What if you aren't MF?

Then in Smith//Approval, what does it take to try to protect your top-set?

The CWs is likely to be in many top-sets, likely in yours. Good rank
methods have som protection for the CWs.

In an FBC method like Bucklin, you can effectively vote,  Aporoval-style.
Top-voting one top-set member won't hurt another one.

In Smith//Approval, that's not so.

In methods with wv strategy, a CW's win isn't lost by truncation from one
side.

...& burial should be deterred if the CWs's voters plump.

...& if burial is done anyway, then that hands the win to the opposite
wing. If that's your wing, then that's better than the CWs.

Smith//Approval doesn't have wv strategy.

Bucklin, & wv- strategy methods,  but not Smith//Approval, let the CWs's
voters protect the CWs by plumping.

By Approval standards, Smith//Approval isn't as good as Approval.

By ranking method goals & considerations, it isn't as good as Bucklin or
wv- strategy methods.

I don't know the answers to the 2 questions below. Replying farther down.

>
> A couple of somewhat relevant questions:
>
> (1)How often is a sincere Condorcet winner who is supported by the
smallest of three factions of voters the Approval winner?
>
> (2) How often is the Approval winner not in the Smith set?
>
> And so what is wrong with Condorcet//Approval?

See above.

Michael Ossipoff

>
>> Jameson:-In Condorcet systems, the CW wins under honesty by definition.
But under SHOS, any CD-compliant method gets the wrong answer.
>>
>> Mike: No. In wv, a CWs isn't vulnerable to truncation. S/he still wins.
...the right answer in such a method.
>
>
>  But Winning Votes isn't a "CD-compliant method" .
>
>
> Jameson wrote:
>
>> Support Accept Reject Abstain voting works as follows:
>>
>> Voters can support, accept, reject, or abstain on each candidate.
Default is abstain.
>> Call a candidate "acceptable" if they are rejected by 50% or less and
supported or accepted by over 25%. If any candidates are acceptable,
eliminate all who aren't.
>> Give remaining candidates 2 points for each "support", 1 point for each
"accept", and half a point for each "abstain". Highest points wins.
>>
> I still hate default being anything other than bottom/"reject", and the
"25%" figure looks arbitrary.
>
> With the default rule as it is, it looks like the method fails
Mono-add-Plump.
>
> Say the winner X has a lower point score than some other candidate Y that
is barely rejected by over 50%. Then adding some ballots that do nothing
> but "support" X could raise the 50% threshold enough so that now Y is no
longer eliminated and wins.
>
> Chris Benham
>
>
>>
>> On Oct 23, 2016 8:21 AM, "Jameson Quinn" <jameson.quinn at gmail.com> wrote:
>> >
>> > Center squeeze occurs in 3-candidate election when the CW is the
plurality loser.
>>
>> So it refers to a special case of the problem of the defensive strategy
needed to protect the CWs's win.
>>
>> Certainly, in IRV, Benham, & Woodall, the main problem is when a middle
Cws has the smallest faction.
>>
>> And, in other pairwise-count methods, when I test for truncation or
burial vulnerability, I use such an example, because it seems more
favorable to finding vulnerability.
>>
>> But a CWs can lose without being smallest, and I don't make a
distinction about that except that it's part of the problem-examples in
IRV, Benham, and Woodall.
>>
>> You wrote:
>>
>> There are different levels of the problem, characterized by the level of
strategy necessary for the CW to win:
>>
>> (endquote)
>>
>> Sure. I was saying that with wv strategy, truncation from one side can't
take the win from the CWs. ...& that the CWs's voters can deter burial by
mere plumping.
>>
>> No need for the non-offensive wing to vote any less than sincerely in
methods with wv strategy.
>>
>> As you know, that isn't so with other Condorcet versions or general
pairwise-count methods.
>>
>> (You wrote):
>>
>> >
>> > -In systems like IRV or plurality, in order for the CW to win even
under honesty, the weaker of the two wings must betray their sincere
favorite.
>>
>> (endquote)
>>
>> Well, it isn't so much the size of the wing factions. It's more a matter
of which wing candidate is preferred by the CWs's voters.
>>
>> That's why I've been saying that IRV is fine for you if you're
majority-favored.
>>
>> If the CWs's voters transfer the other way, then you aren't in a
mutual-majority, and then, for you, IRV is then no better than Plurality.
>>
>> Many of us have been explaining that to Rob Richie for the past 30 years
or so.
>>
>> > -In systems like approval and score, the CW may or may not win under
"honesty" (however defined). Arguably, if they do not, they should not.
>>
>> (endquote)
>>
>> Correct. When sincerely-voted Approval chooses differently from
Condorcet, it's choosing socially better.
>>
>> Sure, not choosing the CWs got IRV repealed in Burlington.
>>
>> Not electing the CWs results in a majority who'd prefer a different
result.
>>
>> But I think IRV was repealed for more than that:
>>
>> Approval let's people protect the CWs, without favorite-burial,  if that
were what they want to do.  ...and I suggest that it isn't, because
electing from your top-set is usually more important than electing the best
particular individual candidate you can.
>>
>> CWs protection is more for rank methods, which are about choosing
_among_ your top-set, to elect the best particular candidate you can.
>>
>> (But see my earlier post yesterday, about that.)
>>
>> If you have a top-set, then, by definition, electing from it is the
important thing.
>>
>> Approve (only) your top-set.
>>
>> (More about that in my post yesterday)
>>
>> But, if it turned out to be really important to avoid having a majority
who'd prefer a different result, then that  could be a reason for voters in
Approval to slightly modify their voting to avoid approving past the
expected CWs (CWse). ...though that would lower your Pt.
>>
>> When approving your top-set, you'd just stop short of approving past the
CWse.
>>
>> Yesterday, too, I spoke of the possibility of your wing-faction having a
social agreement to that effect.
>>
>> What you're saying doesn't contract what I've been saying.
>>
>> (You wrote):
>>
>> But even if they do win under honesty, they will probably lose under
"semi-honest offensive strategy" (SHOS) in which the stronger wing
truncates the CW to equal-bottom.
>>
>> (endquote)
>>
>> In rank methods, we've been calling that offensive truncation.
>>
>> In Approval, I don't regard it as an offensive strategy at all. In
Approval, the middle CWs doesn't need support from both sides, as s/he does
in general pairwise-count methods.
>>
>> The larger wing isn't the one that needs the CWs, if you're after the
best particular candidate you can get. But you're talking about a strategy
problem that shouldn't concern voters in Approval.
>>
>> (You wrote):
>>
>> The defensive strategies that work are for the weaker wing to give the
CW near-top score, and/or the CW faction to give the stronger wing
near-bottom score.
>>
>> (endquote)
>>
>> Both, if people are trying for the best candidate they can get. Then
obviously the smaller wing should support the CWs, and the CWs's voters
should plump.
>>
>> But I don't suggest that goal for Approval, for the reasons stated
above, & in earlier posts.
>>
>> > -In Condorcet systems, the CW wins under honesty by definition. But
under SHOS, any CD-compliant method gets the wrong answer.
>>
>> No. In wv, a CWs isn't vulnerable to truncation. S/he still wins. ...the
right answer in such a method.
>>
>> With burial, a potentially  successful buial can succeed or fail,
depending on whether the CWs's voters plump. Either way, the CWs doesn't
win.
>>
>> (You wrote):
>>
>> ; the strategy is effective if the CW faction plumps, and backfires if
the CW faction cooperates with the weaker wing. Note that even a strategy
backfire here is not in my opinion a "good" result; because the strategy
has some chance of winning, it will be sometimes attempted. Strategic
backfire is a good thing if it discourages the strategy, but a bad thing if
it actually occurs.
>>
>> Yes, that's what makes Condorcet at least a bit questionable.
...especially with the possibility of the perpetual burial fiasco.
>>
>> But it can probably be argued that, with rank methods' goal of getting
the best individual candidate you can get, wv strategy improves on Bucklin,
because, even if people misjudge who the CWs is, and so hir voters don't
plump, the mere threat of plumping could be enough to deter burial, because
the would-be buriers don't have better information than the defenders have.
>>
>> > -In most basic Bucklin systems, the CW faction must defensively plump
to counteract SHOS.
>>
>> Yes, in Bucklin, if you aren't majority-favored, and if you rank
sincerely, and if the CWs's voters don't plump, then that can result in the
election of someone in your bottom-set.
>>
>> (You wrote):
>>
>> > -In a system like SARA, in most cases, even if the CW does not
defensively plump, the weaker wing can give enough support to the CW to
protect them from SHOS, without having to rate them equal-top. (If the CW
is only barely a CW, the weak wing may have to rate equal-top for a
successful defense in this case. However, I think that very
tightly-balanced situations like this are not a large concern.)
>>
>> (endquote)
>>
>> That makes me sit up & take notice, because such an improvement on
Bucklin strategy, while keeping FBC, would be major, & could outweigh the
lack of CD enough to compete with the best methods.
>>
>> Michael Ossipoff
>>
>>
>
>
> ----
> Election-Methods mailing list - see http://electorama.com/em for list info
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20161025/1b719fc1/attachment-0001.htm>