[EM] Voting-System Choice for Polls (Just one more thing I want to say)

[Michael Ossipoff]

Hi Juho--

It isn't that I can't stop, can't quit discussing voting-systems :^)  It's
just that this good idea occurred to me, because I'm polling on another
(non-political) subject at other websites, and I want the count to be as
strategy-proof as possible.

And so of course I wanted to mention the good idea (for poll
strategy-resistance) at EM.

I suggested Pairwise-Winner (MAM,  Smith//MMPO).

But, even though FBC isn't needed (because I want sincere rankings, and not
equal-top-ranking), it might still be undesirable if Smith//MMPO could do
things that the voter doesn't intend or like.  A ballot that gets X into
the Smith set (by eliminating one of her defeats by Smith-set members)
could allow hir strong pairwise opposition against Y (though Y pairbeats X)
to give the win to someone worse than both.

Maybe that's undesirable even if FBC isn't needed.

So maybe Pairwise-Winner (MAM,  Smith,MMPO)?

Even though Smith,MMPO could fail to elect a CWv, that's ok, because MAM
would elect hir, and s/he'd pairbeat the Smith,MMPO winner.

So, which would be better for polls:

Pairwise-Winner (MAM,   Smith//MMPO)

or

Pairwise-Winner (MAM,  Smith,MMPO)?

Forest?  Kevin?

Any suggestions? I want the best strategy-proofness for polls that might
need that, but without unnecessarily irrational, unintended and undesirable
results.

Left to myself on this, I'll probably go with

Pairwiise-Winner (MAM,   Smith,MMPO)

But what are some opinions on this?

Replying farther down:

On Sun, Dec 18, 2016 at 6:09 AM, Juho Laatu <juho.laatu at gmail.com> wrote:

> On 18 Dec 2016, at 08:05, Michael Ossipoff <email9648742 at gmail.com> wrote:
>
> Before quitting EM & retiring from voting-systems, there's one more thing
> I'd like to say:
>
> What's the best voting-system for polls?
>
>
> Mostly exactly the same as for elections.
>




> Polls may however often be less competitive than elections. (Although some
> elections can be less competitive than some polls.) This means that you
> could emphasize performance with sincere votes more, and put less weight on
> performance against strategic voters.
>

Yes, in some polling electorates, like CIVS. But in rank-balloting polls,
you want to find the CWs, and so you don't want equal-top-ranking. You
might even not allow equal-ranking. So FBC isn't needed.


>
>
> Though FBC is important for official political elections
>
>
> No, FBC is not important. It is important not to be vulnerable to
> "betrayal strategies", but that doesn't necessarily mean FBC compliance. :-)
>

Well, for elections, with all kinds of voters, especially under current
conditions, I'd want for even the worst overcompromisers to not have any
incentive to vote someone over their favorite.  Besides, as an Approvalist,
I want the opportunity to vote fully-effective equal-top-ranking.

[Replying farther down] :


>
>
>
> As I've said, I haven't noticed any sign (top-cycles for 1st place) of
> strategy in polls.
>
>
> Has somebody sometimes noticed a strategically generated top-cycle or any
> other cycle in any elections or polls?
>

No, but in one EM poll, the other co-conductor of the poll told me that
someone had requested permission to vote anonymously, suggesting an
intention to use burial strategy.

At CIVS (Condorcet Internet Voting-Service) I've never noticed a top-cycle
for 1st place. But for lower-place, later-finishers, there are often
cycles, when there are lots of candidates, probably because of short
rankings, or people guessing about their lower choices.



>
>
> I suggest that, for sincere electorates, MAM is the ideal best. That means
> it's best for polls at the Condorcet Internet Voting Service, where there's
> been no sign of strategy.
>
>
> When strategic intentions are gone, or low enough, one should focus on
> which candidate is best according to the needs of the society. I have no
> good idea on what MAM tries to achieve (= what kind of winner to elect)
> with sincere votes. In some polls Range could be the best (if you want the
> winner to have high average support), or maybe Approval (= elect the most
> approved candidate) or some Condorcet (= elect the one who is preferred
> over the others) method.
>

Yes, certainly, for elections. But when I use a rank-method for polling,
it's because I want to find out the CWs. That's my goal with a
rank-balloting poll.

[Replying farther down] :


>
> But what if you're doing a poll among people who are highly involved in
> the subject that you're polling about, and have strong committment to some
> alternatives? Like, for example, suppose you're pollng at EM about
> voting-systems?
>
>
> I guess you are talking about competitive polls here. I.e. EM members are
> expected not only to have strong opinions on the alternatives but also
> strong intentions to make their favourite win (by voting strategically).
>

Yes, that would be the concern.


>
>
> So you can't really be sure that there'll be no chicken-dilemma defection.
>
>
> Among this group of people, I'm sure you couldn't be sure of anything :-)
> , and you should protect yourself against all possible strategies. Or at
> least against those strategies that EM people are likely to master. That is
> probably a smaller set than the largest set of all theoretically possible
> strategies. :-)
>

Invulnerability to truncation & chicken-dilemma defection, and resistance
to (& penalizing of) burial would be good enough. That's what I'm trying
for with
Pairwise-Winner (MAM,   Smith,MMPO).




>
>
> So maybe, instead of MAM, Smith//MMPO should be used.
>
>
> Ok, now you are talking about highly competitive polls. Why did you
> introduce Smith set here? Is that to safeguard against strategies? I'm not
> knowledgeable enough to tell what kind of strategic threats Smith set would
> eliminate. Maybe you introduced it to have good performance with sincere
> votes.
>

Yes, partly.

>
> This is a sidetrack, but I must note that I'm not a believer in Smith set.
> My opinion is that it tries to force linearity to group opinions. You can
> expect that (linear opinions) from individual voters, but in groups there's
> no need to assume that all the Smith set members would be better than the
> other candidates (with sincere votes). You can as well think than losses to
> other candidates are a bad thing, irrespective of whether they occur in
> cycles or not. Losses within a cycle could be considered worse than losses
> outside of that cycle.
>

When offensive strategy is used against the CWs, s/he'll be in the voted
Smith-set. That's a good reason to elect from the voted Smith-set, so that
the CWs can maybe be rescued.

[Replying farther down] :


>
> It is automatically resistant to chicken-dilemma defection.
>
> But, for burial, it isn't as good as MAM. With MAM, a candidate that you,
> & sufficiently-many others, don't rank can't beat the CWs by burial.
>
> ...but it can in MMPO, though there's a lot of uncertainty & risk in
> trying burial in MMPO.
>
> Maybe Smith//MMPO's reliable automatic chicken-dilemma protection is more
> important, because defection is easier & less drastic a strategy than
> burial.
>
> But maybe MAM's better burial protection is more important, because burial
> temptation & opportunity is a lot more common than a chicken-dilemma
> situation.
>
> My suggestion: Use both.
>
> Do the count by Smith//MMPO, & by MAM. Of the winners by those 2 methods,
> the final winner is the one that pairwise-beats the other.
>
> That's a solid good solution, because:
>
> In the chicken-dilemma example, and also in a burial example, the intended
> victim of the offensive strategy pairbeats the perps' candidate.
>
> So, declaring, as winner, the one of those 2 winners that pairbeats the
> other is definitely the best solution, if MAM & Smith//MMPO are the best
> choices, each of which offers better protection in different ways.
>
> Of course, it's been pointed out that methods that elect the pair-winner,
> among the winners by 2 different methods, tend to fail FBC.
>
> But FBC isn't needed in polls, where you want sincere ranking, not
> equal-top-ranking.
>
>
> I'm not able to take position on how good that combination would be
> against strategies that might emerge among the EM list voters. The
> discussed environment is assumed to be very competitive, and I guess
> there's no difference any more on whether we had a poll or an election.
> Both would be competitive, and similar methods would be needed to make the
> system work.
>
> ...except that, for polling, I don't object to losing FBC for a
rank-balloting poll.

For elections, for a ranking-method, I like MMDA & MMDAsc.



> Or actually I do believe that often we do not really need extreme
> defensive measures, and putting more emphasis on electing the best winner
> with sincere votes is possible.
>

That seems so at CIVS. I always use MAM there.


> I mean for example that I have not seen any good general guidance (to
> regular voters with no exact knowledge on how other voters are going to
> vote) on how to vote strategically in some very basic Condorcet elections.
> My best guess is that even if EM list would vote on which single winner
> election method is the best, using Condorcet, I would not expect people to
> identify any good strategy that they could use to (effectively,
> intentionally) manipulate the results and change the winner from the
> sincere winner to some alternative that they like better.
>

Well, the feeling to want to do burial might be natural. Likewise
truncation, which might not even be strategically-intended. Llikewise
chicken-dilemma defection. People at EM would be the most familiar with
those strategies, but anyone could feel the incentive for them.

Most especially if you're polling, on some topic, a group of people who
have strongly-preferred  favorites among the alternatives




>
> For these reasons I tend to favour methods that are good with sincere
> votes, and that are designed to protect us only against some very basic and
> very obvious strategic voting opportunities. I'm somewhere at the level of
> "in most cases Condorcet is enough".
>

Certainly at CIVS. When there are strong preferences, when people might be
strongly committed to favorites, I'd like a little more in polls.

Pairwise-Winner (MAM,  Smith,MMPO)

or

Pairwise-Winner (MAM,  Smith//MMPO)

would be an effort to achieve that.

Any opinions regarding which of those two would be better?

Michael Ossipoff

>
> *B*R, Juho
>
>
> Michael Ossipoff
> ----
> Election-Methods mailing list - see http://electorama.com/em for list info
>
>
>
> ----
> 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/20161218/75a6ee55/attachment-0001.htm>