[EM] Turkey-raising. Options. MJ majority protection.

[Jameson Quinn]

2012/2/24 MIKE OSSIPOFF <nkklrp at hotmail.com>

>  Another comment on the turkey-raising criticism of
> conditionality-by-mutuality:
>
> Sure, if several rival candidates' supporters all give a conditional vote
> to the same loser, in order to get a mutual vote thereby, then that loser
> will
> get more votes in those transactions than any one of the rivals, and could
> thereby win.
>
> 1. This requires that only one loser is getting those conditional
> turkey-raising votes. Or at least that their number is significantly less
> than that of the
> rivals. In reality, if you're going to insincerely vote one loser over
> your rivals for that reason, you'd also vote the whole set of losers
> similarly.
>
> 2. This consequence of turkey-raising could also be referred to as
> backfiring burial. You know that when burial backfires in Beatpath, that's
> considered
> a virtue, a deterrent. Then it's a virtue and a deterrent in the
> mutually-conditional methods too. Voters attempting burial strategy are
> asking for it. Don't
> expect the method to protect voters practicing burial.
>

Would the method be even worse if burial were a universally good strategy?
Yes. The problem is the burial; the DH3 is the symptom. Your argument here
is like saying "Yes, this food causes vomiting. But that's a good thing,
because otherwise it would poison you."

>
> A more uncertain question is: As a vote management option in an Approval
> election, is there a different option, other than the MTAOC, MCAOC  and
> AOCBucklin options, that is more in the option-user's interest--without
> causing an FBC violation?
>

Yes, SODA, with a certain N^2 summable, polytime FBC fix for approval
ballots...  more on this later. But even without detailing the fix or
proving FBC for it, it's not hard to understand how delegation can be in
the voter's interest, both for fixing the chicken dilemma without causing a
burial incentive, and for voters who honestly would prefer to trust the
candidate than evaluate every option.


> I don't like to leave uncertainties or open questions when I leave this
> mailing list, but those uncertainties will remain when I (soon) quit the
> list, unless someone else
> has answered them by that time.
>
> MJ majority protections:
>
> Yes, a majority who rate x,y and z over a, b and c still does so even if
> it rates x, y and z differently, as is the case in ABucklin.
>
> But the fact remains that if you rate them all fairly closely, then
> there's a good chance that you're rating them all over where their
> medians would otherwise be, and, therefore, are not raising x,y and z's
> medians any more than you're raising a, b and c's medians.
>

Look in the previous paragraph: you said "a majority". And you were right
to do so, because we're discussing a property that includes this in the
definition.

If a majority ranks x, y, and z above a certain rating, and a, b, and c
below that rating, then they will raise the former group's scores above and
lower the latter group below that rating.


>
> So sure, it's like Bucklin in one respect, but there's a (sometimes good)
> chance that you aren't helping one candidate over the other at all.
>

The latter half of your sentence is irrelevant to the property being
discussed, and also just as true of Bucklin as of MJ.


>
> MJ is like Approval in which you have the option of only uncertainly
> (maybe and maybe not) raising a candidate's final count score
> against that of another candidate.  ...and/or of only uncertainly raising
> a candidate's final count score at all, instead of lowering it.
>

I've explained before that, once you have a history to go on, you will be
able to be virtually certain that the winning rating will not go outside of
a certain range, so that if you give a rating outside of that range, you
can be confident of its effect.


> In RV, too, if you rate a candidate below hir mean rating, you lower hir
> mean rating. And, additionally, the less extreme your rating,
> the less effect it has.
>
> But, as I said, I'd be glad for the enaction of MJ or RV,  (instead of
> Plurality, IRV, or any FBC-failing method) because they both essentially
> are Approval with the above options. You don't have to use those options.
>
> And they both meet FBC. RV also has Forest's solution to the ABE problem.
> Approval has it too, implemented probabilistically.
>

And MJ has it too. In MJ, if the chicken players are collectively a
majority (which is true in all common statements of the dilemma), they can
rate all non-chicken candidates at F and use Forest's solution for voting
the other chicken group at D/F. The largest chicken group wins.


>
> So I stand by my ranking of:
>
> 1. Approval
> 2. RV
> 3. MJ
>
> --among the methods in Ruderman's poll.
>
> A more complete merit ranking:
>
> 1. MMPO2
> 2. MDDTR
> 3. optionally-conditional methods
> 4. automatically-conditional methods
> 5. ICT
>
> (Approval and Bucklin versions below are ordinary, don't have
> conditionality)
>
> 6. ABucklin
> 7. Approval, MTA and MCA
> 8. RV
> 9. MJ
>

Again, you've forgotten to even list SODA.


>
> Among the optionally or automatically conditional
> methods, I'd rank their Approval, MTA, MCA and ABucklin
> versions in the same order as ordinary Approval, MTA, MCA and
> ABucklin are ranked.
>
> For public proposals, though, I'd say that ordinary Approval is at the
> top, with AOC next.
>
> Though AOC is better, brings big improvement to Approval, it's also true
> that ordinary Approval already
> had at least 3 other ways to deal with co-operation/defection:
>
> 1. Public declaration of principled refusal to accept a compromise or
> co-operate with a faction (Maybe under
> specified conditions)
>
> 2. The consequences, in subsequent elections, of defection
>
> 3. Forest's solution (probabilistically-implemented in Approval)
>
> Regarding #2, a good strategy for a defected-against faction would be to
> refuse to help the defectors
> in the next election. Then give them another chance to co-operate in the
> election after that. Factions should make
> it clear that that will be their strategy.
>
> In the co-operation/defection tournament for computer programs, described
> in Scientific American some time ago,
> the winner was a program called "Tit-For-Tat". I believe that the strategy
> that I described in the above paragraph
> is the Tit-For-Tat strategy: Vote the same strategy that the other faction
> did in the previous election.
>
> But SciAm later described another strategy that worked even better than
> Tit-For-Tat. I don't remember what it was,
> but it was nearly as simple as Tit-For-Tat.
>
> The point is that there are good deterrent strategies for
> co-operation/defection.
>

Tit-for-tat is a two-player strategy. It's not even clear what it means in
a multiplayer context, much less that you can apply results from two-player
studies.


>
> The broader point is that, even without AOC's conditionality, ordinary
> Approval can deal very well with the
> co-operation/defection problem. So ordinary Approval has merit, as will as
> winnability, as a first proposal.
>
> Mike 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/20120224/a62affeb/attachment-0003.htm>