[EM] Symmetric ICT reformulation and exploration

[Michael Ossipoff]

I forgot to add Mono-Add-Plump to the advantages of Buclin over MDDTR. So
it should say:

Bucklin:

* Mono-Add-Plump

* Use-Precedence

* Easier protection of the CWs

MDDTR:

*CD

*LNHa

* Precinct-Summabilty

-----------------------------------------

If Bucklin's easier protection is in question, then the comparison is
especially more favorable to MDDTR.

Michael Ossipoff



On Sat, Nov 12, 2016 at 4:32 PM, Michael Ossipoff <email9648742 at gmail.com>
wrote:

> Yes, that 2/3 majority rule would avoid having to say:
>
> "(If each candidate has someone rated over hir by a majority, then the
> winner is the most top-rated candidate.)"
>
> Tantalizingly greater simplicity, regrettably not workable, as you said.
>
> ICT would avoid the Mono-Add-Plump criticism, but at the cost of
> truncation-vulnerability.
> I'd rather have the Mono-Add-Plump criticism and truncation-proofness.
>
> The ICT wording you described comes closer to the brevity of MDDTR, but
> MDDTR doesn't need a separate "beat" definition at all. ...just the use of
> the already-understood majority.
>
> I suggested the 3-slot version of MDDTR because I felt that it should be
> used only as an Approval-version, with the Middle rating reserved for the
> special chicken-dilemma situation.   ...because I felt that MDDTR, with its
> complete vulnerability to burial (like every pairwise-count method),
> wouldn't be good as a rank-method.
>
> But maybe that should be reconsidered. Why would it be worse to rank your
> inbetween candidates in order of preference, than to rate them all together
> at middle?
>
> The burial-vulnerability, the fact that "wv-like" didn't mean as much as
> I'd believed it did,  was such a disappointment that it at first made me
> not appreciate the fact that MDDTR still has truncation-resistance. Burial
> vulnerability isn't a complete disaster:
>
> For one thing, to bury the CWs, you have to know who is the CWs. And if
> you know it, then the defending wing knows it too, because the same
> predictive information is available to everyone.
>
> If the CWs is more  with your wing, and it's the opposite wing that
> dislikes the CWs, and is likely to bury, you can prevent successful burial
> by equal-top-ranking the CWs. That was pointed out a long time ago, as
> general pairwise-count defensive strategy.
>
> You could protect the CWs in that way in Bucklin too. (in case people
> might rank past the CWs).
>
> Of course the difference is that, in Buclin you & the others in your wing
> can also just avoid ranking past the CWse (expected or evident CWs).
>
> MDDTR, and pairwise-count methods in general, don't have that protection,
> and you only have the defensive strategy of equal-top-ranking the CWse.
>
> So, as regards protection of the CWs, Bucklin is better than the
> pairwise-count methods. MDDTR's tradeoff-advantage is its CD.  ...in return
> for being able to protect the CWs only by equal-top-ranking.
>
> Conditional Bucklin's and Conditional Approval's FBC failure is of a
> different kind than Condorcet's FBC failure, it seems to me. With
> Conditional Bucklin, the effectiveness of my equal-top-ranking isn't
> diminished by the FBC failure. The FBC failure merely gives me a trick that
> I could use, with sufficient predictive information, to gain advantage. Not
> a problem. But the problem is that the serious overcompromiser would still
> have incentive to rank Hillary alone at top, over the overompromiser's
> favorite. So I guess I reluctantly have to not advocate Conditional Bucklin
> or Conditional Approval.
>
> ...meaning that evidently Bucklin can't have CD, and MDDTR's CD is an
> advantage for MDDTR over Bucklin.  So it's MDDTR's CD vs Bucklin's easier
> protection of the CWs.
>
> All that time I was calling wv burial-deterrent, because of the
> 3-candidate example, where deterrence is achieved by merely not ranking the
> would-be buriers' candidate, I never considered a 4-candidate example.
>
> It's easy to make a 4-candidate example where, in wv (& in MDDTR), that
> defense won't work: If B is the CWs, and the A voters are going to bury by
> insincerely ranking C over B, then just add D, between B and C.
>
> The halfway point between D & C is of course way C-ward from the median,
> and so D will have a majority against C, if voters are
> uniformly-distributed. (...and probably could, with suitable
> distance-relations, even if the voter-distribution is Gaussian).
>
> So, when the A voters make B majority-beaten, by ranking C over B, C
> remains majority-beaten (by D), and so now everyone is majority-beaten, and
> A wins if s/he's the most top-rated.
>
> I haven't looked at what it would take for A to win in wv, because I no
> longer propose wv for official public political elections (or any where
> offensive strategy is likely). That would be something for a wv advocate to
> discuss.
>
> Maybe I should list, together, some relative advantages of Bucklin & MDDTR:
>
> Bucklin:
>
> * Easier protection of the CWs (relevant if one of your inbetween might be
> the CWs)
>
> * Use-precedence
>
> MDDTR:
>
> * CD
>
> * LNHa
>
> * Precinct-Summability
>
> ----------------------------------------------------------------
>
> As for Bucklin's easier protection of the CWs, I'm not entirely sure,
> because there's some reason for the individual to rank all of the best
> candidates (instead of only ranking down to the CWse), to improve,
> somewhat, the probabiliy of electing one of them (but not as much as
> equal-top-ranking them all). So I don't know if Bucklin's easier protection
> of the CWs would materialize..
>
> Michael Ossipoff
>
>
>
>
>
>
>
> Michael Ossipoff
>
>
>
>
>
> Maybe a toss-up. Bucklin has the use-precedence advantage, but MDDTR has
> the precinct-summability adantage. And I consider MDDTR's LNHa an advantage
> too, when you don't have to hesitate to append less-liked inbetweens to
> your ranking for fear that you'll help the beat better candidaes.
>
> In Bucklin, when skipping is permitted, you could make sure that, above
> some inbetween, you skip enough levels that the better candidates will have
> enough rounds to accumulate the coalescing lower-choices that are coming to
> them from other candidates' preferrers.
>
> In MDDTR, & maybe in Bucklin, I'd likely top-rate the CWse, along with the
> very best of the strong top-set, even if s/he isn't really among those, and
> even if I felt like down-rating some of that top-set a bit because of some
> fault, or because of likely defection-inclination of their voters.
>
>
>
>
>
>
> On Sat, Nov 12, 2016 at 2:56 PM, Forest Simmons <fsimmons at pcc.edu> wrote:
>
>> Perhaps we could modify (non-symmetric) ICT in order to have a less wordy
>> definition of "strongly beat."
>>
>> Candidate X *strongly beats* candidate Y iff  X is preferred over Y on
>> more ballots than Y is* ranked* equal to or above X.
>>
>> All strongly beaten candidates are disqualified unless that would
>> disqualify all of them.
>>
>> Elect the qualified candidate ranked top on the most ballots.
>>
>> This definition makes it slightly harder for X to strongly beat Y than in
>> standard ICT, because all equal rankings have to be overcome, not only
>> those at the top.
>>
>> But it changes nothing in our standard CD examples, because in those
>> examples there are no equal rankings (only equal truncations, which don't
>> contribute to the strongly beat definition).
>>
>> It should preserve the FBC and perhaps even introduce a stronger
>> property: if some candidate X is raised to the level of the winner on some
>> ballots, then the winner is unchanged unless the new winner is X.
>>
>> I see the wisdom in saying "disqualified" instead of "eliminated."  If we
>> said "eliminated," then some people would wrongly think that "favorite"
>> refers to the highest among the remaining candidates (after their original
>> favorite was stricken from the ballot).
>>
>> Also a comment about three slot methods in general:
>>
>> With three slots it is impossible for every candidate to be eliminated by
>> a two-thirds majority.  So the following method would be even simpler to
>> define in the context of 3 slot ballots:
>>
>> Elect the favorite candidate who is not beaten by a two-thirds majority.
>>
>> Of course, for all practical purposes that would be the same as "elect
>> the candidate ranked top on the greatest number of ballots," which doesn't
>> satisfy the CD criterion.
>>
>>
>> On Sat, Nov 12, 2016 at 9:07 AM, Michael Ossipoff <email9648742 at gmail.com
>> > wrote:
>>
>>> Yes, but ICT defines "beat" in a wordier way, that people hear as
>>> complicated.
>>>
>>> For people who are into voting-systems, I can say "majority-beaten", &
>>> they know what I mean...that I'm talking about pairwise defeats.
>>>
>>> So, here's how I'd define 3-Slot MDDTR, to the public:
>>>
>>> You rate each candidate as  "Top", "Middle", or "Bottom". If you don't
>>> rate someone, that counts as rating hir at Bottom.
>>>
>>> The winner is the most    favorite candidate who doesn't have anyone
>>> rated over hir by a majority.
>>>
>>> (If everyone has someone rated over hir by a majority, then the winner
>>> is the most favorite candidate.)
>>>
>>> (end of definition)
>>>
>>> I'd just call it " Majority Disqualification".
>>>
>>> Michael Ossipoff
>>> On Nov 11, 2016 4:39 PM, "Forest Simmons" <fsimmons at pcc.edu> wrote:
>>>
>>>> You wrote in part ...
>>>>
>>>> >Another advantage that it has over 3-Slot ICT is that 3-Slot MDDTR has
>>>> a much >simpler definition:
>>>>
>>>> >The winner is the most favorite candidate who isn't majority-beaten.
>>>>
>>>> Three slot ICT could be defined in the same way;
>>>>
>>>> Elect the most favorite candidate who isn't strongly beaten.
>>>>
>>>> Neither definition tells what to do when every candidate is beaten
>>>> (majority beaten or strongly beaten, respectively).  But that is just a
>>>> detail of the definition that doesn't have to be mentioned immediately.
>>>>
>>>> Here's a more complete definition that works in both cases:
>>>>
>>>> Eliminate all candidates that are {majority, strongly} beaten unless
>>>> that would eliminate all candidates.  Elect the most favorite among the
>>>> remaining.
>>>>
>>>> So ordinary ICT and MDDTR are equally easy to define.  It's a matter of
>>>> which has the best properties.
>>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20161112/699734ca/attachment.htm>