[EM] (3) MJ -- The easiest method to 'tolerate'

[Kristofer Munsterhjelm]

On 09/22/2016 03:38 AM, Jameson Quinn wrote:
> 
> 
> 2016-09-21 18:59 GMT-04:00 Kristofer Munsterhjelm <km_elmet at t-online.de
> <mailto:km_elmet at t-online.de>>:
> 
>     On 09/06/2016 01:29 AM, Jameson Quinn wrote:
>     >     ...
>     >     If that's true, then there's some kind of threshold of number of grades,
>     >     below which there aren't enough grades for the ballot format to
>     >     encourage grading against a common standard. So three-slot methods would
>     >     have to be tested to see if voters would vote grading-style or
>     >     relative-rank style with three grades, or if three grades are still
>     >     too few.
>     >
>     >
>     > Personally, I'd guess that three would be enough. I think the four most
>     > common voting heuristics, in descending order, would be:
>     >
>     > First, as I've said: approve of one, disqualify greater evil among the
>     > frontrunners and any evils greater still.
>     >
>     > Second: Approve of one, disqualify both frontrunners.
>     >
>     > Third: if one side of the left-right divide got used to habitually
>     > losing, the centrists among them might begin to extend approval all the
>     > way to the most-centrist on the other side.
>     >
>     > Fourth: Approve of one, disqualify unqualified/unserious candidates.
>     >
>     > Any of the heuristics above would tend to lead to a "successful"
>     > resolution of the Chicken Dilemma.
> 
>     That's kind of going into the domain of "manual DSV", though, which I'm
>     not as much a fan of. MJ has this going for it that there are ways for
>     honest voters to vote as long as the categories are well-defined.
>     Approval is a *lot* muddier, because it's much less certain that honest
>     voters can divide the candidates into "I like these" and "I definitely
>     don't like these". Three-slot methods could work if the heuristics are
>     common and intuitive enough, but they're kind of in-between MJ's
>     expressiveness and having Approval's manual DSV needs.
> 
> 
> Median-based methods, especially those which break ties using
> above-median votes, can do well with chicken dilemma scenarios. The
> flipside of this is that in order to handle center-squeeze scenarios,
> they require the voters who favor the centrist CW to put all other
> candidates below the winning median; for safety, at bottom-rating.
> 
> I think that's somewhat plausible as a naive strategy ("Since I'm
> Center, I think Left and Right are equally bad, so even though I don't
> hate them as much as they hate each other, I might as well put both of
> them at bottom rank"). I also think that when Center voters don't so
> strategize, the outcome is not too bad; generally, that will tend to be
> in cases where Center might even have lost a score election, and the
> candidate who beats Center might even be the utility winner or at least
> close to it.
> 
> But yes, there is this much "manual" strategy necessary, and I don't see
> any way around it. I think a system can't be robust to strategy in both
> Chicken and Center Squeeze scenarios, and I'd rather get Chicken right
> in spite of some strategy (and thus not encourage strategies which, when
> over-applied, can end up electing the CL) instead of getting Center
> Squeeze right without any strategy.

Do you think there's a way to formalize vulnerability to center squeeze?
I imagine there would be two criteria:

- No center squeeze outright (e.g. IRV)
- No center squeeze with strategy

And I suppose for the latter, you *could* make a rather heavyhanded
criterion saying something like: if C is center and strategy can make
not-C win but not make C win, then it's vulnerable to strategic center
squeeze...

but that doesn't feel very elegant. Any ideas? Perhaps something
involving votes being drawn from a spatial model so that the idea of
"center" is relatively apparent?

The point is that if we had a formal model, I could check what methods
pass the criterion (or often pass it), and then see if there's anything
that both passes CD and whatever it ends up being.

Do you think hybrids that use both Condorcet and positional data (like
my fpA-fpC stub method or Benham/Woodall methods) pass both CD and
center squeeze? They do at least elect the centrist when there is a CW.
I guess the question would be how badly they degrade when there is no CW.