[EM] The general form of Quick Runoff

[Kevin Venzke]

Hi Jameson,

--- En date de : Lun 24.5.10, Jameson Quinn <jameson.quinn at gmail.com> a écrit :
>I like this method, and would like to discuss the behavior with the simple 
>three-candidate "center squeeze" scenario, as compared to IRV.
>
>If IRV falls victim to center squeeze, then the centrist candidate is 
>candidate C in QR. Thus, for them to win, two things must happen: B has a 
>majority win over A, and C over B. If you locate the three in one-
>dimensional issue space, and label the median voter X, this means 
>something like the following:
>
>...----B-C-X--A-----...
>
>If C is at the median voter, then they cannot win unstrategically, because 
>whichever of the other two candidates is closer to the median will win 
>both first preferences and pairwise against the other one. Thus, C must be 
>on the same side of the median as B, in order to give A "more room" to win 
>the first preferences.

Yes. I think it's quite typical that the first-preference winner would 
be on the opposite side of the median as the centrist/smaller clone.

>Can a truly-centrist (median) candidate win strategically? If C voters 
>know how the other two candidates will fall in first preferences, their 
>best vote for electing C will be C>B>A, even if their honest preference is 
>C>A>B. Note that this is fundamentally a dangerous strategy, as, in order 
>to work, it depends on B beating A; and so if C does not beat B, it helps 
>the worst candidate B win the election. Thus, the A>C>B voters can 
>spitefully defend against this by announcing their intention to bullet-
>vote for A. This defense means nothing to the honest C>B>A voters, who 
>still prefer B to A; but it does deter strategic C>B>A votes from honestly-
>C>A>B voters. Thus, if the A voters' threat is plausible, a truly->centrist 
>candidate is still vulnerable to center squeeze, and cannot win.

Yes, though I have two thoughts:
1. It is odd that, with a truly centrist candidate, it can be predicted
which of the other two candidates will place first. This seems like an
error by whoever nominated B.
2. When this strategy backfires, it at least elects a candidate with
*more* first preferences than intended, rather than fewer, which is
typical with burial strategies in other methods.

I can imagine that someone who favors a weaker candidate could attempt
to vote lower preferences in completely reversed first-preference order.
But he has to calculate the risk that he fails to deliver the win all
the way to the target.

And on the other side, supporters of stronger candidates could believe
that there is no way certain preferences should be necessary in a given
*sincere* election, and so decline to state the lower preference.

>So really, the advantage of QR over IRV is not so much in traditional 
>center squeeze, as in "clonelike center squeeze". That is, where IRV (like 
>closed primaries) often elects the more-radical of two near-clones, QR 
>generally elects the more-centrist one.

Yes, I agree, but I don't find this very disappointing because I think
this would often be the situation.

>(Note that "DNA" analysis misses this point because it never includes >both 
>C>B>A and C>A>B preferences in the same scenario. I like the "DNA" idea, 
>but this is its weakness.)

It does miss that, which is why it can't distinguish DSC/SPST, but it
does tell you that A and B can't bury, and it could tell you that C could
bury, if I had programmed that as a criterion.

Kevin