[EM] Critique of FairVote's "approval voting" report

[Kristofer Munsterhjelm]

robert bristow-johnson wrote:
> dunno if i can do much critiquing of that particular doc.  what i
> like is in FairVote's page:
> 
> http://www.fairvote.org/single-winner-voting-method-comparison-chart
> 
> 
> where they claim that IRV will do a better job getting the Condorcet
> winner than does Condorcet (sometimes the Condorcet method will
> *fail* to elect the Condorcet winner for those who didn't know that):
> 
> "IRV will generally elect a Condorcet winner, ... IRV may actually do
> a better job of electing Condorcet winners that nominal Condorcet
> voting methods, because of the incentives for strategic voting under
> Condorcet rules that are absent under IRV. ... Condorcet voting is
> designed specifically to find and elect a Condorcet winner whenever
> such a candidate exists. Ironically, due to incentives for strategic
> voting inherent in Condorcet methods, they may in fact fail to elect
> the Condorcet winner, even when one exists."
> 
> i know, so up is down, black is white, etc.

It's not so surprising that they claim this. They can't claim that IRV 
passes Condorcet, because it's an easily demonstrable fact that IRV 
doesn't - any ballot set that leads to IRV electing a different 
candidate than the CW constitutes a proof that IRV fails the Condorcet 
criterion.

Therefore, they have to focus their efforts along two lines: that the 
Condorcet criterion is undesirable and so that IRV failing Condorcet is 
a *good* thing, or that while the letter-of-the-law is that IRV fails 
Condorcet, if one alters the playing field, then the situation changes 
to the benefit of IRV.

The first approach is done through emphasis on "weak winners" and 
"Condorcet winners that are nobody's favorite", "core support", and 
similar objections. FV asks us to envision a flip-flopper that is bland 
enough to not be greatly disliked and so wins even though nobody liked 
him, either, and then they imply that if you choose a Condorcet method, 
that's what you'll get.

The second approach is what you see here. By redefining the playing 
field to be "in the case of strategic voters", they can say that IRV 
passes Condorcet and does so even more often than does ordinary 
Condorcet methods. Even if that doesn't hold, it's no longer a matter of 
outright fact any more - simulating strategic behavior is hard and so 
they might claim that IRV elects Condorcet winners more often under 
certain models, and then redirect the argument to a more exotic one 
about which models are "realistic".

I don't even think the claim is correct. If strategy resilience makes 
IRV elect Condorcet winners more often than Condorcet, and JGA's paper 
suggests one can preserve IRV's strategy resistance while also getting 
Condorcet (by prefixing IRV with logic that turns it Condorcet 
compliant), then it is not at all clear that the Condorcet-IRV methods 
would rarely elect true Condorcet winners. Both of the methods JGA 
investigated resisted simple strategy better than did the "unadorned" IRV.

Incidentally, Rob Richie commented (on the discussion page for Approval 
voting on Wikipedia) that to consider restricted situations where 
Approval voting would pass certain criteria it otherwise would not, only 
muddied the waters and so shouldn't be done. Yet, this "more CW than 
Condorcet" seems to do something similar: one considers a voting method 
over a limited subset of voting behavior, and then states that the 
method passes a certain criterion (which it might, given the limited 
subset) that it otherwise would not.

(Also, FairVote is simply wrong when they say "Only the plurality, 
two-round runoffs and IRV have ever actually been used in U.S. 
governmental elections". If they only consider national/presidential 
elections, then neither top-two nor IRV has been used; and if they 
consider local elections, too, then Nanson's method, which they consider 
a Condorcet method, has been used.)