[Election-Methods] Election-Methods Digest, Vol 45, Issue 18

[Greg Dennis]

Great comments, Steve. I think you're basically right, though I'll
mildly disagree on a couple things . . .

>  However, IRV is worse at eliminating spoilers than some other methods.
>  It also undermines candidates who take centrist compromise positions, by
>  defeating them and making them appear unpopular.  As a consequence, we
>  can expect IRV would continue the "two big polarized parties, each
>  nominating one candidate per office" system (including its haphazard
>  primary elections).

It's true that IRV will not satisfy IIA to the degree that Condorcet
methods will. It can fail to elect compromise candidates, but I'm not
too concerned about that because the available evidence from real
elections suggests it's exceedingly rare in practice. It has yet to
happen in any IRV election in the US for which we have data (San
Francisco, Burlington, Cambridge). Moreover, where it's implemented,
it tends to pave the way for PR-STV and coalition governments, not
increased polarization.

Also, while I agree that there could be some compromise candidates
that should win that do not under IRV, I'm not completely convinced
every compromise candidate should. The idea of electing an
inexperienced nice guy to an important administrative office because
he happens to be everyones second-choice, for example, makes me a bit
nervous. That said, we should continue to keep our eyes on the ranked
data as it accumulates to see what happens in practice.


>  IRV can be significantly improved by letting candidates withdraw from
>  contention after the votes are cast.  At the end of election day, the
>  votes would be published in a format that candidates (and others) can
>  download.  Then the candidates would be given a few days to decide
>  whether to withdraw.  They could use those days to calculate what the
>  result would be with or without themselves (and/or some other
>  candidates) in the voters' rankings, and to negotiate with supporters
>  and with other candidates about who, if anyone, should withdraw.  The
>  official winner would not be tallied until after the withdrawal period.

Yes, I believe Thomas Round was the first to propose that idea in his
thesis. His proposal would have allowed candidates to "stand down", as
he put it, and let a candidate with fewer first-choices take their
place in the next round. Interesting idea.

Putting aside the theory for a moment, as a political strategy, I
think it's advantageous we ride the IRV wave to greater third-party
participation and hopefully STV proportional representation. Once we
reach that goal, which will be difficult enough, we can take a closer
look at how often compromise candidates are being eliminated in early
rounds. If it's significant, we can look towards amending on the votes
are tallied, perhaps by moving to CPO-STV.

Thanks again for your thoughts.


>  -----------------
>  Greg @ Somerville for IRV wrote:
>  > I don't believe that Range Voting will eliminate even the kind of
>  > spoilers that IRV does away with. Consider two-candidate race between
>  > Bush and Gore in which 51 voters prefer Gore to Bush and 49 prefer
>  > Bush to Gore. Following the directions given on the RangeVoting.org
>  > website, voters should give a '10' to their favorite candidate and a
>  > '0' to their least favorite:
>  >
>  >   51 people rate Gore=10, Bush=0
>  >   49 people rate Bush=10, Gore=0
>  >   the result: Gore=510 > Bush=490 (Gore wins)
>  >
>  > Now we re-run the election with Nader in the race. When Nader enters,
>  > 5 people for whom Gore was their first choice switch to preferring
>  > Nader. Those 5 voters rate Nader a '10' and push Gore down to '5':
>  >
>  >   46 people rate Gore=10, Bush=0, Nader=0
>  >   49 people rate Bush=10, Gore=0, Nader=0
>  >   5 people rate Nader=10, Gore=5, Bush=0
>  >   result: Bush=490 > Gore=485 > Nader=50 (Bush wins)
>  >
>  > How could the 5 Nader voters suddenly have a different utility for
>  > Gore? In reality, the rating a voter will give to a candidate isn't
>  > some fixed objective measurement of their happiness with a candidate,
>  > but a rating relative to the options available. In short, everyone
>  > will naturally grades on a curve. To say that RV satisfies IIA is to
>  > ignore the reality that ratings will inevitably be relative.
>  >
>  > Thus, Nader would still be labeled a spoiler, because his entrance
>  > into a race causes some to rate Gore lower.
>  >
>  > Agree?
>  >
>  > Greg