[EM] Why IRV is better than Condorcet

[Bill Clark]

[Note: I recently switched my list address to my new gmail account,
and because of some confusion on my part about how it works,
mistakenly took this discussion off-line when I meant to keep it on
the list.  Next I'll forward my earlier post, to which Warren is
replying below.  Incidentally, is there any way to set the default
Reply-To: field to the list address, so that when I hit "reply" it
actually goes to the list, rather than the author of the post to which
I'm replying?]


On Sun, 8 Aug 2004 12:34:55 -0400 (EDT), Warren Schudy <wschudy at wpi.edu> wrote:

> A system is unpredictable if you cannot determine how the system will
> behave without doing too much work. (Yes, too much is in the eye of the
> beholder. Squaring a number is usually considered an ok amount of work,
> but evaluating a large summation term by term is often too much).

> IRV is this way - most questions about an IRV election cannot be answered
> without running a simulation.

How is this any different from any other election system?

The paper under discussion (
http://www.isye.gatech.edu/~jjb/papers/stv.pdf ) only purports to
demonstrate that one particular question is difficult to answer --
namely, whether or not an optimal strategy exists for IRV, given any
particular set of voter preferences.

Anyway, I'm still not clear on what you mean by "unpredictable" since
it would seem to me that even as simple a question as who won the
election would often require "running a simulation" no matter what
election method were being used.  You could only get away with using
partial information if the election were particularly lopsided.

I think perhaps I could understand your point better if you could
provide some examples for comparison.  What are some other questions
for which IRV (but not other systems) exhibits "unpredictable"
behavior?

-Bill Clark