[EM] More non-altruistic attacks on IRV usage.

[David L Wetzell]

On Fri, Dec 2, 2011 at 7:31 AM, Kristofer Munsterhjelm <km_elmet at lavabit.com
> wrote:

> We're still hitting the same disagreements. I say "look at the others",
> you say "this time it'll be different", I say "Condorcet >> IRV", you say
> marketing differences are great while in practice, there's no difference
> between Condorcet and IRV large enough to make a difference.
>
> Thus, let me do some asking, because we're not getting anywhere. Consider
> in your mind: what kind of data could I show that would change your mind
> about whether IRV is stronger in the hegemonic direction than PR is in the
> enabling-contesting-parties direction?
>
> Furthermore: On what do you base that reality is:
>
> 0 Plurality
> 0.7 IRV
> 0.72 Condorcet
>
> rather than:
>
> 0 Plurality
> 0.25 IRV
> 0.72 Condorcet?
>
> You keep saying that X_Condorcet - X_IRV is small. Is that just a belief,
> or do you have something on which to support it?
>

Here's an 'AV faith article that looks at how the use of IRV would have
changed outcomes in UK MP elections.
http://www.newstatesman.com/uk-politics/2011/04/seats-party-election-majority

For a heuristic, we could consider the Bayesian Regret measures with only 4
candidates, instead of the 7 candidates typically used.  Condorcet doesn't
do best in this procedure, but FPTP and purely random voting do a lot
better with only 4 serious candidates, as is more realistic for
single-winner/member elections.  Thus, if the worse election rule does
considerably better under more realistic assumptions, it stands to reason
that the diffs among all of the election rules will be lowered considerably.

Let's say that in a close 3-way election that it's .25 IRV and .75 Cond.
 Let's say that in other elections that it's .70 IRV and .72 Cond then the
appropos question is how often are there 3-way close elections?  In the US,
not very often and that is the context that I am presuming.  So the
weighted average is going to be closer to .7 IRV and .72 Cond.

Moreover, the "biases" from IRV will get averaged out over time and place
and a bias against centrists won't matter so much if the biases to the
right and left cancel...  This makes the X_IRV closer to X_other alts to
FPTP.

I'll reply to the below later.

> dlw
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20111202/2ed34520/attachment-0004.htm>