[EM] IRV vs Condorcet vs Range/Score

[Raph Frank]

On 10/13/08, Aaron Armitage <eutychus_slept at yahoo.com> wrote:
> Any strategically informed voter will give approval-like ratings, and will
>  approve the top two frontrunner he likes better, as well as any candidates
>  he prefers to that frontrunner, which means that the stated ratings will
>  look like that regardless of the real ones and a range election will tend
>  to produce that result regardless of how much utility voters attach to
>  each candidate. And, of course, IRV produces the same result.

Under perfect strategy, it doesn't actually matter what voting method
is used.  A condorcet winner would be elected every time.

However, in practice, strategy only occurs at a local level.  Each
voter decides how to vote based on how the other voters are going to
vote.

This means that the optimal strategy is a Nash equilibrum.  Even if C
is a condorcet winner, the majority won't switch to to him under
plurality if he isn't one of the top 2.

However, under approval, he will get approved by a majority of the
voters, as they will approve him in additiona to one of the top 2.

Also, with approval polls, a condorcet winner should become one of the
top 2 automatically.

A system should tend towards the condorcet (or utility) winner under a
process of poll -> poll -> ... -> poll  -> election.

A system that performs poorly under strategy fails that requirement.
Voters in the (early) poll stages would be reasonably likely to give
honest ratings and thus will move the result faster to the condorcet
winner.