[EM] why can't we have the Ranked Ballot (even IRV) for primaries?

[Kristofer Munsterhjelm]

Raph Frank wrote:
> On Wed, Aug 25, 2010 at 9:41 PM, Kristofer Munsterhjelm
> <km-elmet at broadpark.no> wrote:
>> Third, the primary is not open and so
>> even if a good ranked method were used, it would elect the candidate closest
>> to the party's median, not that of the electorate in general.
> 
> Not necessarily.  The candidates could easily argue (as now), that
> they have a better change of being elected if they are closer to the
> national median.  Party voters would have to trade off getting a
> candidate who reasonably represents the party's views with one who has
> a reasonable change of being elected.

That's true, but then the voters are acting strategically. Although the 
strategy isn't the familiary selfish sort, they are reporting a 
different rank than if they were to just rank by their own preference.

> Another option would be to give party members 2 votes and everyone
> else 1 vote.  This would give a median that is between the national
> and party median.

You probably have a continuum here. At one end, you would have the party 
leadership just decide upon the common candidate. A little further, you 
have an ordinary (closed primary), in effect giving party members 1 vote 
and everybody else 0; then you have all sorts of weighting up to open (1 
vote for members and 1 for everybody else), and if you go even further, 
no primary at all.

With methods like Schulze and RP, an open primary distorts the final 
election less than in Plurality, because these methods fail IIA less 
than does Plurality. That is, if you have an open primary and the voters 
would vote identically in the real election (with respect to the 
candidates running in the primary) as in the primary, then that primary 
is closer to having no primary at all and just one election with every 
possible candidate than would be the case for Plurality.

The parties still might want to run primaries, though, so that they have 
one candidate to unify behind, run ads for, and so on - but on the 
scale, you'd expect the open primary and no primary points to be closer 
to each other for the advanced Condorcet methods (and probably cardinal 
ratings too) than for Plurality.