Re: [EM] 2-step CR: A proposal

[Alex Small]

> However, if I understand correctly, the reason for increasing the
> number of levels is to give less incentive to rate compromise right up
> there with favorite.  This could be accomplished with 3 level CR by
> assigning values of 10, 9, and 0 to the three levels.

That is certainly simpler.  The main objection I have is that it doesn't
provide the same level of flexibility.  We can arbitrarily decide to go
with {10,9,0} or {5,4,0} or even {2,1,0} as the values of each of the 3
ratings (assuming nobody gets a majority of 1st place ratings).  The
question of which is best has no clear answer in my mind.  Rather than
trying to sell any particular set of values, by using CR we give each
individual voter the decision.

Some voters might want their lower-tier votes to have maximum impact if
they are used, and in that case they might decide to rate every candidate
as 10, 9, or 0.  Others might want to reserve more advantage to their
favorite, and assign ratings across the whole scale.

On a tangentially related note, I know it's been argued that CR and
Approval are strategically equivalent.  I have to beg to differ, at least
for the case where voters or blocs of voters have only partial information:

Suppose that in a CR election with 3 candidates the voters are trying to
decide their strategies.  Suppose that polling data makes it quite clear
that most people have already made up their minds about my least favorite
and compromise candidates, but it's impossible to tell right now how my
favorite will fare.  Suppose also that my least favorite is slightly ahead
of my compromise.

People who share my set of relative preferences have a strong incentive to
give our compromise candidate just enough votes so that he barely finishes
ahead of our least favorite.  So, on a scale of 0 to 10 we might all rate
our compromise as a 3.  We rate our favorite a 10, of course.  If enough
people from other camps rate our favorite highly then he'll win.
Otherwise, we'll at least block the election of our least favorite.

All of this assumes that we can coordinate, of course.

One other note:  Demorep once asked what scenario would generate
probabilistic strategies.  Suppose that the above election had been held
with Approval rather than CR.  In that case everybody who shares my
preference order would give a vote to our favorite, and 30% of us would
also give a vote to our compromise so he finishes just ahead of our least
favorite.

Alex

----
For more information about this list (subscribe, unsubscribe, FAQ, etc), 
please see http://www.eskimo.com/~robla/em