[EM] CR vs. Condorcet

[Forest Simmons]

See below for my two cents worth.

On Thu, 28 Nov 2002, Douglas Greene wrote:

> This is primarily directed to Mike, but I'd like to know why list members
> support Condorcet over cardinal rankings/range voting.  BTW, I've posted
> Warren Smith's work on range voting to our Yahoo!Group.

There are at least two questions here.  (1) What kind of ballot? and (2)
How to use the ballot information to pick the winner?

Personally, I prefer the greater freedom of the range/CR/grade ballot, but
not the scoring method that is usually assumed as part of the package,
namely giving the win to the candidate with the highest average
rating/grade.

The trouble with that scoring method is that voters soon see the
advantages of rating/grading almost exclusively at the extremes of the
allowed range.  The strategic cost of greater expressivity is high.

A better way of scoring is to count all ratings above midrange as passing
grades, and all below as failing.  The candidate with the highest number
of passing grades wins the election.  This method is statistically and
strategically equivalent to the standard version of CR, but does not
penalize expressivity.

There are other ways of scoring these types of ballots that may offer
additional advantages, including ways that give the win to the pairwise
beats all candidate if there is one, in other words, ways satisfying
various possible generalized Condorcet Criteria.

Grade ballots are simpler, less confusing, and more flexible than ballots
that require strict rankings (with or without possible truncations), and
any reasonable method of scoring a ranked ballot can be adapted to grade
ballots.

The rank scoring method that is most analogous to the highest average
rating is the highest average ranking, i.e. the Borda Count.  The
inflative pressure to the extremes in the CR version shows up in Borda as
high pressure incentive for insincere ranking.

In the context of rankings, good Condorcet scoring methods relieve most of
this pressure to the point where the cost of sincere ranking is balanced
by the satisfaction of sincere expression.

In the context of ratings, Condorcet analogous methods relieve most of the
pressure to the point where the cost of uninflated ratings is balanced by
the satisfaction of sincere expression.


>From another point of view, the simplest CR method, namely Approval (How
do you rate the candidates on a scale of zero to one?) is favored over the
various Condorcet methods by many list members, especially when it comes
to public proposals for the near future.

So perhaps this list does give range voting its appropriate due.

Forest

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