[EM] multi-method combo
araucaria.araucana at gmail.com
Wed Jun 7 15:17:51 PDT 2006
Hey there, you all. I've been off the list for a while.
Recently I read The Wisdom of Crowds (
), and one thing that struck me is how important it is that when
aggregating the opinions of a group, it is crucial that each vote be
independent. Each voter should express his/her preference without
considering what other voters or blocks are going to do, as much as
Unfortunately, given Arrow's theorem, there is going to be something
unsatisfactory about any election methods proposal, and in every
proposed method on this list, there is some situation in which a
faction could successfully game the system based on what they think
other voters are going to do.
If the object is to discourage strategic voting and encourage
complete preferences, it seems like you want characteristics like
Later no harm
Later no help
in addition to the usual criteria we desire in a "strong"
Condorcet-completion scheme, to discourage bullet voting. But you
want all the other strong Condorcet properties, because you want to
avoid pushover voting, like you get with IRV.
So I've been thinking about combining several methods. No method
would have all the desirable characteristics, but in combination there
would be a partial satisfying of most criteria. Use a simple ballot
that can be interpreted many ways, tabulate the winner using several
methods, and then choose randomly from among those winners, weighted
by the number of methods that pick that winner. Here's an example:
- Ratings ballot,
0 (least preferred) to 100 (most preferred),
equal ratings allowed
- Use M methods simultaneously. For example, M=3:
1) Range Voting / Cardinal Ratings
2) IRV [ER(whole) variation]
- Calculate the 3 different winners.
- Pick 1, 2 or 3 with uniformly random probability, and the
overall winner is the winner of the corresponding tabulation
This incurs extra cost for multiple tabulations, with IRV-ER(whole)
being the most expensive, but the advantage is that any strategy that
would increase the odds of winning under one method would have only a
1/3 chance of succeeding.
So some questions for you all:
Is this idea completely off the wall, or does it look like it might
have some merit?
If it has some merit, what would be a good combination of methods?
araucaria dot araucana at gmail dot com
More information about the Election-Methods