[EM] Voting Systems Study of the League of Women Voters of Minnesota

James Green-Armytage jarmyta at antioch-college.edu
Mon Oct 11 05:03:54 PDT 2004


	I think that the problem with range voting, as with approval voting, is
that it fails the majority mutual majority criterion, which states that if
there is a set of candidates S1 such that a single majority prefers every
candidate in S1 to every candidate not in S1, then the winner should come
from S1. It also fails the majority criterion, which is a special case of
mutual majority where S1 has only one member, who is the first choice of a
majority of the electorate. I would not consider approval voting or RV to
be a truly superior method for this reason, although I would prefer both
of them to plurality.
	I recommend a Condorcet-efficient method for any voting situations where
majority rule is appropriate. The remaining challenge, in my opinion, is
to temper the strategic vulnerability of Condorcet-efficient methods to
the "burying" strategy.

my best,
James

>The assumption that the mathematical analysis of voting systems has
>nothing to do with actual voting systems is mistaken; there seems to
>be a level of cognitive resolution below which inferior systems (such
>as IRV) will appear "good" because the mathematical evidence is
>incomprehensible. They also mistakenly assert that
>IRV eliminates the spoiler problem.
>
>Warren D. Smith has proposed (on the approval voting list) range
>voting (RV), which is the continuous analog of approval voting (or
>cardinal voting); a voter can vote any number in the unit interval for
>each candidate. A common criticism of approval voting, which occurs in
>the LWV article--that approvals all have the same intensity (equal to
>one)--is eliminated in range voting. However, it is more involved to
>implement than approval voting.
>
>Smith shows that IRV will lead to a 2-party duopoly over time (his is
>expected), whereas RV will not. Smith defines a decision theoretic
>notion called Bayesian regret and has computer simulations which show
>that range voting minimizes the average Bayesian regret over all
>systems compared (including plurality, IRV, approval voting...). One
>would like to see statements and proofs of the RV-analogues of
>theorems about approval voting that have appeared in the literature,
>though certain nice properties are immediate (monotonicity,
>summability).
>
>-FL




More information about the Election-Methods mailing list