[EM] Automated CR Strategy

[Forest Simmons]

The standard use of cardinal ratings (CR) ballots (awarding the win to the
candidate with the highest average rating) encourages strategic voting so
that voters in the know tend to vote exclusively at the extremes.

Is it possible to automate CR strategy well enough to eliminate the
advantage of the strategic voter over the naive voter?

Perhaps not in all cases, but in the case of large public elections, where
there is always an irreducible residue of statistical uncertainty, no
matter how carefully and honestly the polls are conducted, the answer to
this question is undoubtedly, yes!

What I have in mind is this: voters submit CR ballots in the form of
letter grades (A thru F or A thru Z, I don't care at this point), and then
these ballots are sampled statistically in a way that is an improvement on
any possible pre-election poll, but not so perfectly as to remove all
uncertainty.

This statistical information is used to get winning probabilities for the
various candidates.  If this is done correctly, these probabilities will
be more reliable than any that could be calculated from pre-election poll
results.

Each ballot is then transformed into an optimal ballot relative to these
estimated probabilities.

The candidate with the highest average rating on these transformed ballots
is the winner.

The only way a voter could get more reliable probability estimates would
be through some oracle or time machine.

If a voter takes it upon himself to vote at the extremes, these choices
will automatically be preserved by the transformation, since optimal
strategy preserves ratings at the extremes.

In summary, my idea is to take advantage of the pocket of uncertainty
inherent in public or private polls of large numbers of people to
virtually eliminate the advantage of sophisticated voters over naive
voters in the context of cardinal ratings with large numbers of voters.


Forest