[EM] Why use more than two grades?

[MIKE OSSIPOFF]

>It's refreshing that someone would sacrifice this influence to express
>himself more accurately, but until a huge majority can be trusted to
>take this same attitude, the carefully thought out degrees of approval
>will be lost in the din of the masses.

Yes. The people who voted for Gore in Plurality would give Gore
maximum points, insincerely, in Cardinal Ratings. If they think they
need Gore, if they think that Gore & Bush will be the frontrunners,
then they'll give Gore maximum points. Of course they'll give the
same to everyone whom they like more than Gore too.

Joe argues for the right to vote poor strategy, and I agree that
people have a right to drive without a seatbelt, ride a motorcycle
without a helmet, take heroin or crack, or commit suicide (if they
claim to have justification). But the balloting for CR is more expensive
than that for Approval, and so it's as if Joe is saying that someone
has a right to demand that the public pay for him to start a heroin habit.

People certainly have the right to make their own life choices and
make their own mistakes, but that doesn't include a right to have the
public finance the mistakes.

Remember that to use CR for state
elections, it would be necessary to set up the new balloting in every
county of the state. I like some rank methods, which of course
likewise need new balloting. But if we're going to use new balloting,
let's make the _best_ use of it.

But Joe has underscored something that I said earlier: Maybe CR will
have more appeal than Approval, for the reason Joe talked about, and
because people are used to rating things numerically. If CR is more
winnable than Approval or Condorcet, then I'd be glad to get CR.

Replying to Joe's statement that Condorcet voters should vote only
2 preference levels, I point out that the 2-level strategy is optimal
for CR, but not for Condorcet. Well under some conditions, with 0-info,
with 3 candidates it could be optimal in Condorcet & BeatpathWinner too,
but that's the exception.

>
>Of course, if it is a matter of honesty and conscience, that is reason
>enough to disregard what other voters might do. I voted for Nader for this
>reason.  Otherwise, I felt, some day God would ask me why I voted for one
>of the evils when there was a perfectly good choice available.

That's right. Principle matters. And as long as we keep voting for
the same thing, the same 2 identical parties, how can we expect anything
to ever improve socially? So the reasons for voting honestly include
both principle & practicality.

>It might be practical to have a non-binding poll based on honest approval
>levels, just to help everybody know where they stand before the big
>election

Organizations could do that, nongovernmentally. That should be
proposed to organizations that want someting other than the big-2.

The simulation that you describe below is social utility simulation.
Voters sincere ratings of the candidates in the simulation are
calculated based on their distances in issue-space. Based on those
ratings, rankings & Approval strategy can be determined. It's
assumed that voters rank candidates in order of how they rate them,
and that, in Approval, they vote for the above-mean candidates
(that's the optimal 0-info Approval strategy).

Pairwise-count methods do very well by SU. Borda often does better
(but would be a strategic mess in our actual elections, which aren't
0-info). SSD & BeatpathWinner do noticibly, slightly, better than
Tideman. But when information for strategy is available, all the
pairwise-count methods other than Condorcet & BeatpathWinner can be
expected to do worse. I should add that the simulations don't
include equal ranking, and that optimal 0-info strategy in BeatpathWinner & 
Condorcet sometimes involves equal ranking, but I don't
know how that would affect SU. Anyway, as I said, our elections aren't
0-info.

Approval consistently & robustly does better than IRV in social utility
simulation studies, due to IRV's jumps to extremes.

>One not so obvious application is to use a high resolution system when
>evaluating the performance of a low resolution method during simulation.
>Start the simulation with the honest levels of approval of the
>hypothetical voters.  From these, figure out how the voters would mark the
>more restrictive ballots, taking into account the likely strategies of the
>voters. Using only these ballots (without the more expressive information)
>try all of the competing methods. When the results are in for the
>different methods, see how they stack up to a higher resolution method
>based on the honest values.  Repeat the simulations until the competing
>methods start to show patterns of strengths and weaknesses.  The higher
>resolution method has inside information that the competing methods do not
>have, so it stands a chance of being a standard of comparison.
>
>Forest

_________________________________________________________________
Get your FREE download of MSN Explorer at http://explorer.msn.com