[EM] 03/30/02 - Re: How to vote in Approval:

[Forest Simmons]

On Sun, 31 Mar 2002, Adam Tarr wrote:

> Dave wrote:
>
> >I have nothing nice to say about Approval, noting that no one seems able
> >to give me useful guidelines as to how well I should have to like a
> >candidate to say "approve".
>
> Rob LeGrand came up with a decent strategy that makes a lot of
> sense.  Approve every candidate you like more than the front runner, and
> approve the front runner if you like him/her more than the second-place
> candidate.  This requires knowing who's top two in the polls, but this is a
> pretty good bet in a large election.

Here's another approach that is near optimal in practice:

(1) rank the candidates as you would for your favorite Condorcet method.
(2) convert the ranks to Borda scores by subtracting from the number of
candidates.
(3) adjust the numbers so that they better reflect your relative
intensities of preference or "utilities" for the candidates. Call these
numbers "ratings."
(4) assign subjective probabilities of winning for each candidate.
(5) multiply the ratings by the respective probabilities and add the
results to get what we will call the expected winner rating (EWR).
(6) approve all of the candidates that you have rated above this EWR.

In the case of zero information all of the probabilities would be equal,
so you would be approving by the "above mean utility strategy" which is
often advocated as near optimal for the zero-info case. In other words,
this strategy is a straight-forward extension of the above mean utility
strategy.

<snip>

> Approval certainly has its places.  If I had to advocate wide-scale single
> winner voting reform to a large group that was not familiar with it, my
> inclination is to advocate approval voting.  It's simple and it's an
> obvious improvement over plurality.  But if I'm trying to turn around an
> IRV advocate, I would definitely jump straight to advocating Condorcet
> voting.  Once someone has accepted the concept of ranked ballots, they have
> a certain intuitive appeal that's hard to give up.  It's hard to convince
> someone who's bought into IRV that you get better results with plain old
> unranked approval ballots.  But a few simple examples will convince all but
> the most hardheaded IRV advocate that Condorcet is superior.

Yes, if they are willing to fill out the IRV ballots, show them what those
ballots can do, given half a chance.

Better yet, show them the advantages of CR ballots where you don't have to
pretend that all of your preferences are of equal intensity.

Most people don't have the patience to rank two dozen candidates. Giving
each a grade of A,B,C,D,F or (default) E, is easier.

[The reason I believe the default grade should be E, is that F should be
reserved for someone that you are sure about being no good. But beware,
this convention might backfire if you allow election of someone with a
majority of below D grades.]


Forest