[EM] Truncation and Ballot Type

[Forest Simmons]

On Wed, 18 Sep 2002, Adam Tarr wrote:

<snip>

> There's no way to distinguish between the lazy voter and the sincerely
> indifferent voter, or between the informed voter and the haphazard
> voter.

The interactive ballot that queries the voter on all pairwise comparisons
and distinguishes between A=B and A?B as suggested by Stephane and
others might be able to effectively discern these differences.

<snip>

>
>  ...  The simplicity to
> the voter of ABCD(E)F voting is worth it.  The voters who are interested
> and involved enough to actually need six distinct levels of approval are
> the same voters who will understand that the unmarked candidate will get
> the E grade.
>
> I spent some time a few months back seeing if approval-completed Condorcet
> could be made to work using this sort of ballot, but in the end I realized
> that the only time approval completion worked like I thought it should was
> when the approval votes supported the sincere Condorcet winner.  When the
> approval votes did not do so, the system encouraged all sorts of strategic
> manipulations.  So I gave up on ACC, and I now hang my hat with
> winning-votes based Condorcet, which seems to be the system that is most
> resistant to strategic manipulation by a wide margin.

It is true that ACC was once one of the leading suggestions for scoring
Grade Ballots.

But don't forget that Grade Ballots can be scored pairwise.  Once the
pairwise matrix is constructed, the winner can be extracted from that
matrix according to the rules of SSD or Ranked Pairs.

It may seem like a defect that grade ballots cannot fully rank more than
six candidates, but I believe that

(1) the simplicity and clarity of the ballot far outweigh the supposed
benefits of fully ranking more than six candidates,

(2) forcing the voter to collapse preferences down to six levels makes the
surviving comparisons more meaningful,

and

(3) therefore, in the statistically few cases when different results might
obtain from the two distinct ballot types, as likely as not the grade
ballot will give the superior result.

A note about the danger of ranked preference ballots:

When I was in Vietnam, the phrase "number one" meant best, while "number
ten" meant worst.

If memory serves me correctly, a few years later a movie came out in which
Bo Derek played the part of a woman who was "a perfect ten," as in the
scoring of Olympic skating, diving, gymnastics, etc.

Would you rather have a 4.0 grade point, or a 1.5 ?

It doesn't matter how prominently the instructions say to rank the
candidates from one to N on the "Vietnamese system" where small is good
and large is bad, a significant percentage of American voters will revert
to the dominant "large=good, small=bad" mode of thought while filling out
their ballots.

It wouldn't do any good to change the convention either; there would still
be plenty of confused people marking their ballots the exact opposite of
their intentions.

The confusing nature of ranked preference ballots is probably one of the
main reasons that voters in Australia use candidate supplied "voter cards"
in single winner elections and vote "above the line" for their
party-approved order in multi-winner elections.

In effect they have a proxy system that could be implemented more simply
with plurality ballots: "Choose one voter card" or "Choose one party
approved ranking."

When interactive ballots become common, then we can start thinking about
fully ranked preference ballots in public elections.

By the way, here in Oregon the IRV initiative (which never made it to the
ballot) was only proposing to rank the top three candidates, with
mandatory truncation for the rest, due to supposed technological
limitations, even though the public schools all have optical scanners for
the ballots that we call "grade reports" filled out for each class each
term by the teachers. In addition to the standard grades of A through F
they allow for half a dozen other options (Incomplete, Pass, Withdrawal,
etc.) without requiring more than one line per candidate (i.e. student).

If more than six levels are really needed, we could minimize confusion by
adding the familiar plus and minus options to the Grade Ballot rather than
going over to a ranked system.  A Grade Ballot with the plus/minus options
could distinguish sixteen levels (because the default grade E has no
plus/minus option), many more than the magical seven recommended by the
psycho-metric folks.

How many single winner elections have more than sixteen candidates worth
distinguishing by rank?

Ranked ballots (usually) do not give the option of collapsing any
preferences except at the lowest level. A sixteen level grade ballot
allows collapsing preferences at any level, as well as fully ranking up to
sixteen candidates.

So from now on, when discussing various Condorcet methods, why not keep in
mind the option of collapsing preferences at any level, as Rob LeGrand has
argued for in the past?  The use of Grade Ballots makes allowing it easier
than prohibiting it.

Does this have any affect on the wv versus margins debate?

Forest

----
For more information about this list (subscribe, unsubscribe, FAQ, etc), 
please see http://www.eskimo.com/~robla/em