[EM] What exactly is an "Election Method"?
kislanko at airmail.net
Thu Aug 19 12:57:33 PDT 2004
The exercise to describe Condorcet in 12 words or less emphasizes something
I've been thinking of for a while. A "method" is a collection of processes,
and what the list came up with for a short description of a Condorcet-based
method makes very clear what the highest-level subdivision of a an "election
"Tally the round-robin pairings using the voters' orders of preference."
You need a process to count the votes, and you need a process to collect the
input to the tally process. That's my definition of an EM, although I'd word
it the other way around.
n. A combination of a procedure for collecting voter preferences and an
algorithm for counting votes.
>From a purely analytical standpoint, these can be dealt with separately and
I believe it is much easier to do so. I take no sides on the IRV vs
Condorcet discussion, but observe that you can do neither if the collection
process only supports plurality. Any improvement over plurality needs some
of ranked-ballot input, and it seems to me to be pointless to be debating
the attributes of tallying algorithms if there's no input to talley except
Folks on this list tend to concentrate on the properties of the algorithm
for the "counting votes algorithm", and that is proper. But in order to do
so there are unstated assumptions regarding how the preferences were
collected. Any Condorcet-based method that only notes voter x preferred A
to B and voter y preferred A to B but doesn't take into account that voter x
preferred C to B while voter y preferred B to C will necessarily be wrong in
assuming that x's and y's A>B votes are equal, and there's no way to resolve
that just from the pairwise matrix. It just can't show the difference
between A>C>B and A>B ballots. (Yes, voter x's C>B vote will show up there,
but B still gets the same benefit from x and y, even though they didn't
intend to equally support B).
>From a terminology standpoint, IRV has an instant advantage. Every voter I
know understands run-offs, and every voter I know would prefer to only vote
once to express their preferences. So IRV advocates need the voters to adopt
ranked-ballots. And of course, without ranked-ballots, you can't have
Condorcet-based tallying procedures. Ranked ballots also support plurality
and approval -based methods. So pretty much everyone should work toward
getting the collection procedure to provide ranked-ballots as input to an
arbitrary tally algorithm..
To re-open an old discussion, I would argue strongly that an even more
general ballot would be appropriate for any Condorcet-based method. If there
are 5 candidates in the race I might select the A>B>C>D>E ballot, but if you
take A, B, and E out of the race, in the "round robin" I might rationally
choose D>C, because all of the issues that made C better than D were covered
by A and B. With them out of the picture, D may agree with me on more
high-priority issues than C.
So I would argue quite strongly that if you use a pairwise matrix as your
tally mechanism, you allow each voter to fill in their pair-wise
preferences. If you don't, the system is not "transparent" and can't be
backed-up by voters' ballots. That would be a lot harder to sell (but it's
trivial to implement)
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Election-Methods