Davison Run-Off is Coombs' Method. Not Bad.

[Mike Ossipoff]

Not bad at all, and I say that with no sarcasm. You've re-invented
the Coombs method. So you can't call it "Davison Runoff", and]
you've got to call it "Coombs"--unless you modify it into something
else.

Coombs is very much better than MPV (I'm not calling MPV
"Instant Runoff" anymore, because it's now doubtful whether
MPV is the most deserving of that name).

In fact, Coombs is about the best method short of Condorcet
& Simpson-Kramer, & the methods mitigated by GMC & the Condorcet
Criterion or by NML & NPL. At least if we're only talking about
1-balloting count rules. But if I had to propose to the public
a non-Condorcet 1-balloting method, it wouldn't be Coombs,
even though I like it; it would just be Approval or "Yes/No"
(where anyone getting "No" from a majority is disqualified, and
the winner is the un-disqualified candidate with the most
"Yes"s).

Actually, if I couldn't interest the public in Condorcet,
then I'd try for BeatsAll//Approval, a (potentially) 2-balloting
method. If I couldn't get that either, I'd try for
Inclusive 2nd Balloting--in particular, Majority//Approval
(where a majority wins the 1st balloting, otherwise all
candidates compete in a 2nd balloting by Approval).

The reason I wouldn't propose Coombs, even though I like
it, is that its strategy is of an unfamiliar kind. I personally
feel that Coombs' defenseive strategy need isn't as bad as
the non-ranked methods that I'd propose instead, but those
non-ranked methods have a familiar kind of strategy need,
and are obviously better than Plurality. Coombs's strategy
need is so different that people might consider its strategy
unacceptably strange, and might screw up badly, strategically.

Because, Coombs does require defensive strategy where Condorcet
doesn't, and even when they both do (order-reversal conditions)
Coombs requires more complicated defensive strategy.

Coombs is at its best when there's a 1-dimensional political
spectrum. The beauty of it is that, under those conditions
it will always be the extreme candidates who will be the
last choice of the most people. So Coombs eliminates from the
extremes, nibbling its way to the middle, and finally leaving
the Condorcet winner as the only un-eliminated candidate. 

Sounds pretty good, and, as I said, it would be my choice after
Condorcet & Simpson-Kramer, & the criterion-mitigated methods,
if it weren't for unfamiliar strategy needs.

The problem, you see, in a big election, is that it's necessary
for (e.g.) the progressives to agree on their lower choices. Now,
in a _big_ election, with lots of candidates, as we'd have
in a rank-balloting election, is everyone going to agree on
all their chocies, right down to their last? Not that likely.
But, on the other hand, asking progressive to agree artificially
on their lower choices is a lot easier than asking them to
falsify their 1st choice, as MPV sometimes requires. So
Coombs is an immense improvement over MPV.

When order-reversal is attempted, things get hairier, and
rankings have to deviate from sincerity, based on a knowledge
of which way (along the 1-dimensional spectrum) the Condorcet
winner is (the candidate who'd beat each of the others in
separate 2-candidate elections). I don't know whether or not
it's necessary to actually know which candidate is CW, and
if not, that's a strong point in Coombs' favor. But defensive
strategy under order-revesal conditions requires knowing which
direction the CW is, and reversing one's preference orderings
sometimes, as called for by Coombs' anti-order-reversal defensive
strategy.

But, as I've said before, I don't consider order-reversal to be
a likely problem in public elections anyway, and Bruce has agreed
with me on that. So Coombs is actually pretty good, though not
as good as Condorcet, because Coombs requires strategy as soon
as your policy grouping doesn't agree precisely on its entire
ranking. So Condorcet is better.

But Coombs is an improvement, and I won't argue against
Don if he promotes it, because it's pretty good. I have no
problem with Coombs being offered alongside Condorcet as
alternatives to Plurality (to the public) & as alternatives
to MPV (to CVD & other electoral reformers). They're both
good; let the reform opponents be the ones to complain about
Coombs' strategy; nothing wrong with offering more than 1
good method; may the best method (Condorcet) win.

I only argue against unacceptable methods like MPV. But
I continue to claim that Condorcet is best, due to its
properties that I've been talking about. For instance,
Coombs can fail GMC, Condorcet's Criterion, NML, &
NPL when voters are split regarding their last choice.
Not as good as Condorcet, but still not bad, if voters
can artificially agree on the same spectrum-ordering.
Or if they naturally agree on that.


Mike


Mike





--